chore(release): seal 1.25.3 model catalog delivery and QA evidence

This commit is contained in:
Agent Workbench Dev
2026-11-05 03:02:21 +08:00
parent 1bd94856d3
commit cef636fdbd
1673 changed files with 1876129 additions and 11 deletions
+6
View File
@@ -267,3 +267,9 @@ docs/reliability-1.25.2-release-identity.json -text
docs/reliability-1.25.2-offline.txt binary
docs/reliability-1.25.2-git-index.txt binary
docs/LIVE-DELIVERY-X-PREFREEZE-REFUSAL.txt binary
docs/model-catalog-1.25.3-evidence/** binary
docs/model-catalog-1.25.3-qa-follow-up/** binary
docs/model-catalog-1.25.3-release-identity.json -text
docs/model-catalog-1.25.3-offline.txt binary
docs/model-catalog-1.25.3-git-index.txt binary
+2
View File
@@ -52,3 +52,5 @@
!/docs/task-agreement-1.25-evidence/**/*.log
!/docs/task-agreement-1.25-supplement/**/*.log
!/docs/reliability-1.25.2-evidence/**/*.log
!/docs/model-catalog-1.25.3-evidence/**/*.log
!/docs/model-catalog-1.25.3-qa-follow-up/**/*.log
+3 -1
View File
@@ -1,5 +1,7 @@
# Agent Workbench项目执行规范
1.25.3已完整本地交付:最终来源1bd94856/候选20261104181218275,上游目录搜索多选/同源凭据引用/手动与自动选模、配套推理开关强度与空响应诊断可用。完整源码926/926实际来自6741510b;最终候选仅两个验收驱动差异,全部产品/测试输入与22编译相同,严格差异已封存。657输入源码ZIP独立离线重建22/22、三归档、源码78组/便携126组、真实隔离安装12生命周期及安装副本103UI全部通过。唯一正式升级283191ms,原PID36552/instance/Provider与75项目91会话6723记录14320事件及outbox、全部原行/导出/工件/公开配置在网页1440/390和真实Windows回看后再次exact,窗口0/9229关闭、hasKey=true。主1635份138998818字节已离线通过;封存后仅QA元数据修复另跑6组10图和6个实际版本门检查,补验16份771613字节另封存,不重制发行。验收docs/QA-MODEL-CATALOG-1.25.3.md、回顾docs/RETROSPECTIVE-MODEL-CATALOG-1.25.3.md、身份docs/model-catalog-1.25.3-release-identity.json。真实供应商目录/模型请求均0,不操作真实Key、不委派/推送、不续跑旧样本;上游真实能力覆盖、供应商稳定/自主交付、原网络/重复读取/真实预留边界/Purge未知保留。以下“正式仍1.25.2/待发行”等为历史时点,当前状态以本段和最终报告为准。
1.25.3最终发行接续:6741510b完整源码926/926、零取消/跳过;候选20261104175949830源码ZIP657输入独立重建22/22和三归档通过。便携首轮被verify-controlled-release旧版本断言(止于1.25.2)拒绝,原release-fixed/portable失败保持;只补1.25.3并在相同候选复验,完整便携/实际SDK/浏览器/控制引擎4组通过。最终差异仅两个验收驱动(此断言和新write-once标签),所有产品/测试输入及22编译不变;不为仅验收范围再跑整套源码,完整源码实际来源仍6741510b。现在本地提交后另冻release-verified、重制候选/独立重建/归档/源码便携UI与真实安装103组,之后才保留原PID36552/Provider正式升级。正式仍1.25.2,真实模型和目录0;不触真实Key、不委派/推送、不续跑旧样本。
1.25.3发行阻断修复:bb4b4977完整源码926项923通过3失败,均为upgrade-running兼容清单漏1.25.3;原full-final.txt与release-final冻结保持,不改成通过。当前将当前版本绑定随包package.json的静态JSON导入(编译时内联,便携脚本无需src),保留历史版本/未知拒绝,构建与安装/版本/实际两次及legacy升级相关36/36、保留对象/工件升级6/6通过。目录和推理实现不重做;阶段回顾后本地提交,另冻release-fixed,重新完整源码与独立发行,原release与release-final不覆盖。正式仍1.25.2/PID36552/Provider,未发真实模型/目录请求、不操作真实Key、不委派/推送、不续跑旧实验。
@@ -166,7 +168,7 @@ D固定可信来源生产策略已完成:实现9106655/最终验收d05a741,6
## 当前本地服务与Windows
- 正式服务入口http://127.0.0.1:4318,当前1.25.1,沿用原PID36552/instance与Provider;来源0a82636d/候选Agent-Workbench-1.25.1-win-x64-20261104103902806,验收见docs/QA-OUTLIER-INSTALL-1.25.1.md。升级后65项目81会话6377记录13484事件及outbox精确保持;W样本另增2自有项目/会话,当前67项目83会话,原行/导出/3720原工件精确一致,窗口0/9229关闭。普通开发不重启服务,不提取/复制/解密/重存Key;V已按诊断partial封存,真实交付另冻新样本,旧V/T/R/P/O不续跑。
- 正式服务入口http://127.0.0.1:4318,当前1.25.3,沿用原PID36552/instance与Provider;来源1bd94856/候选Agent-Workbench-1.25.3-win-x64-20261104181218275,验收见docs/QA-MODEL-CATALOG-1.25.3.md。最终75项目91会话6723记录14320事件及outbox、全部原行/导出/工件和公开配置精确保持,窗口0/9229关闭。普通开发不重启服务,不提取/复制/解密/重存Key;真实目录/参数及工程收益另冻新样本,旧Z3/Z2/Z/Y/W/V/T/R/P/O等不续跑。
- 浏览器/fixture验证使用独立端口与数据目录;用户项目与研究对象的登录/会话配置不用于测试凭据。
- Windows执行环境以实测shell为准。现有工具使用PowerShell 5.1和UTF-8;不用&&/||拼接命令,使用换行和$LASTEXITCODE核对结果。
- 桌面与移动共用执行宿主。手机127.0.0.1指手机,远程接入通过未来经验证的配对网关。
+2 -1
View File
@@ -1,12 +1,13 @@
# Agent Workbench 更新记录
## 1.25.3 · 上游模型目录、推理控制与兼容能力配套(发行验收中)
## 1.25.3 · 上游模型目录、推理控制与兼容能力配套(已本地交付)
- 模型中心增加“推理与能力”:按品牌或自定义兼容格式选择默认、关闭、开启、强度及 Messages 推理预算,保存、导入和实际请求使用同一份参数。
- 从上游获取可搜索的多模型目录,无需先手填型号;批量保存后可手动指定或自动适配。同源型号共用凭据引用,目录有能力与推理档位时同步呈现,缺失项保持待确认。
- DeepSeek JSON/流式工具往返保留所属原生推理状态,受原输出预算约束;接口能力声明、检测结果和本地接入范围可查看。
- Messages 空响应明确显示供应商返回为空,保留原文与用量,不再误报检查未完成。
- 修复当前补丁再次升级被兼容清单拒绝的问题;当前版本随发行信息配套,保留原进程的更新与回退继续核对原数据和凭据。
- 源码完整926/926(实际来源6741510b,最终候选仅验收脚本差异)、独立657输入重建22/22、源码78组/便携126组及实际安装12生命周期/103组UI通过。正式升级283191ms,原PID36552/Provider及75项目91会话全部原状态保持;真实供应商目录/模型请求0,未验证事项继续保留,见[验收](docs/QA-MODEL-CATALOG-1.25.3.md)。
## 1.25.2 · 预留证据、安装边界与读取进展(已本地交付)
+2
View File
@@ -1,5 +1,7 @@
# 本地 Agent 工作台
当前本地正式版 **1.25.3**:[打开工作台](http://127.0.0.1:4318/?release=1.25.3)。模型中心可从上游获取型号目录,搜索、多选后手动指定或加入自动适配池;推理模式/强度按品牌与自定义兼容格式配套,能力元数据缺失时保留待确认。使用见[模型目录](docs/MODEL-CATALOG.md),验证与限制见[验收](docs/QA-MODEL-CATALOG-1.25.3.md)和[回顾](docs/RETROSPECTIVE-MODEL-CATALOG-1.25.3.md)。原进程、连接凭据及全部原数据保持;真实供应商兼容、自主工程交付和 Purge 收益仍待独立验证。以下为各版本历史记录。
1.22.0已本地上线:任务中准备Vite预览,桌面端右侧「浏览」查看真实页面并审批固定操作;原截图、动作和下载可核对,手机可观察同一原图。683项核心测试沿用C完整通过原轮、250组界面沿用B原轮,均已严格绑定输入;D独立重建及实际桌面、便携、安装/升级通过,原连接和全部历史保持。见[工作台](http://127.0.0.1:4318/?release=1.22.0)、[使用说明](docs/EMBEDDED-BROWSER.md)、[验收](docs/QA-1.22.md)、[回顾](docs/RETROSPECTIVE-1.22.md)和[发行身份](docs/embedded-browser-release-identity.json)。新iPhone与真实模型效果单列待验。以下为历史交付记录。
+11 -2
View File
@@ -1,6 +1,15 @@
# 上游模型目录与多模型使用
用户补充:模型应正常从上游获取,不要求手填;可选择多个型号,使用时手动指定或自动适配。本切片并入1.25.3,沿用模型中心、ModelHub和现有模型池,不另建路由体系。正式交付状态以独立发行记录为准。
已随 **1.25.3** 完成本地正式交付,见[发行验收](QA-MODEL-CATALOG-1.25.3.md)及[回顾](RETROSPECTIVE-MODEL-CATALOG-1.25.3.md)。用户要求模型正常从上游获取,不要求手填;可选择多个型号,使用时手动指定或自动适配。实现沿用模型中心、ModelHub和现有模型池,不另建路由体系。
## 使用
1. 打开“模型中心 → 从上游添加模型”。可选已有连接使用原凭据,也可填写新上游的服务地址和鉴权;正常路径无需填写型号。
2. 获取目录后搜索并勾选要使用的型号,确认其能力与参数声明。上游缺少信息时按界面提示补充必要项;未提供目录的服务可用“手动填写(兼容)”。
3. 保存时保留当前选择、指定某一型号用于新任务,或建立自动适配池。之后可在工作台的模型选择入口切换;自动池按任务能力、健康、并发和所设策略选择成员。
4. 在连接的“推理与能力”里选择配置档、推理模式和强度,保存后用于新任务。上游提供的可选项会同步;未填写参数沿用上游默认,推理与答复共用输出预算。
获取目录会访问所选上游;保存配置本身不调用推理。模型列表不保证包含全部功能信息,声明支持与已验证、本地已接入分别展示,具体型号和代理的兼容仍以其实际响应为准。
## 实现
@@ -20,7 +29,7 @@ Chat/Responses和DeepSeek兼容服务使用根地址下的GET /models;Messages
目录品牌补强首轮有一项验收器断言失败:DOM已包含disabled的option,Playwright的toBeDisabled仍返回enabled。原截图和失败保留,改为核对真实HTMLSelectElement选项的disabled布尔属性,原期望/期限不变,最终网页/桌面通过。阶段复核另发现同地址/型号的新账号可能误中旧源匹配,已将去重严格限制到明确源连接,增加真实HTTP两账号隔离;不是将端点或型号当凭据身份。
证据位于`.qa/reasoning-controls/`的新文件。原109e5124范围的完整源码测试因本次需求扩大,在264个完成项、未观察到失败时主动停止所属进程;未完整通过,不作为新范围验收。原日志与取消回执位于`.qa/reasoning-1.25.3/full-source*`。新范围将重新冻结,旧freeze不覆盖。
开发证据位于`.qa/reasoning-controls/`,最终原件已封存到`docs/model-catalog-1.25.3-evidence/`。原109e5124范围的完整源码测试因本次需求扩大,在264个完成项、未观察到失败时主动停止;未完整通过,不作为新范围验收。扩大范围已另冻并完成完整源码、独立发行和正式升级;首取消、版本遗漏及元数据收尾失败与复验分别保留,详见发行报告。
公开契约参考:OpenAI官方GET https://api.openai.com/v1/models的OpenAPI说明;Anthropic官方SDK `https://raw.githubusercontent.com/anthropics/anthropic-sdk-typescript/main/src/resources/models.ts`(原文件已保存于`.qa/reasoning-controls/public-docs/claude-sdk-models.ts`),包括capabilities、effort、thinking.types与分页。兼容代理实际是否提供这些元数据尚未用真实账号核验,本阶段真实模型请求0。
+5 -1
View File
@@ -1,6 +1,10 @@
# 选择和管理模型
左下角「模型与连接」、顶部当前模型按钮和「模型中心」打开同一个模型中心。上方「新任务默认模型」可直接选择已保存连接或模型池,也可点击「添加模型」保存新的接口、协议、请求型号和凭据。保存新连接后,再在选择器中选用它;新增和切换本身不调用模型。
左下角「模型与连接」、顶部当前模型按钮和「模型中心」打开同一个模型中心。上方「新任务默认模型」可直接选择已保存连接或模型池。1.25.3 的主要新增入口是「从上游添加模型」:选择已有上游或填写新上游地址与鉴权,获取目录后搜索并勾选多个型号,无需逐个手填。没有目录的服务仍可用「手动填写(兼容)」,详见[模型目录](MODEL-CATALOG.md)。
批量保存时可保留当前选择、手动指定默认型号或建立自动适配池;之后在工作台模型入口切换。自动池按任务能力、健康、并发和所设策略选择成员,每次实际型号仍进入请求账本。获取目录访问上游,保存与切换本身不调用模型推理。
连接的「推理与能力」可选择推理模式、强度及兼容参数配置档;上游返回的能力与参数声明同步呈现,缺失项保持待确认,本地未接入的功能不会仅凭声明变成可用。保存后用于新任务,原 Run 仍使用其冻结配置。当前版本的验证与限制见[1.25.3验收](QA-MODEL-CATALOG-1.25.3.md)。
需要凭据但尚未填写、或已经停用的连接不能在选择器中选用。可先编辑连接,或在「凭据保存」中保存、加载本机加密凭据。需要实测接口时,「验证响应(调用模型)」会发出请求并记录用量,区别于保存和切换操作。
+74
View File
@@ -0,0 +1,74 @@
# 1.25.3:模型目录、推理控制与兼容能力验收
状态:已独立发行、实际隔离安装并升级正式本机服务至 **1.25.3**。正式原 PID **36552**、instance、Provider 和凭据保持,网页及真实 Windows 窗口只读回看后再次核对全部原状态。入口为 <http://127.0.0.1:4318/?release=1.25.3>。本阶段真实供应商目录请求、真实模型请求均为 **0**;全部协议和选路验证使用自有 HTTP 夹具。
## 用户要求与交付
| 要求 | 交付与范围 |
| --- | --- |
| 正常从上游获取型号,不要求逐个手填 | 模型中心主入口“从上游添加模型”;支持新上游、现有连接、搜索、多选与批量选中。没有目录的服务保留手动兼容入口。 |
| 保存多个模型,使用时手动或自动选择 | 复用现有默认模型和模型池;保存可保留当前选择、指定默认型号或建立自动适配池。自动按任务所需能力、健康、并发及池策略筛选;未宣称语义任务路由已验证。 |
| 品牌与自定义兼容的参数配套 | Chat、Responses、Messages、DeepSeek 共享配置档,贯通品牌建议、手动覆盖、JSON、导入、保存及 Run 冻结。 |
| 可开启推理并选择强度 | 可选择默认、关闭、开启及协议支持的强度;Messages 支持预算或 adaptive。上游有模式/档位声明时限制可选范围;推理共用原输出预算。 |
| 同步能力与功能 | 同步上游实际返回的能力、容量、参数和生命周期元数据,区分声明、实测与本地接入状态。上游未提供的内容保持未知,不按型号名补造。音视频、服务端工具、结构化输出约束等尚未接入的执行能力仍明确列出。 |
| 已有凭据与用户配置保留 | 同源型号引用源连接凭据;不读取或重新保存原 Key,不把相同端点当相同账号。原本地参数与命名不被重新选取目录型号静默覆盖。 |
操作说明见[模型目录与多模型使用](MODEL-CATALOG.md),参数契约与开发验证见[推理阶段验收](QA-REASONING-CONTROLS.md)。目录接口不保证提供所有能力字段;品牌或兼容协议相同,也不能据此认定每个型号支持全部功能。
## 来源与完整源码验证
- 产品切片:Messages 空响应诊断 `3ab3a6e4`,推理与配置档 `f1242ee8`,模型目录 `d0381918`,升级版本配套 `6741510b`。
- 完整源码轮次实际来源 **6741510b2dc9dd1191ae7f459474a981ed1d07c9**:**926/926**,失败、取消、跳过和 todo 均为 0,原总时长 2143077.8077ms。
- 最终候选来源 **1bd94856d31a4d122eb03a9d7d199b42dbabacc6**;候选 `Agent-Workbench-1.25.3-win-x64-20261104181218275`。与完整源码轮次相比只改两个发行验收驱动,全部产品/测试输入及 22 个编译输出相同;差异逐项 SHA 证明封存在 `development/candidate-input-delta.json`。
- 最后证据与回顾提交还包含两处 QA 元数据修复及离线验证器。它们在正式升级和主证据封存后完成,不改变已安装候选或 sourceCommit,不把最终 Git HEAD 称作上述完整源码轮次。
相关开发验证包括目录及路由/凭据/能力/推理 68/68、热升级 4/4、推理相关 85/85 加补强专项 8/8、Messages 28/28,以及版本/安装相关 36/36 与保留对象/工件升级 6/6。完整轮次与这些局部检查分别保留,不相加成新的完整轮次。
## 独立载荷、界面与安装
| 验证 | 实际结果 |
| --- | --- |
| 冻结与离线重建 | 652 个冻结输入;源码 ZIP 解包后 657 个输入独立离线重建,22/22 编译字节一致。 |
| 三归档 | portable/source/evidence 的 CRC、清单与原字节均通过;安装器 SHA256 为 `55f144ea2698ed3ac3da8d74ca9287fc02f7844dc23511113ba098154d8d4be7`。 |
| 实际便携基础 | REA、真实 SDK、浏览器及受控引擎通过。 |
| 源码界面 | 12 套、78 组;目录、推理、要求复核、单轮观察、范围约定与诊断闭环均保留。 |
| 便携界面/文件流程 | 17 套、126 组;含旧原生参数 6 组、Messages 7 组及文件/结果/读取 9/16/10 组。 |
| 真实隔离安装 | 12 生命周期场景通过,安装副本 14 套、103 组 UI 通过;卸载保留数据。 |
| 正式网页 | 1440×1080 与 390×844,旧 Task、推理表单及目录入口只读回看;errors 与 rejectedWrites 均为空。 |
| 正式 Windows 窗口 | 实际安装路径、release=1.25.3、可见 v1.25.3;原完成 Task/历史超时、目录主入口/表单、DeepSeek 推理模式与强度可见。未保存、检测或获取真实目录。 |
最终源码、便携和安装副本的桌面及 390px 关键原图已实际查看。正式原生窗口仅报告其实际桌面尺寸;窄屏证据来自实际隔离 Electron 和正式网页,不冒充正式原生 390px 回看。
旧 fidelity 验收脚本 6 组/10 图已完成,但原结果没有 health version 字段,汇总器因此失败。独立复核原 appRoot、候选清单和全部 22 编译输出后追加 `portable-fidelity-reviewed.json`;其版本依据是载荷归属,不能说原收据已经观察 health。Messages 原字段为 `testedVersion`,独立运行 7 组通过,实际 health=1.25.3,追加 reviewed 收据。两份原结果和首失败均未改写。
主证据封存后,fidelity 脚本补实际 `/health` 版本,发行汇总兼容 `testedVersion` 并拒绝冲突别名。对已验便携再跑 6 组/10 图,实际 health=1.25.3;直接执行修复后的实际版本门,原 Messages 收据通过,缺失/错误/冲突版本均拒绝。仅 QA 变化,未重建、重制发行或再次安装。补验保存在独立 `model-catalog-1.25.3-qa-follow-up/`。
## 正式保留状态升级
唯一正式升级于 2026-11-04 UTC 完成,安装器 exit=0,实测操作 **283191ms**。原 1.25.2 操作 289669ms、1.25.1 的 156369ms 及 1.25.0 的 774098ms 为各自历史观测,本轮不声称进一步性能收益。
| 项目 | 升级前及网页/原生回看后的最终核对 |
| --- | --- |
| 宿主 | PID 36552、原 instance、原 Provider;空闲,hasKey=true。 |
| 原数据 | 75 项目、91 会话、6723 记录、14320 events 和 14320 outbox,全部原表行、session IDs 及 eventCursor exact。 |
| 原证据 | 全部原导出、原工件字节与公开模型中心/配置 SHA exact。 |
| 安装 | 指针 revision 29、previous 1.25.2、operation completed;runtimeRoot 为上述候选。 |
| 收尾 | 正式窗口 0,9229 关闭;Key 未读取、复制、解密或重存。 |
`formal-native/launch.json` 记录安装入口进程,入口可能在转交版本化桌面程序后退出;正常关闭依据实际 Windows 窗口清单和最终正式 host 窗口数,不依赖启动器寿命。
## 首次失败与限制
1. 旧 `109e5124` 范围的完整测试因用户新增目录需求,在 264 个完成项、未观察到失败时主动停止;原取消回执保留,不能算完整通过。
2. `bb4b4977` 完整 926 项为 923 通过、3 失败,均指向升级兼容清单漏 1.25.3。修复后另跑完整 926/926;不倒写首轮。
3. `6741510b` 候选便携被控制引擎验收器旧版本断言拒绝;原失败保留。补版本后另冻最终候选,完成全部独立载荷和安装。
4. fidelity 版本字段缺失、Messages 字段命名差异,以及开发时 option matcher、缺声明、SQLite 夹具清理挂起均保留原记录。正式原生回看一次点击未显示在窗口内的按钮被越界拒绝,重新观察、滚动到可见区域后成功,未保存配置。
5. 本轮新验证不证明真实供应商目录或能力元数据覆盖、接口稳定、模型自主工程交付、间歇网络唯一根因、重复读取真实改善、真实预留异常边界收益或原版 Purge AB/BA。Z2 型号不符、Z3 空响应、DeepSeek length 和其它历史失败均保持;不续跑旧实验。
## 封存与复核
主证据 **1635 份、138998818 字节**(不含 manifest),SHA 清单固定为 `a2e6ac2d982a147557ac256457516082c9b0346bd76f20627b149699e5880751`。QA 补验另 **16 份、771613 字节**(不含其 manifest)。主目录及旧封存目录未改写。
含两个 manifest 和外部发行身份,Git 索引核对 **1654 份、140305327 字节**全部一致;仅为本版两个新目录添加日志例外和二进制属性,避免遗漏或换行规范化改变原件。
独立只读验证命令:`node scripts/verify-model-catalog-evidence.mjs`;暂存后用 `--index` 逐项核对主证据、补验、两个 manifest 和发行身份的 Git blob 原字节。结果见[离线收据](model-catalog-1.25.3-offline.txt)与[索引收据](model-catalog-1.25.3-git-index.txt),[发行身份](model-catalog-1.25.3-release-identity.json)及[回顾](RETROSPECTIVE-MODEL-CATALOG-1.25.3.md)。仅本地提交,不推送。
+2 -2
View File
@@ -1,6 +1,6 @@
# 推理与能力配套:开发阶段验收
状态:源码已实现并验证,正式仍1.25.2;独立发行验收另记。本阶段真实模型请求0,未改正式连接、Provider或真实Key。来源为b74e5edc后的本地开发切片,先提交空响应修复3ab3a6e4。
状态:本文件保留开发阶段记录,现已随 **1.25.3** 独立发行并升级正式本机,见[最终验收](QA-MODEL-CATALOG-1.25.3.md)。本阶段真实模型请求0,正式连接、原Provider与凭据保持。来源为b74e5edc后的本地开发切片,先提交空响应修复3ab3a6e4。
用户要求推理可开启、选择强度,并与品牌或自定义兼容连接配套。模型中心现有入口改为“推理与能力”,新增一份共享配置档贯通自动匹配、手动覆盖、保存、导入、Run快照及实际请求。原JSON仍是同一份参数,可与控件双向编辑;其它参数、凭据引用、旧记录和池保持。导入保留推理字段、输出字段和能力声明,来源变更需重新预览,本地编辑冲突保留。
@@ -26,4 +26,4 @@
- 新推理网页6组、实际Electron6组(各9自有HTTP请求、0真实模型),1440/390px保存/刷新/默认/关闭/开启/强度/手动预算/JSON错误/保护字段/空响应反馈通过,原图实际回看。
- 旧原生参数/上游错误网页6组、Messages网页7组通过;来源日志在.qa/reasoning-controls/,旧证据目录未写入。
阶段回顾:已修本地参数、状态回传与诊断缺口,尚未证明真实供应商稳定、自主Node交付、网络间歇唯一根因、重复读取收益或Purge AB/BA。Z3原空响应、DeepSeek原length、全部旧失败及预算保持,不续跑旧实验、不推送、不委派。下一独立版本冻结、归档重建/实际安装及保留状态升级。
阶段回顾:已修本地参数、状态回传与诊断缺口,尚未证明真实供应商稳定、自主Node交付、网络间歇唯一根因、重复读取收益或Purge AB/BA。Z3原空响应、DeepSeek原length、全部旧失败及预算保持,不续跑旧实验、不推送、不委派。后续独立冻结、归档重建/实际安装及保留状态升级现已完成,范围与来源差异以最终验收为准。
+2
View File
@@ -1,5 +1,7 @@
# 推理开关、强度与空响应修复
最终状态:已随1.25.3完整本地交付,正式PID36552/Provider与全部原数据保持。完整源码926/926实际来自6741510b,最终候选1bd94856仅验收驱动差异;源码ZIP657输入重建22/22、源码78组/便携126组/安装12生命周期及103UI、正式网页/原生回看通过。主1635份与QA补验16份分别封存并离线核对,见[验收](QA-MODEL-CATALOG-1.25.3.md)与[回顾](RETROSPECTIVE-MODEL-CATALOG-1.25.3.md)。本阶段真实目录和模型请求均0。以下“尚未部署/下一发行”保留为当时阶段记录。
起点:`b74e5edc`,工作树干净,正式服务仍为1.25.2。用户补充:推理应由使用者开启并选择强度,不能只接入DeepSeek关闭推理。沿用模型中心原生参数保存接口和每次实际请求封存,不另立配置体系。
1. 核对官方Chat、Responses、Messages与DeepSeek契约。按协议提供默认、关闭和开启/强度;保留高级JSON及已有参数,明确不同型号的支持范围。DeepSeek工具往返必须保留所属响应的reasoning_content,不能只发开关。
@@ -0,0 +1,36 @@
# 1.25.3 模型目录与推理配套回顾
本轮已交付上游目录搜索多选、同源凭据引用、手动选择/自动适配及按协议配套的推理控件;正式本机为 1.25.3,原 PID36552/instance/Provider 与全部原数据保持。详细结果、首次失败、身份及证据见[验收报告](QA-MODEL-CATALOG-1.25.3.md)。
## 原问题与结果
之前的模型中心主要要求手填型号,只提供高级 JSON,用户看不到推理模式/强度。DeepSeek 原样本因默认推理耗尽 2048 输出而 length,Messages 空响应被解释成检查未完成。这些证据说明本地兼容和交互存在缺口,不能笼统归为模型行为或网络不稳定。
本轮把获取目录作为主要新增模型路径,参数配置档贯通品牌、自定义兼容、导入、JSON 和实际请求;修复所属推理状态回传与空响应诊断。真实 HTTP 夹具验证参数、历史、凭据代次、选路、取消和错误,桌面及手机入口验证保存刷新和失败恢复。未增加真实模型消费,也未降低原预算、权限、停止或原要求检查。
“从上游同步”需要区分型号目录、上游明确声明的能力及本地实际执行能力。目录缺字段时保留未知;声明没有变成实测。尚未接入的音视频、服务端工具和结构化输出执行不因有字段就显示可用。自动适配复用现有能力/健康/并发/池策略,未把它包装成已验证的语义路由。
## 验证为何分开记录
完整 926/926 实际来自 6741510b;最终发行来自 1bd94856,只有两个验收驱动变化,产品/测试及全部编译字节相同。独立源码 ZIP 657 输入重建、三归档、真实便携、源码 78 组、便携 126 组和真实安装 12 生命周期/103 UI 都在最终候选上完成。最终 QA 元数据修复在安装和主封存之后另记,不改写候选身份。
验收中确实发现当前版本兼容清单和控制引擎旧断言遗漏,前者导致首完整源码 923/926,后者拦住便携。第一处用随包版本配套并保留未知拒绝修复,第二处补验收范围;保留原失败、复核差异后重新冻结,而不是放行未验载荷。
旧 fidelity 已完成 6 组却在元数据汇总失败,说明 exit=0 或裸 success 不足以绑定版本。先以载荷归属和 22 输出 SHA 追加复核,明确原结果缺 health;封存后永久补实际 health 字段,再对同便携跑 6 组/10 图。Messages 的实际 `testedVersion` 得到兼容,错误、缺失及冲突别名仍拒绝。没有为 QA 元数据变化重跑 926 项或重新升级正式服务。
## 原状态与证据
升级及最终回看后,75 项目/91 会话/6723 记录/14320 事件和 outbox、全原行/全导出/工件字节/公开模型中心 exact,窗口 0、9229 关闭。没有读取、复制、解密或重存真实 Key,既有连接不会因本轮验收被切换或保存。
主封存 1635 份/138998818 字节,QA 补验 16 份/771613 字节,各有独立清单。开发的首取消、测试失败、清理挂起、版本字段收尾失败和原生越界点击均保留。离线验证不启动宿主、不重发网络/模型,不靠 `.qa` 原目录或仍存在的安装副本才能核对历史。
两个清单及外部发行身份一并计算,1654 份/140305327 字节 Git 索引原字节全部相同。日志例外和字节属性仅覆盖本轮目录,不更改旧封存记录。
## 尚未解决与下一步
- 真实供应商是否返回完整目录/元数据、是否接受具体型号的推理选项,尚未实测;官方或兼容协议说明不是该账号的验收结果。
- Z2 型号不符、Z3 空响应、DeepSeek length、间歇网络及旧高负载启动/审批超时的原结果不变。空响应诊断修复也不能让空内容成为成功答复。
- 真实自主 Node 交付、重复读取收益、真实预留异常边界和原版 Purge AB/BA 仍未证明。旧样本不续跑,后续真实实验需要另冻来源、线路、目标、预算及停止规则;Node 闭环后才 React/Python/Purge。
- Tianshu 工具 surface/前缀实测 C、提醒回读 D 与 E/F 仍按采用计划和前置独立推进,本轮模型功能交付不代表整个研究计划完成。
后续先用新冻结的当前配置验证真实目录/参数兼容,再决定有界的新模型工程样本;保留首失败与实际成本,不以无限重试寻找成功记录。本轮按用户授权本地提交,不委派、不推送。
@@ -0,0 +1,30 @@
{
"_approval-presentation-QyCkfhWb.js": {
"file": "assets/approval-presentation-QyCkfhWb.js",
"name": "approval-presentation"
},
"index.html": {
"file": "assets/desktop-ByCeFnDP.js",
"name": "desktop",
"src": "index.html",
"isEntry": true,
"imports": [
"_approval-presentation-QyCkfhWb.js"
],
"css": [
"assets/desktop-DT9Zs6lq.css"
]
},
"mobile.html": {
"file": "assets/mobile-DdtBN08G.js",
"name": "mobile",
"src": "mobile.html",
"isEntry": true,
"imports": [
"_approval-presentation-QyCkfhWb.js"
],
"css": [
"assets/mobile-joLHctQw.css"
]
}
}
File diff suppressed because one or more lines are too long
File diff suppressed because one or more lines are too long
File diff suppressed because one or more lines are too long
File diff suppressed because one or more lines are too long
File diff suppressed because one or more lines are too long
@@ -0,0 +1 @@
<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 64 64"><rect x="2" y="2" width="60" height="60" rx="14" fill="#154b4a"/><path d="M12 44l8-25h6l8 25h-6l-2-6H18l-2 6zm8-11h5l-2.5-8zm14-14h5l3 17 3-13h5l3 13 3-17h5l-5 25h-6l-2.5-13L45 44h-6z" fill="#fcfdfc"/></svg>

After

Width:  |  Height:  |  Size: 268 B

@@ -0,0 +1,8 @@
<!doctype html>
<html lang="zh-CN">
<head><meta charset="UTF-8" /><meta name="viewport" content="width=device-width, initial-scale=1.0" /><meta name="theme-color" content="#154b4a" /><link rel="icon" type="image/svg+xml" href="/favicon.svg" /><title>Agent Workbench · 本地编程工作台</title> <script type="module" crossorigin src="/assets/desktop-ByCeFnDP.js"></script>
<link rel="modulepreload" crossorigin href="/assets/approval-presentation-QyCkfhWb.js">
<link rel="stylesheet" crossorigin href="/assets/desktop-DT9Zs6lq.css">
</head>
<body><div id="root"></div></body>
</html>
@@ -0,0 +1,5 @@
[
"/assets/approval-presentation-QyCkfhWb.js",
"/assets/mobile-DdtBN08G.js",
"/assets/mobile-joLHctQw.css"
]
@@ -0,0 +1 @@
<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 192 192"><rect width="192" height="192" rx="36" fill="#172b43"/><path d="M42 55h108v82H42z" fill="none" stroke="#f3f6fa" stroke-width="9"/><path d="m60 78 16 18-16 18m32 0h35" fill="none" stroke="#8bd0b9" stroke-width="9" stroke-linecap="square"/></svg>

After

Width:  |  Height:  |  Size: 307 B

@@ -0,0 +1,12 @@
const CACHE = 'workbench-mobile-65469da75610f50cf284';
const FILES = ["/","/mobile.webmanifest","/mobile-icon.svg","/mobile-touch-icon.png","/assets/approval-presentation-QyCkfhWb.js","/assets/mobile-DdtBN08G.js","/assets/mobile-joLHctQw.css"];
self.addEventListener('install', event => event.waitUntil(caches.open(CACHE).then(cache => cache.addAll(FILES)).then(() => self.skipWaiting())));
self.addEventListener('activate', event => event.waitUntil(caches.keys().then(names => Promise.all(names.filter(name => name.startsWith('workbench-mobile-') && name !== CACHE).map(name => caches.delete(name)))).then(() => self.clients.claim())));
self.addEventListener('fetch', event => {
const url = new URL(event.request.url);
if (event.request.method !== 'GET' || url.origin !== self.location.origin || url.search || !FILES.includes(url.pathname)) return;
event.respondWith(fetch(event.request).then(response => {
if (response.ok) { const copy = response.clone(); event.waitUntil(caches.open(CACHE).then(cache => cache.put(event.request, copy))); }
return response;
}).catch(async () => (await caches.match(event.request)) || Response.error()));
});
Binary file not shown.

After

Width:  |  Height:  |  Size: 1.0 KiB

@@ -0,0 +1,5 @@
<!doctype html>
<html lang="zh-CN"><head><meta charset="UTF-8" /><meta name="viewport" content="width=device-width, initial-scale=1, viewport-fit=cover" /><meta name="theme-color" content="#172b43" /><meta name="apple-mobile-web-app-capable" content="yes" /><meta name="apple-mobile-web-app-title" content="Workbench" /><link rel="manifest" href="/mobile.webmanifest" /><link rel="icon" type="image/svg+xml" href="/mobile-icon.svg" /><link rel="apple-touch-icon" href="/mobile-touch-icon.png" /><title>Agent Workbench · 手机伴随端</title> <script type="module" crossorigin src="/assets/mobile-DdtBN08G.js"></script>
<link rel="modulepreload" crossorigin href="/assets/approval-presentation-QyCkfhWb.js">
<link rel="stylesheet" crossorigin href="/assets/mobile-joLHctQw.css">
</head><body><div id="mobile-root"></div></body></html>
@@ -0,0 +1 @@
{"id":"/","name":"Agent Workbench 手机伴随端","short_name":"Workbench","lang":"zh-CN","start_url":"/","scope":"/","display":"standalone","background_color":"#f1f4f7","theme_color":"#172b43","icons":[{"src":"/mobile-icon.svg","sizes":"any","type":"image/svg+xml","purpose":"any"},{"src":"/mobile-touch-icon.png","sizes":"256x256","type":"image/png","purpose":"any"}]}
Binary file not shown.

After

Width:  |  Height:  |  Size: 1.1 KiB

Binary file not shown.

After

Width:  |  Height:  |  Size: 1.0 KiB

File diff suppressed because it is too large Load Diff
@@ -0,0 +1,15 @@
"use strict";
// src/desktop/preload.cjs
var { contextBridge, ipcRenderer } = require("electron");
contextBridge.exposeInMainWorld("desktopWorkbench", Object.freeze({
protocol: "agent-workbench-desktop-v1",
status: () => ipcRenderer.invoke("awb:desktop-status"),
pickProject: () => ipcRenderer.invoke("awb:desktop-pick-project"),
projectFolder: (projectId) => ipcRenderer.invoke("awb:desktop-project-folder", { projectId }),
openFilePreview: (input) => ipcRenderer.invoke("awb:desktop-file-preview", input),
reviewTask: (input) => ipcRenderer.invoke("awb:desktop-review-task", input),
revealFile: (input) => ipcRenderer.invoke("awb:desktop-reveal-file", input),
stopHost: () => ipcRenderer.invoke("awb:desktop-stop-host"),
browsingLayout: (input) => ipcRenderer.invoke("awb:browsing-layout", input)
}));
@@ -0,0 +1,478 @@
// src/desktop/host-cli.mjs
import fs5 from "node:fs";
import path6 from "node:path";
import os from "node:os";
import { fileURLToPath } from "node:url";
// src/core/desktop/host-controller.mjs
import fs4 from "node:fs";
import path5 from "node:path";
import { spawn, execFileSync as execFileSync2 } from "node:child_process";
import { randomBytes, randomUUID as randomUUID2 } from "node:crypto";
import { setTimeout as delay } from "node:timers/promises";
// src/shared/contracts/desktop-host.mjs
import { createHmac, timingSafeEqual } from "node:crypto";
var HOST_PROTOCOL = "agent-workbench-host-v1";
var HOST_REGISTRATION_FILE = "host-authority-v1.json";
var hostError = (message, code = "HOST_IDENTITY_REJECTED", status = 409) => Object.assign(new Error(message), { code, status });
var canonicalHost = (value) => Array.isArray(value) ? value.map(canonicalHost) : value && typeof value === "object" ? Object.fromEntries(Object.keys(value).sort().map((key) => [key, canonicalHost(value[key])])) : value;
function hostProof(identity, challenge, secret) {
if (!/^[a-f0-9]{64}$/.test(challenge) || !Buffer.isBuffer(secret) || secret.length !== 32) throw hostError("\u5BBF\u4E3B\u6838\u9A8C\u6311\u6218\u683C\u5F0F\u65E0\u6548\u3002", "HOST_CHALLENGE_INVALID", 400);
return createHmac("sha256", secret).update(JSON.stringify(canonicalHost({ identity, challenge }))).digest("hex");
}
function verifyHostProof(reply, challenge, secret) {
if (!reply?.identity || reply.challenge !== challenge || !/^[a-f0-9]{64}$/.test(reply.proof ?? "")) throw hostError("\u5BBF\u4E3B\u6CA1\u6709\u63D0\u4F9B\u5339\u914D\u7684\u5B9E\u4F8B\u6838\u9A8C\u3002");
const actual = Buffer.from(reply.proof, "hex"), expected = Buffer.from(hostProof(reply.identity, challenge, secret), "hex");
if (!timingSafeEqual(actual, expected)) throw hostError("\u7AEF\u53E3\u54CD\u5E94\u4E0D\u5C5E\u4E8E\u767B\u8BB0\u7684\u6267\u884C\u5BBF\u4E3B\u3002");
return reply.identity;
}
// src/core/process-environment.mjs
import path from "node:path";
function projectEnvironment(extra = {}) {
const allowed = /^(path|pathext|systemroot|systemdrive|windir|comspec|temp|tmp|userprofile|homedrive|homepath|home|appdata|localappdata|programfiles(?:\(x86\))?|commonprogramfiles(?:\(x86\))?|lang|lc_all)$/i;
const environment = Object.fromEntries(Object.entries(process.env).filter(([name]) => allowed.test(name)));
const inheritedPath = Object.entries(environment).find(([name]) => name.toLowerCase() === "path")?.[1] || "";
for (const name of Object.keys(environment)) if (name.toLowerCase() === "path") delete environment[name];
return { ...environment, PATH: [path.dirname(process.execPath), inheritedPath].filter(Boolean).join(path.delimiter), ...extra };
}
// src/core/desktop/process-identity.mjs
import fs2 from "node:fs";
import path3 from "node:path";
import { execFileSync } from "node:child_process";
import { createHash } from "node:crypto";
// src/core/desktop/file-state.mjs
import fs from "node:fs";
import path2 from "node:path";
import { randomUUID } from "node:crypto";
var normalized = (value) => process.platform === "win32" ? value.toLowerCase() : value;
var samePath = (a, b) => normalized(path2.resolve(a)) === normalized(path2.resolve(b));
function assertLocalTree(absolute, { missing = false } = {}) {
const resolved = path2.resolve(absolute), parsed = path2.parse(resolved);
let current = parsed.root;
const components = resolved.slice(parsed.root.length).split(path2.sep).filter(Boolean);
for (const component of components) {
if (component.includes(":") || component.includes("\0")) throw hostError("\u672C\u673A\u72B6\u6001\u8DEF\u5F84\u683C\u5F0F\u65E0\u6548\u3002");
current = path2.join(current, component);
let stat;
try {
stat = fs.lstatSync(current);
} catch (cause) {
if (missing && cause.code === "ENOENT") return resolved;
throw cause;
}
if (stat.isSymbolicLink() || !samePath(current, fs.realpathSync.native(current))) throw hostError("\u672C\u673A\u72B6\u6001\u4E0D\u63A5\u53D7\u94FE\u63A5\u6216\u91CD\u89E3\u6790\u76EE\u5F55\u3002");
if (stat.isFile() && stat.nlink !== 1) throw hostError("\u672C\u673A\u72B6\u6001\u6587\u4EF6\u5177\u6709\u591A\u4E2A\u8DEF\u5F84\u3002");
}
return resolved;
}
function ownedChild(root, relative, { missing = false } = {}) {
if (typeof relative !== "string") throw hostError("\u672C\u673A\u72B6\u6001\u8DEF\u5F84\u65E0\u6548\u3002");
assertLocalTree(root);
const base = fs.realpathSync.native(root), target = path2.resolve(base, relative);
if (path2.isAbsolute(relative) || path2.win32.isAbsolute(relative) || relative.includes("\0") || !normalized(target).startsWith(normalized(base + path2.sep))) throw hostError("\u672C\u673A\u72B6\u6001\u8DEF\u5F84\u8D85\u51FA\u6240\u5C5E\u76EE\u5F55\u3002");
let existing = target;
while (!fs.existsSync(existing)) {
if (!missing) throw hostError("\u672C\u673A\u72B6\u6001\u6587\u4EF6\u4E0D\u5B58\u5728\u3002", "HOST_REGISTRATION_MISSING");
const parent = path2.dirname(existing);
if (parent === existing) throw hostError("\u672C\u673A\u72B6\u6001\u8DEF\u5F84\u65E0\u6CD5\u6838\u5BF9\u3002");
existing = parent;
}
const real = fs.realpathSync.native(existing);
if (!samePath(base, real) && !normalized(real).startsWith(normalized(base + path2.sep))) throw hostError("\u672C\u673A\u72B6\u6001\u94FE\u63A5\u8D85\u51FA\u6240\u5C5E\u76EE\u5F55\u3002");
if (fs.lstatSync(existing).isSymbolicLink()) throw hostError("\u672C\u673A\u72B6\u6001\u4E0D\u63A5\u53D7\u94FE\u63A5\u6216\u91CD\u89E3\u6790\u76EE\u5F55\u3002");
const stat = fs.statSync(existing);
if (stat.isFile() && stat.nlink !== 1) throw hostError("\u672C\u673A\u72B6\u6001\u6587\u4EF6\u5177\u6709\u591A\u4E2A\u8DEF\u5F84\u3002");
return assertLocalTree(target, { missing });
}
function treeFile(base, relative) {
if (typeof relative !== "string" || !relative || path2.isAbsolute(relative) || path2.win32.isAbsolute(relative) || relative.includes("\0") || relative.includes(":")) throw hostError("\u672C\u673A\u72B6\u6001\u8DEF\u5F84\u65E0\u6548\u3002");
const target = path2.resolve(base, relative);
if (!normalized(target).startsWith(normalized(base + path2.sep))) throw hostError("\u672C\u673A\u72B6\u6001\u8DEF\u5F84\u8D85\u51FA\u6240\u5C5E\u76EE\u5F55\u3002");
let stat, real;
try {
stat = fs.lstatSync(target);
real = fs.realpathSync.native(target);
} catch (cause) {
if (cause.code === "ENOENT") throw hostError("\u672C\u673A\u72B6\u6001\u6587\u4EF6\u4E0D\u5B58\u5728\u3002", "HOST_REGISTRATION_MISSING");
throw cause;
}
if (stat.isSymbolicLink() || !samePath(target, real)) throw hostError("\u672C\u673A\u72B6\u6001\u4E0D\u63A5\u53D7\u94FE\u63A5\u6216\u91CD\u89E3\u6790\u76EE\u5F55\u3002");
if (!stat.isFile()) throw hostError("\u672C\u673A\u72B6\u6001\u8DEF\u5F84\u4E0D\u662F\u5E38\u89C4\u6587\u4EF6\u3002");
if (stat.nlink !== 1) throw hostError("\u672C\u673A\u72B6\u6001\u6587\u4EF6\u5177\u6709\u591A\u4E2A\u8DEF\u5F84\u3002");
return { path: target, stat };
}
function readState(file, maximum = 4 * 1024 * 1024) {
assertLocalTree(file);
const stat = fs.lstatSync(file);
if (!stat.isFile() || stat.isSymbolicLink() || stat.nlink !== 1 || stat.size > maximum) throw hostError("\u672C\u673A\u72B6\u6001\u6587\u4EF6\u683C\u5F0F\u6216\u4F5C\u7528\u8303\u56F4\u65E0\u6548\u3002");
try {
return JSON.parse(fs.readFileSync(file, "utf8").replace(/^\uFEFF/, ""));
} catch {
throw hostError("\u672C\u673A\u72B6\u6001\u6587\u4EF6\u635F\u574F\uFF0C\u8BF7\u6838\u5BF9\u539F\u8BB0\u5F55\u3002");
}
}
function writeState(file, value) {
assertLocalTree(file, { missing: true });
const directory = path2.dirname(file), temporary = path2.join(directory, "." + path2.basename(file) + "-" + randomUUID() + ".tmp");
if (fs.existsSync(file)) readState(file);
try {
const descriptor = fs.openSync(temporary, "wx", 384);
try {
fs.writeFileSync(descriptor, JSON.stringify(value, null, 2) + "\n");
fs.fsyncSync(descriptor);
} finally {
fs.closeSync(descriptor);
}
assertLocalTree(file, { missing: true });
fs.renameSync(temporary, file);
} finally {
if (fs.existsSync(temporary)) fs.unlinkSync(temporary);
}
}
// src/core/desktop/process-identity.mjs
function windowsProcessObservationScript(pid) {
if (!Number.isInteger(pid) || pid < 1 || pid > 2147483647) throw hostError("\u5BBF\u4E3BPID\u65E0\u6548\u3002");
const query = 'Get-CimInstance Win32_Process -Filter "ProcessId=' + pid + '"';
return "$ErrorActionPreference='Stop';$ProgressPreference='SilentlyContinue';[Console]::OutputEncoding=New-Object System.Text.UTF8Encoding($false);$awbProcess=" + query + ';if($null -eq $awbProcess){Write-Output "null";exit 0};try{$awbOwner=Invoke-CimMethod -InputObject $awbProcess -MethodName GetOwnerSid;if($awbOwner.ReturnValue -ne 0 -or -not $awbOwner.Sid){throw "Owner identity unavailable"};}catch{$awbRemaining=' + query + ';if($null -eq $awbRemaining){Write-Output "null";exit 0};throw};[pscustomobject]@{pid=[int]$awbProcess.ProcessId;parentPid=[int]$awbProcess.ParentProcessId;createdAt=$awbProcess.CreationDate.ToUniversalTime().ToString("yyyy-MM-ddTHH:mm:ss.fffZ");executable=$awbProcess.ExecutablePath;userSid=$awbOwner.Sid}|ConvertTo-Json -Compress';
}
function observeWindowsProcess(pid) {
if (process.platform !== "win32") throw hostError("\u6B64\u684C\u9762\u5BBF\u4E3B\u6838\u9A8C\u9700\u8981Windows\u3002", "HOST_PLATFORM_UNSUPPORTED");
const script = windowsProcessObservationScript(pid);
try {
const shell = path3.join(process.env.SystemRoot || "C:\\Windows", "System32/WindowsPowerShell/v1.0/powershell.exe");
const output = execFileSync(
shell,
["-NoProfile", "-NonInteractive", "-EncodedCommand", Buffer.from(script, "utf16le").toString("base64")],
{ windowsHide: true, encoding: "utf8", stdio: ["pipe", "pipe", "pipe"], timeout: 15e3, maxBuffer: 65536, env: projectEnvironment() }
);
const value = JSON.parse(output.trim().replace(/^\uFEFF/, ""));
if (value === null) return null;
if (value.pid !== pid || !Number.isFinite(Date.parse(value.createdAt)) || !/^S-1-\d+(?:-\d+)+$/.test(value.userSid) || !value.executable || !fs2.existsSync(value.executable)) throw new Error("Invalid observation");
return { ...value, executable: fs2.realpathSync.native(value.executable) };
} catch {
throw hostError("\u65E0\u6CD5\u6838\u5BF9Windows\u5BBF\u4E3B\u521B\u5EFA\u3001\u7528\u6237\u6216\u53EF\u6267\u884C\u8EAB\u4EFD\uFF1B\u4E0D\u4F1A\u590D\u7528\u6216\u505C\u6B62\u672A\u77E5\u8FDB\u7A0B\u3002", "HOST_PROCESS_UNVERIFIED");
}
}
function treeFileIdentity({ path: absolute, stat }) {
if (stat.size > 256 * 1024 * 1024) throw hostError("\u5BBF\u4E3B\u53EF\u6267\u884C\u6587\u4EF6\u8EAB\u4EFD\u65E0\u6548\u3002");
const bytes = fs2.readFileSync(absolute);
if (bytes.length !== stat.size) throw hostError("\u5BBF\u4E3B\u6587\u4EF6\u5728\u6838\u5BF9\u671F\u95F4\u53D1\u751F\u53D8\u5316\u3002");
return { path: absolute, bytes: bytes.length, sha256: createHash("sha256").update(bytes).digest("hex") };
}
function binaryIdentity(file) {
assertLocalTree(file);
const absolute = fs2.realpathSync.native(file), stat = fs2.lstatSync(absolute);
if (!stat.isFile() || stat.isSymbolicLink() || stat.nlink !== 1 || stat.size > 256 * 1024 * 1024) throw hostError("\u5BBF\u4E3B\u53EF\u6267\u884C\u6587\u4EF6\u8EAB\u4EFD\u65E0\u6548\u3002");
return { path: absolute, bytes: stat.size, sha256: createHash("sha256").update(fs2.readFileSync(absolute)).digest("hex") };
}
// src/core/desktop/app-installation.mjs
import fs3 from "node:fs";
import path4 from "node:path";
import { createHash as createHash2 } from "node:crypto";
function inspectHostInstallation(directory, { full = false, onProgress = null } = {}) {
assertLocalTree(directory);
const root = fs3.realpathSync.native(path4.resolve(directory));
if (!fs3.statSync(root).isDirectory()) throw hostError("\u5BBF\u4E3B\u5B89\u88C5\u76EE\u5F55\u65E0\u6548\u3002");
const digest = (value) => createHash2("sha256").update(value).digest("hex");
const manifestFile = path4.join(root, "MANIFEST.json"), manifest = fs3.existsSync(manifestFile) ? readState(manifestFile, 16 * 1024 * 1024) : null;
if (manifest && (manifest.app !== "agent-workbench" || !Array.isArray(manifest.files) || manifest.files.length > 6e4)) throw hostError("\u5BBF\u4E3B\u53D1\u884C\u6E05\u5355\u65E0\u6548\u3002");
const version = manifest?.version ?? (fs3.existsSync(path4.join(root, "package.json")) ? readState(path4.join(root, "package.json")).version : null);
if (typeof version !== "string" || !/^\d+\.\d+\.\d+$/.test(version)) throw hostError("\u5BBF\u4E3B\u5B89\u88C5\u7248\u672C\u65E0\u6548\u3002");
const files = /* @__PURE__ */ new Set(["dist-server/server.mjs", "dist-server/hot-upgrade.mjs", "dist-client/index.html"]);
if (!manifest || fs3.existsSync(path4.join(root, "dist-server/host-launcher.mjs"))) files.add("dist-server/host-launcher.mjs");
if (fs3.existsSync(path4.join(root, "dist-server/install-manager.mjs"))) files.add("dist-server/install-manager.mjs");
for (const relative of ["dist-desktop/main.cjs", "dist-desktop/preload.cjs"]) if (fs3.existsSync(path4.join(root, relative))) files.add(relative);
const names = /* @__PURE__ */ new Map();
for (const row of manifest?.files ?? []) {
if (!row || typeof row.path !== "string" || row.path.includes("\\") || row.path.split("/").includes("..") || names.has(row.path) || !/^[a-f0-9]{64}$/.test(row.sha256 ?? "") || !Number.isSafeInteger(row.bytes) || row.bytes < 0) throw hostError("\u5BBF\u4E3B\u6E05\u5355\u5305\u542B\u91CD\u590D\u6216\u8D8A\u754C\u8F93\u5165\u3002");
names.set(row.path, row);
if (full || /^(dist-client|dist-server|dist-desktop)\//.test(row.path)) files.add(row.path);
}
if (!manifest || manifest.kind === "source") for (const entry of fs3.readdirSync(path4.join(root, "dist-client/assets"))) files.add("dist-client/assets/" + entry);
const selected = [...files].sort(), totalBytes = selected.reduce((sum, name) => sum + (names.get(name)?.bytes ?? 0), 0);
let completedFiles = 0, completedBytes = 0;
const report = (currentPath) => onProgress?.({ scope: full ? "full-manifest" : "loaded-inputs", root, totalFiles: selected.length, totalBytes: manifest ? totalBytes : null, completedFiles, completedBytes, currentPath });
report(null);
const inputs = selected.map((relative) => {
const identity = treeFileIdentity(treeFile(root, relative)), declared = names.get(relative);
if (manifest?.kind === "portable" && !declared) throw hostError("\u5B9E\u9645\u5BBF\u4E3B\u7A0B\u5E8F\u4E0D\u5728\u539F\u53D1\u884C\u6E05\u5355\u4E2D\u3002");
if (declared && (identity.bytes !== declared.bytes || identity.sha256 !== declared.sha256)) throw hostError("\u5B9E\u9645\u5BBF\u4E3B\u7A0B\u5E8F\u4E0E\u539F\u53D1\u884C\u6E05\u5355\u4E0D\u4E00\u81F4\u3002");
completedFiles++;
completedBytes += identity.bytes;
report(relative);
return { path: relative, bytes: identity.bytes, sha256: identity.sha256 };
});
assertLocalTree(root);
if (!samePath(fs3.realpathSync.native(root), root)) throw hostError("\u5BBF\u4E3B\u5B89\u88C5\u76EE\u5F55\u5728\u6838\u5BF9\u671F\u95F4\u53D1\u751F\u53D8\u5316\u3002");
const result = {
root,
version,
kind: manifest?.kind ?? "development-build",
manifestSha256: manifest ? digest(fs3.readFileSync(manifestFile)) : null,
inputs,
verification: full ? "full-manifest" : "loaded-host-and-client-inputs"
};
return { ...result, identitySha256: digest(JSON.stringify(canonicalHost(result))) };
}
// src/core/desktop/host-controller.mjs
var HOST_ENVIRONMENT = Object.freeze(["AGENT_WORKBENCH_DEEPSEEK_ROOT", "AGENT_WORKBENCH_PURGE_ROOT", "AGENT_WORKBENCH_PYTHON_PATH", "AGENT_WORKBENCH_TEST_PYTHON"]);
async function jsonResponse(response) {
if (!response.ok) throw hostError("\u672C\u673A\u5BBF\u4E3B\u63A5\u53E3\u62D2\u7EDD\u64CD\u4F5C\uFF08HTTP " + response.status + "\uFF09\u3002", "HOST_API_REJECTED", response.status);
const chunks = [];
let length = 0;
for await (const chunk of response.body) {
length += chunk.length;
if (length > 4 * 1024 * 1024) throw hostError("\u672C\u673A\u5BBF\u4E3B\u54CD\u5E94\u8FC7\u5927\u3002");
chunks.push(chunk);
}
try {
return JSON.parse(Buffer.concat(chunks).toString("utf8"));
} catch {
throw hostError("\u7AEF\u53E3\u672A\u8FD4\u56DE\u6709\u6548\u7684\u5BBF\u4E3B\u534F\u8BAE\u3002");
}
}
var HostController = class {
constructor({ appRoot: appRoot2, dataRoot: dataRoot2, port = 4318, nodePath: nodePath2 = null, timeoutMs = 45e3, environment = {} }) {
if (!Number.isInteger(port) || port < 1 || port > 65535 || typeof dataRoot2 !== "string" || !path5.isAbsolute(dataRoot2)) throw hostError("\u684C\u9762\u5BBF\u4E3B\u76EE\u5F55\u6216\u7AEF\u53E3\u65E0\u6548\u3002");
this.installation = inspectHostInstallation(appRoot2);
this.appRoot = this.installation.root;
assertLocalTree(dataRoot2, { missing: true });
fs4.mkdirSync(dataRoot2, { recursive: true });
this.dataRoot = fs4.realpathSync.native(dataRoot2);
const bundled = path5.join(this.appRoot, "runtime/node.exe");
const executable = nodePath2 ?? (fs4.existsSync(bundled) ? bundled : !process.versions.electron ? process.execPath : null);
if (!executable) throw hostError("\u684C\u9762\u5B89\u88C5\u7F3A\u5C11\u72EC\u7ACBNode\u5BBF\u4E3B\u3002");
this.node = binaryIdentity(executable);
const version = execFileSync2(this.node.path, ["--version"], { encoding: "utf8", windowsHide: true, timeout: 1e4, env: projectEnvironment() }).trim();
if (!/^v24\./.test(version)) throw hostError("\u684C\u9762\u5BBF\u4E3B\u9700\u8981\u5DF2\u7ECF\u6838\u9A8C\u7684Node24\u8FD0\u884C\u5668\u3002");
this.node.version = version;
this.port = port;
this.url = "http://127.0.0.1:" + port;
this.timeoutMs = Math.min(6e4, Math.max(1e3, timeoutMs));
const allowed = new Set(HOST_ENVIRONMENT);
if (Object.keys(environment).some((key) => !allowed.has(key) || typeof environment[key] !== "string" || !path5.isAbsolute(environment[key]))) throw hostError("\u5BBF\u4E3B\u542F\u52A8\u53EA\u63A5\u53D7\u5DF2\u58F0\u660E\u7684\u672C\u673A\u8FD0\u884C\u5668\u8DEF\u5F84\u3002");
this.environment = environment;
}
async api(route, { method = "GET", input = null, timeoutMs = 15e3 } = {}) {
if (typeof route !== "string" || !route.startsWith("/") || route.startsWith("//")) throw hostError("\u5BBF\u4E3B\u63A5\u53E3\u8DEF\u5F84\u65E0\u6548\u3002");
return jsonResponse(await fetch(this.url + "/api" + route, {
method,
headers: {
Connection: "close",
...this.token ? { "X-Workbench-Token": this.token } : {},
...input ? { "Content-Type": "application/json" } : {}
},
body: input ? JSON.stringify(input) : void 0,
signal: AbortSignal.timeout(timeoutMs)
}));
}
async health() {
try {
return await this.api("/health", { timeoutMs: 1500 });
} catch (cause) {
if (cause.cause?.code === "ECONNREFUSED" || cause.cause?.errors?.every((value) => value.code === "ECONNREFUSED")) return null;
throw hostError("\u7AEF\u53E3\u88AB\u5360\u7528\u6216\u54CD\u5E94\u65E0\u6CD5\u6838\u5BF9\uFF1B\u4E0D\u4F1A\u542F\u52A8\u53E6\u4E00\u5957\u5BBF\u4E3B\u3002", "HOST_PORT_UNVERIFIED");
}
}
async connect() {
const health = await this.health();
if (!health) return null;
if (health.app !== "agent-workbench" || health.ready !== true || !Number.isInteger(health.pid)) throw hostError("\u7AEF\u53E3\u5C5E\u4E8E\u672A\u77E5\u670D\u52A1\u3002");
const bootstrap = await this.api("/bootstrap");
if (bootstrap.version !== this.installation.version || typeof bootstrap.token !== "string") throw hostError("\u8FD0\u884C\u7248\u672C\u4E0D\u5339\u914D\uFF0C\u8BF7\u4F7F\u7528\u660E\u786E\u7684\u5347\u7EA7\u6D41\u7A0B\u3002", "HOST_VERSION_MISMATCH");
this.token = bootstrap.token;
const challenge = randomBytes(32).toString("hex"), reply = await this.api("/host/identity?challenge=" + challenge);
ownedChild(this.dataRoot, ".secrets");
const registration = readState(ownedChild(this.dataRoot, ".secrets/" + HOST_REGISTRATION_FILE), 16384);
if (registration.schema !== 1 || registration.protocol !== HOST_PROTOCOL || !/^[A-Za-z0-9+/]{43}=$/.test(registration.secret ?? "")) throw hostError("\u672C\u673A\u539F\u6CE8\u518C\u683C\u5F0F\u65E0\u6CD5\u6838\u5BF9\u3002");
const identity = verifyHostProof(reply, challenge, Buffer.from(registration.secret, "base64"));
const actual = observeWindowsProcess(identity.pid), caller = observeWindowsProcess(process.pid);
if (!actual || !caller || identity.pid !== health.pid || identity.pid !== registration.pid || identity.createdAt !== actual.createdAt || identity.createdAt !== registration.createdAt || identity.instanceId !== registration.instanceId || identity.ownerId !== registration.ownerId || identity.userSid !== actual.userSid || identity.userSid !== registration.userSid || identity.userSid !== caller.userSid || identity.protocol !== HOST_PROTOCOL || identity.version !== this.installation.version || identity.port !== this.port || !samePath(identity.dataRoot, this.dataRoot) || !samePath(identity.runtimeRoot, this.appRoot) || identity.runtimeIdentity !== this.installation.identitySha256 || !samePath(identity.executable.path, actual.executable) || identity.executable.sha256 !== this.node.sha256 || binaryIdentity(actual.executable).sha256 !== this.node.sha256) throw hostError("\u5B9E\u9645PID\u521B\u5EFA\u3001\u7528\u6237\u3001\u7A0B\u5E8F\u3001\u5B89\u88C5\u6216\u6570\u636E\u76EE\u5F55\u4E0E\u539F\u5BBF\u4E3B\u767B\u8BB0\u4E0D\u4E00\u81F4\u3002");
const info = readState(ownedChild(this.dataRoot, "service-info.json"));
if (info.pid !== identity.pid || info.instanceId !== identity.instanceId || info.createdAt !== identity.createdAt || info.port !== this.port || info.userSid !== identity.userSid || info.ownerId !== identity.ownerId || info.runtimeIdentity !== identity.runtimeIdentity || !samePath(info.appRoot, identity.appRoot) || !samePath(info.dataRoot, this.dataRoot) || !samePath(info.runtimeRoot, this.appRoot)) throw hostError("\u672C\u673A\u5BBF\u4E3B\u7D22\u5F15\u4E0E\u5B9E\u9645\u5B9E\u4F8B\u4E0D\u4E00\u81F4\u3002");
this.identity = identity;
return { url: this.url, identity };
}
async lock(action) {
const lockDirectory = ownedChild(this.dataRoot, ".desktop-launch.lock", { missing: true }), nonce = randomUUID2(), observed = observeWindowsProcess(process.pid);
const deadline = Date.now() + this.timeoutMs;
while (true) {
try {
fs4.mkdirSync(lockDirectory, { mode: 448 });
break;
} catch (cause) {
if (cause.code !== "EEXIST") throw cause;
ownedChild(this.dataRoot, ".desktop-launch.lock");
const ownerFile2 = path5.join(lockDirectory, "owner.json");
if (fs4.existsSync(ownerFile2)) {
ownedChild(this.dataRoot, ".desktop-launch.lock/owner.json");
const owner = readState(ownerFile2, 16384);
if (!Number.isInteger(owner.pid) || typeof owner.createdAt !== "string" || typeof owner.nonce !== "string") throw hostError("\u542F\u52A8\u9501\u7684\u539F\u8EAB\u4EFD\u65E0\u6CD5\u6838\u5BF9\u3002");
const active = observeWindowsProcess(owner.pid);
if (active && active.createdAt === owner.createdAt && (owner.userSid !== active.userSid || owner.userSid !== observed.userSid)) throw hostError("\u542F\u52A8\u9501\u4E0D\u5C5E\u4E8E\u5F53\u524DWindows\u7528\u6237\u3002");
if (!active || active.createdAt !== owner.createdAt) {
if (readState(ownerFile2).nonce !== owner.nonce) continue;
const archive = ownedChild(this.dataRoot, ".desktop-launch-stale-" + randomUUID2(), { missing: true });
try {
fs4.renameSync(lockDirectory, archive);
} catch (error) {
if (!["ENOENT", "EEXIST"].includes(error.code)) throw error;
}
continue;
}
} else if (Date.now() - fs4.statSync(lockDirectory).mtimeMs > 1e4) throw hostError("\u542F\u52A8\u9501\u7F3A\u5C11\u539F\u8BB0\u5F55\uFF0C\u8BF7\u6838\u5BF9\u8BE5\u76EE\u5F55\u3002");
if (Date.now() >= deadline) throw hostError("\u6B63\u5728\u7B49\u5F85\u539F\u542F\u52A8\u64CD\u4F5C\uFF0C\u8BF7\u6838\u5BF9\u5F53\u524D\u5BBF\u4E3B\u3002", "HOST_START_PENDING");
await delay(100);
}
}
const ownerFile = path5.join(lockDirectory, "owner.json");
try {
writeState(ownerFile, { schema: 1, pid: process.pid, createdAt: observed.createdAt, userSid: observed.userSid, nonce });
return await action();
} finally {
if (fs4.existsSync(ownerFile) && readState(ownerFile).nonce === nonce) {
fs4.unlinkSync(ownerFile);
fs4.rmdirSync(lockDirectory);
}
}
}
async start() {
return this.lock(async () => {
const existing = await this.connect();
if (existing) return { ...existing, reused: true };
const intentFile = ownedChild(this.dataRoot, "desktop-launch-state.json", { missing: true });
if (fs4.existsSync(intentFile)) {
const intent = readState(intentFile);
if (intent.pid && !["exited", "failed"].includes(intent.state)) {
const actual = observeWindowsProcess(intent.pid);
if (actual && actual.createdAt === intent.createdAt) throw hostError("\u539F\u542F\u52A8\u8FDB\u7A0B\u4ECD\u6D3B\u52A8\u4F46\u6CA1\u6709\u53EF\u6838\u9A8C\u63A5\u53E3\uFF0C\u8BF7\u6838\u5BF9\u539F\u65E5\u5FD7\uFF1B\u4E0D\u4F1A\u518D\u542F\u52A8\u53E6\u4E00\u8FDB\u7A0B\u3002", "HOST_START_UNVERIFIED");
}
}
const logs = ownedChild(this.dataRoot, "logs", { missing: true });
if (!fs4.existsSync(logs)) fs4.mkdirSync(logs);
const id = randomUUID2(), stdoutPath = path5.join(logs, "desktop-host-" + id + "-stdout.log"), stderrPath = path5.join(logs, "desktop-host-" + id + "-stderr.log");
let output, errors, child;
try {
output = fs4.openSync(stdoutPath, "wx", 384);
errors = fs4.openSync(stderrPath, "wx", 384);
child = spawn(this.node.path, [ownedChild(this.appRoot, "dist-server/server.mjs")], {
cwd: this.appRoot,
windowsHide: true,
detached: true,
env: projectEnvironment({ ...this.environment, AGENT_WORKBENCH_APP_DIR: this.appRoot, AGENT_WORKBENCH_DATA_DIR: this.dataRoot, AGENT_WORKBENCH_PORT: String(this.port) }),
stdio: ["ignore", output, errors]
});
} finally {
if (output !== void 0) fs4.closeSync(output);
if (errors !== void 0) fs4.closeSync(errors);
}
let failure = null;
child.on("error", (cause) => {
failure = cause;
});
child.unref();
const observedChild = child.pid ? observeWindowsProcess(child.pid) : null;
writeState(intentFile, {
schema: 1,
state: observedChild ? "starting" : "failed",
id,
pid: child.pid ?? null,
createdAt: observedChild?.createdAt ?? null,
appRoot: this.appRoot,
dataRoot: this.dataRoot,
port: this.port,
stdoutPath,
stderrPath
});
const deadline = Date.now() + this.timeoutMs;
while (Date.now() < deadline) {
if (failure || child.exitCode !== null) throw hostError("\u6240\u5C5E\u5BBF\u4E3B\u542F\u52A8\u5931\u8D25\uFF0C\u8BF7\u67E5\u770B\u672C\u673A\u542F\u52A8\u8BB0\u5F55\u3002", "HOST_START_FAILED");
await delay(100);
let health;
try {
health = await this.health();
} catch (cause) {
if (cause.code !== "HOST_PORT_UNVERIFIED") throw cause;
const waiting = observeWindowsProcess(child.pid);
if (!waiting || !observedChild || waiting.createdAt !== observedChild.createdAt || waiting.userSid !== observedChild.userSid || !samePath(waiting.executable, this.node.path)) throw hostError("\u6240\u5C5E\u5BBF\u4E3B\u542F\u52A8\u8FDB\u7A0B\u5DF2\u7ED3\u675F\u6216\u8EAB\u4EFD\u53D8\u5316\uFF0C\u8BF7\u6838\u5BF9\u539F\u65E5\u5FD7\u3002", "HOST_START_FAILED");
continue;
}
if (!health) continue;
if (health.pid !== child.pid) throw hostError("\u542F\u52A8\u671F\u95F4\u7AEF\u53E3\u88AB\u5176\u4ED6\u8FDB\u7A0B\u5360\u7528\u3002");
const connected = await this.connect();
if (connected) {
writeState(intentFile, { ...readState(intentFile), state: "ready", instanceId: connected.identity.instanceId });
return { ...connected, reused: false, logs: { stdoutPath, stderrPath } };
}
}
throw hostError("\u6240\u5C5E\u5BBF\u4E3B\u5C1A\u672A\u5B8C\u6210\u6CE8\u518C\uFF1B\u4E0D\u4F1A\u91CD\u65B0\u542F\u52A8\u6216\u7ED3\u675F\u672A\u77E5\u8FDB\u7A0B\u3002", "HOST_START_UNVERIFIED");
});
}
async status() {
if (!await this.connect()) return { state: "not-running", url: this.url };
return this.api("/host/status");
}
async stop() {
return this.lock(async () => {
const connection = await this.connect();
if (!connection) return { state: "not-running", url: this.url };
const original = connection.identity;
await this.api("/host/stop", { method: "POST", input: {
instanceId: original.instanceId,
createdAt: original.createdAt,
runtimeIdentity: original.runtimeIdentity,
confirmation: "stop-owned-host"
}, timeoutMs: this.timeoutMs });
const deadline = Date.now() + this.timeoutMs;
while (Date.now() < deadline) {
await delay(200);
const observed = observeWindowsProcess(original.pid);
if (!observed) {
if (await this.health()) throw hostError("\u539F\u5BBF\u4E3B\u5DF2\u7ED3\u675F\uFF0C\u4F46\u7AEF\u53E3\u5B58\u5728\u5176\u4ED6\u670D\u52A1\u3002");
return { state: "stopped", identity: original, processExited: true };
}
if (observed.createdAt !== original.createdAt) throw hostError("\u539FPID\u5DF2\u88AB\u590D\u7528\uFF0C\u4E0D\u4F1A\u7ED3\u675F\u65B0\u7684\u8FDB\u7A0B\u3002");
}
throw hostError("\u6240\u5C5E\u5BBF\u4E3B\u9000\u51FA\u4ECD\u5F85\u6838\u5BF9\uFF0C\u8BF7\u4FDD\u7559\u539F\u5B9E\u4F8B\u8BB0\u5F55\u3002", "HOST_STOP_UNVERIFIED");
});
}
};
// src/desktop/host-cli.mjs
var args = process.argv.slice(2);
if (args.includes("--help")) {
console.log("node host-launcher.mjs <start|status|stop> [--app-root <absolute>] [--data-root <absolute>] [--node <absolute>] [--port <1..65535>]");
process.exit(0);
}
var command = args.shift();
if (!["start", "status", "stop"].includes(command)) throw new Error("Use --help for a supported desktop host operation.");
var options = {};
while (args.length) {
const flag = args.shift(), value = args.shift();
if (!["--app-root", "--data-root", "--node", "--port"].includes(flag) || !value || options[flag] !== void 0) throw new Error("Invalid or repeated host option.");
options[flag] = value;
}
var defaultRoot = path6.resolve(path6.dirname(fileURLToPath(import.meta.url)), "..");
var appRoot = options["--app-root"] || process.env.AGENT_WORKBENCH_APP_DIR || defaultRoot;
var dataRoot = options["--data-root"] || process.env.AGENT_WORKBENCH_DATA_DIR || path6.join(process.env.LOCALAPPDATA || path6.join(os.homedir(), ".local/share"), "AgentWorkbench");
var nodePath = options["--node"] || (fs5.existsSync(path6.join(appRoot, "runtime/node.exe")) ? path6.join(appRoot, "runtime/node.exe") : process.execPath);
try {
const controller = new HostController({
appRoot,
dataRoot,
nodePath,
port: Number(options["--port"] || process.env.AGENT_WORKBENCH_PORT || 4318),
environment: Object.fromEntries(HOST_ENVIRONMENT.filter((key) => process.env[key]).map((key) => [key, process.env[key]]))
});
console.log(JSON.stringify(await controller[command]()));
} catch (cause) {
console.error(JSON.stringify({ success: false, code: cause.code || "HOST_FAILED", message: cause.message }));
process.exitCode = 1;
}
File diff suppressed because one or more lines are too long
File diff suppressed because it is too large Load Diff
File diff suppressed because one or more lines are too long
@@ -0,0 +1,49 @@
import assert from 'node:assert/strict';
export const groups = {
'source-reasoning': 6, 'source-reasoning-desktop': 6, 'source-catalog': 8, 'source-catalog-desktop': 8,
'source-a': 7, 'source-a-desktop': 3, 'source-b': 9, 'source-b-desktop': 9, 'source-c': 6, 'source-c-desktop': 6, 'source-d': 5, 'source-d-desktop': 5,
'portable-reasoning': 6, 'portable-reasoning-desktop': 6, 'portable-catalog': 8, 'portable-catalog-desktop': 8,
'portable-fidelity': 6, 'portable-messages': 7,
'portable-a': 7, 'portable-a-desktop': 3, 'portable-b': 9, 'portable-b-desktop': 9, 'portable-c': 6, 'portable-c-desktop': 6, 'portable-d': 5, 'portable-d-desktop': 5,
'portable-files': 9, 'portable-results': 16, 'portable-reading': 10,
};
export const stages = ['rebuild', 'archives', 'portable', ...Object.keys(groups), 'install'];
export const installedGroups = [16, 9, 7, 3, 9, 9, 6, 6, 5, 5, 6, 6, 8, 8];
export function suiteStats(text) {
const stats = {};
for (const name of ['tests', 'pass', 'fail', 'cancelled', 'skipped', 'todo', 'duration_ms']) {
const match = text.match(new RegExp('^[ℹ#] ' + name + ' ([0-9.]+)\\s*$', 'm'));
assert.ok(match, 'Missing complete test footer: ' + name); stats[name] = Number(match[1]);
}
return stats;
}
export function checkReceipt(stage, receipt, freeze, candidate) {
assert.equal(receipt.success, true, stage); assert.equal(receipt.sourceCommit, freeze.sourceCommit);
assert.equal(receipt.candidate, candidate.portableRoot); assert.equal(receipt.realModelRequests, 0);
const proof = receipt.result;
assert.equal(proof.success, true, stage);
if (groups[stage]) {
assert.equal((proof.checks ?? proof.passed).length, groups[stage], stage);
assert.deepEqual(proof.errors, [], stage); assert.equal(proof.realModelRequests, 0);
if (stage !== 'portable-reading') assert.equal(proof.version, '1.25.3');
}
if (stage === 'rebuild') {
assert.equal(proof.sourceCommit, freeze.sourceCommit); assert.equal(proof.inputCount, freeze.inputs.length + 5);
assert.equal(proof.compiled.length, 22); assert.deepEqual(proof.mismatches, []);
for (const row of freeze.inputs) assert.equal(proof.inputs.find(item => item.path === row.path)?.sha256, row.sha256, row.path);
for (const row of proof.compiled) assert.deepEqual(row.actual, row.expected);
}
if (stage === 'archives') for (const kind of ['portable', 'source', 'evidence']) {
const archive = proof.archives.find(row => row.kind === kind);
assert.equal(archive.crcPassed, true); assert.equal(archive.manifestAndDirectoryMatch, true); assert.equal(archive.sha256, candidate[kind].sha256);
}
if (stage === 'install') {
assert.equal(proof.results.length, 12); assert.ok(proof.results.every(row => row.passed));
assert.equal(proof.installedUI.proofs.length, installedGroups.length);
for (const [index, item] of proof.installedUI.proofs.entries()) {
assert.equal(item.proof.success, true); assert.equal(item.proof.version, '1.25.3');
assert.equal((item.proof.checks ?? item.proof.passed).length, installedGroups[index]);
assert.deepEqual(item.proof.errors, []); assert.equal(item.proof.realModelRequests, 0);
}
}
}
@@ -0,0 +1,20 @@
> local-agent-workbench@1.25.3 build
> tsc --noEmit && vite build && node scripts/build-mobile.mjs && node scripts/build-server.mjs
vite v7.3.7 building client environment for production...
transforming...
✓ 1770 modules transformed.
rendering chunks...
computing gzip size...
dist-client/index.html 0.60 kB │ gzip: 0.41 kB
dist-client/.vite/manifest.json 0.65 kB │ gzip: 0.24 kB
dist-client/mobile.html 0.84 kB │ gzip: 0.47 kB
dist-client/assets/mobile-joLHctQw.css 12.39 kB │ gzip: 3.22 kB
dist-client/assets/desktop-DT9Zs6lq.css 178.76 kB │ gzip: 33.75 kB
dist-client/assets/mobile-DdtBN08G.js 26.97 kB │ gzip: 9.31 kB
dist-client/assets/approval-presentation-QyCkfhWb.js 227.03 kB │ gzip: 71.74 kB
dist-client/assets/desktop-ByCeFnDP.js 456.90 kB │ gzip: 129.54 kB
✓ built in 1.29s
Mobile shell built: 3 allowed assets; API and project data are never cached.
Server bundle built.
@@ -0,0 +1,17 @@
import fs from 'node:fs';
import path from 'node:path';
import assert from 'node:assert/strict';
import { createHash } from 'node:crypto';
const root = '.qa/reasoning-1.25.3';
const read = file => JSON.parse(fs.readFileSync(path.join(root, file), 'utf8'));
const before = read('release-fixed/freeze.json'), after = read('release-verified/freeze.json');
assert.equal(before.sourceCommit, '6741510b2dc9dd1191ae7f459474a981ed1d07c9');
assert.deepEqual(before.inputs.map(row => row.path), after.inputs.map(row => row.path));
const changes = after.inputs.flatMap(row => { const old = before.inputs.find(item => item.path === row.path); return old.sha256 === row.sha256 ? [] : [{ path: row.path, beforeSha256: old.sha256, afterSha256: row.sha256 }]; });
assert.deepEqual(changes.map(row => row.path), ['scripts/reasoning-release.mjs', 'scripts/verify-controlled-release.mjs']);
assert.deepEqual(after.compiled, before.compiled);
for (const row of after.inputs) assert.equal(createHash('sha256').update(fs.readFileSync(row.path)).digest('hex'), row.sha256, row.path);
const result = { success: true, before: before.sourceCommit, after: after.sourceCommit, changes, compiledExact: after.compiled.length,
unchangedProductInputs: true, unchangedTestInputs: true, fullSuiteSourceCommit: before.sourceCommit,
scope: 'Only two acceptance scripts changed after the 926/926 source run: current-version acceptance and a fresh write-once release stage. All product/test/build inputs and all 22 compiled files remain byte-identical. Final source ZIP still requires independent rebuild, archive, portable and installed acceptance.' };
fs.writeFileSync(path.join(root, 'candidate-input-delta.json'), JSON.stringify(result, null, 2) + '\n', { flag: 'wx' }); console.log(JSON.stringify(result));
@@ -0,0 +1,22 @@
{
"success": true,
"before": "6741510b2dc9dd1191ae7f459474a981ed1d07c9",
"after": "1bd94856d31a4d122eb03a9d7d199b42dbabacc6",
"changes": [
{
"path": "scripts/reasoning-release.mjs",
"beforeSha256": "002440c9226c7824fe49725abd9d027527373adfa9c5dc144c35ffe3bd934fa4",
"afterSha256": "ce38a40fb9140f1be8a305b7e0226c4c8fa3d4162f192c03c9132a55b81e2523"
},
{
"path": "scripts/verify-controlled-release.mjs",
"beforeSha256": "dd24176e7173350b514df79f255d86daffee1a037dd5d9df4e71e1db414a8a55",
"afterSha256": "d80b2f410214662b040ec7dcaf508721aafe96a51f6667aa7ed609b545d909d1"
}
],
"compiledExact": 22,
"unchangedProductInputs": true,
"unchangedTestInputs": true,
"fullSuiteSourceCommit": "6741510b2dc9dd1191ae7f459474a981ed1d07c9",
"scope": "Only two acceptance scripts changed after the 926/926 source run: current-version acceptance and a fresh write-once release stage. All product/test/build inputs and all 22 compiled files remain byte-identical. Final source ZIP still requires independent rebuild, archive, portable and installed acceptance."
}
@@ -0,0 +1,20 @@
> local-agent-workbench@1.25.3 build
> tsc --noEmit && vite build && node scripts/build-mobile.mjs && node scripts/build-server.mjs
vite v7.3.7 building client environment for production...
transforming...
✓ 1769 modules transformed.
rendering chunks...
computing gzip size...
dist-client/index.html 0.60 kB │ gzip: 0.41 kB
dist-client/.vite/manifest.json 0.65 kB │ gzip: 0.24 kB
dist-client/mobile.html 0.84 kB │ gzip: 0.47 kB
dist-client/assets/mobile-joLHctQw.css 12.39 kB │ gzip: 3.22 kB
dist-client/assets/desktop-BS17Pjhl.css 176.98 kB │ gzip: 33.46 kB
dist-client/assets/mobile-DdtBN08G.js 26.97 kB │ gzip: 9.31 kB
dist-client/assets/approval-presentation-QyCkfhWb.js 227.03 kB │ gzip: 71.74 kB
dist-client/assets/desktop-hMug9Qj3.js 444.55 kB │ gzip: 126.20 kB
✓ built in 1.32s
Mobile shell built: 3 allowed assets; API and project data are never cached.
Server bundle built.
@@ -0,0 +1,20 @@
> local-agent-workbench@1.25.2 build
> tsc --noEmit && vite build && node scripts/build-mobile.mjs && node scripts/build-server.mjs
vite v7.3.7 building client environment for production...
transforming...
✓ 1769 modules transformed.
rendering chunks...
computing gzip size...
dist-client/index.html 0.60 kB │ gzip: 0.41 kB
dist-client/.vite/manifest.json 0.65 kB │ gzip: 0.24 kB
dist-client/mobile.html 0.84 kB │ gzip: 0.47 kB
dist-client/assets/mobile-joLHctQw.css 12.39 kB │ gzip: 3.22 kB
dist-client/assets/desktop-BS17Pjhl.css 176.98 kB │ gzip: 33.46 kB
dist-client/assets/mobile-DdtBN08G.js 26.97 kB │ gzip: 9.31 kB
dist-client/assets/approval-presentation-QyCkfhWb.js 227.03 kB │ gzip: 71.74 kB
dist-client/assets/desktop-B2BXMdVg.js 444.50 kB │ gzip: 126.17 kB
✓ built in 1.57s
Mobile shell built: 3 allowed assets; API and project data are never cached.
Server bundle built.
@@ -0,0 +1,20 @@
> local-agent-workbench@1.25.2 build
> tsc --noEmit && vite build && node scripts/build-mobile.mjs && node scripts/build-server.mjs
vite v7.3.7 building client environment for production...
transforming...
✓ 1769 modules transformed.
rendering chunks...
computing gzip size...
dist-client/index.html 0.60 kB │ gzip: 0.41 kB
dist-client/.vite/manifest.json 0.65 kB │ gzip: 0.24 kB
dist-client/mobile.html 0.84 kB │ gzip: 0.47 kB
dist-client/assets/mobile-joLHctQw.css 12.39 kB │ gzip: 3.22 kB
dist-client/assets/desktop-BS17Pjhl.css 176.98 kB │ gzip: 33.46 kB
dist-client/assets/mobile-DdtBN08G.js 26.97 kB │ gzip: 9.31 kB
dist-client/assets/approval-presentation-QyCkfhWb.js 227.03 kB │ gzip: 71.74 kB
dist-client/assets/desktop-Cgrdhtk2.js 444.54 kB │ gzip: 126.20 kB
✓ built in 1.27s
Mobile shell built: 3 allowed assets; API and project data are never cached.
Server bundle built.
@@ -0,0 +1,20 @@
> local-agent-workbench@1.25.2 build
> tsc --noEmit && vite build && node scripts/build-mobile.mjs && node scripts/build-server.mjs
vite v7.3.7 building client environment for production...
transforming...
✓ 1769 modules transformed.
rendering chunks...
computing gzip size...
dist-client/index.html 0.60 kB │ gzip: 0.41 kB
dist-client/.vite/manifest.json 0.65 kB │ gzip: 0.24 kB
dist-client/mobile.html 0.84 kB │ gzip: 0.47 kB
dist-client/assets/mobile-joLHctQw.css 12.39 kB │ gzip: 3.22 kB
dist-client/assets/desktop-BS17Pjhl.css 176.98 kB │ gzip: 33.46 kB
dist-client/assets/mobile-DdtBN08G.js 26.97 kB │ gzip: 9.31 kB
dist-client/assets/approval-presentation-QyCkfhWb.js 227.03 kB │ gzip: 71.74 kB
dist-client/assets/desktop-hMug9Qj3.js 444.55 kB │ gzip: 126.20 kB
✓ built in 1.30s
Mobile shell built: 3 allowed assets; API and project data are never cached.
Server bundle built.
@@ -0,0 +1,9 @@
> local-agent-workbench@1.25.2 build
> tsc --noEmit && vite build && node scripts/build-mobile.mjs && node scripts/build-server.mjs
src/client/NativeParameters.tsx(3,81): error TS7016: Could not find a declaration file for module '../shared/contracts/model-parameter-profiles.mjs'. 'D:/Ai-project/REA逆向/Agent-Workbench/src/shared/contracts/model-parameter-profiles.mjs' implicitly has an 'any' type.
src/client/NativeParameters.tsx(20,98): error TS2554: Expected 0-3 arguments, but got 4.
src/client/ReasoningSettings.tsx(1,118): error TS7016: Could not find a declaration file for module '../shared/contracts/model-parameter-profiles.mjs'. 'D:/Ai-project/REA逆向/Agent-Workbench/src/shared/contracts/model-parameter-profiles.mjs' implicitly has an 'any' type.
src/client/ReasoningSettings.tsx(17,53): error TS7006: Parameter 'value' implicitly has an 'any' type.
src/client/ReasoningSettings.tsx(37,116): error TS7006: Parameter 'row' implicitly has an 'any' type.
@@ -0,0 +1,20 @@
> local-agent-workbench@1.25.3 build
> tsc --noEmit && vite build && node scripts/build-mobile.mjs && node scripts/build-server.mjs
vite v7.3.7 building client environment for production...
transforming...
✓ 1770 modules transformed.
rendering chunks...
computing gzip size...
dist-client/index.html 0.60 kB │ gzip: 0.41 kB
dist-client/.vite/manifest.json 0.65 kB │ gzip: 0.24 kB
dist-client/mobile.html 0.84 kB │ gzip: 0.47 kB
dist-client/assets/mobile-joLHctQw.css 12.39 kB │ gzip: 3.22 kB
dist-client/assets/desktop-DT9Zs6lq.css 178.76 kB │ gzip: 33.75 kB
dist-client/assets/mobile-DdtBN08G.js 26.97 kB │ gzip: 9.31 kB
dist-client/assets/approval-presentation-QyCkfhWb.js 227.03 kB │ gzip: 71.74 kB
dist-client/assets/desktop-B1oRE1ea.js 456.85 kB │ gzip: 129.52 kB
✓ built in 1.27s
Mobile shell built: 3 allowed assets; API and project data are never cached.
Server bundle built.
@@ -0,0 +1,20 @@
> local-agent-workbench@1.25.3 build
> tsc --noEmit && vite build && node scripts/build-mobile.mjs && node scripts/build-server.mjs
vite v7.3.7 building client environment for production...
transforming...
✓ 1770 modules transformed.
rendering chunks...
computing gzip size...
dist-client/index.html 0.60 kB │ gzip: 0.41 kB
dist-client/.vite/manifest.json 0.65 kB │ gzip: 0.24 kB
dist-client/mobile.html 0.84 kB │ gzip: 0.47 kB
dist-client/assets/mobile-joLHctQw.css 12.39 kB │ gzip: 3.22 kB
dist-client/assets/desktop-DT9Zs6lq.css 178.76 kB │ gzip: 33.75 kB
dist-client/assets/mobile-DdtBN08G.js 26.97 kB │ gzip: 9.31 kB
dist-client/assets/approval-presentation-QyCkfhWb.js 227.03 kB │ gzip: 71.74 kB
dist-client/assets/desktop-ByCeFnDP.js 456.90 kB │ gzip: 129.54 kB
✓ built in 1.30s
Mobile shell built: 3 allowed assets; API and project data are never cached.
Server bundle built.
@@ -0,0 +1,20 @@
> local-agent-workbench@1.25.3 build
> tsc --noEmit && vite build && node scripts/build-mobile.mjs && node scripts/build-server.mjs
vite v7.3.7 building client environment for production...
transforming...
✓ 1770 modules transformed.
rendering chunks...
computing gzip size...
dist-client/index.html 0.60 kB │ gzip: 0.41 kB
dist-client/.vite/manifest.json 0.65 kB │ gzip: 0.24 kB
dist-client/mobile.html 0.84 kB │ gzip: 0.47 kB
dist-client/assets/mobile-joLHctQw.css 12.39 kB │ gzip: 3.22 kB
dist-client/assets/desktop-DT9Zs6lq.css 178.76 kB │ gzip: 33.75 kB
dist-client/assets/mobile-DdtBN08G.js 26.97 kB │ gzip: 9.31 kB
dist-client/assets/approval-presentation-QyCkfhWb.js 227.03 kB │ gzip: 71.74 kB
dist-client/assets/desktop-ByCeFnDP.js 456.90 kB │ gzip: 129.54 kB
✓ built in 1.25s
Mobile shell built: 3 allowed assets; API and project data are never cached.
Server bundle built.
@@ -0,0 +1,20 @@
> local-agent-workbench@1.25.3 build
> tsc --noEmit && vite build && node scripts/build-mobile.mjs && node scripts/build-server.mjs
vite v7.3.7 building client environment for production...
transforming...
✓ 1770 modules transformed.
rendering chunks...
computing gzip size...
dist-client/index.html 0.60 kB │ gzip: 0.41 kB
dist-client/.vite/manifest.json 0.65 kB │ gzip: 0.24 kB
dist-client/mobile.html 0.84 kB │ gzip: 0.47 kB
dist-client/assets/mobile-joLHctQw.css 12.39 kB │ gzip: 3.22 kB
dist-client/assets/desktop-DT9Zs6lq.css 178.76 kB │ gzip: 33.75 kB
dist-client/assets/mobile-DdtBN08G.js 26.97 kB │ gzip: 9.31 kB
dist-client/assets/approval-presentation-QyCkfhWb.js 227.03 kB │ gzip: 71.74 kB
dist-client/assets/desktop-ByCeFnDP.js 456.90 kB │ gzip: 129.54 kB
✓ built in 1.26s
Mobile shell built: 3 allowed assets; API and project data are never cached.
Server bundle built.
@@ -0,0 +1,19 @@
✔ upstream declarations include brand metadata and remain distinct from adapter support or verification (1.2515ms)
✔ new upstream retrieves more than 100 models without a manual model; choosing several stores one key and frozen metadata (674.9784ms)
✔ Responses and DeepSeek list use bearer; Messages root and v1 endpoints paginate with selected authentication (575.9843ms)
✔ refresh preserves existing key objects, encrypted block, parameters and user capability overrides without re-encryption (1663.0548ms)
✔ shared credentials restore from the original saved block; source disable, endpoint changes and key replacement cannot revive frozen runs (3172.4491ms)
✔ automatic selection reuses capability and health filtering, manual selection sends the exact chosen model and shared auth (784.1553ms)
✔ unknown catalog capabilities require explicit declarations; upstream negative and retired models cannot be overridden (566.2462ms)
✔ model reasoning controls use declared effort and mode ranges without silently enabling unsupported options (600.1784ms)
✔ cancelled, expired, changed-source and failed catalogs cannot save or overwrite a later selection (751.0821ms)
✔ pagination loops, conflicting models and upstream errors fail once without creating members or model attempts (633.5754ms)
✔ limits and transaction failures preserve original registry, config, cipher and preview for explicit correction (1402.5623ms)
ℹ tests 11
ℹ suites 0
ℹ pass 11
ℹ fail 0
ℹ cancelled 0
ℹ skipped 0
ℹ todo 0
ℹ duration_ms 10987.1744
@@ -0,0 +1,8 @@
EVIDENCE D:\Ai-project\REA逆向\Agent-Workbench\.qa\reasoning-controls\catalog-electron-fb5944e4-de27-4333-b3e0-46ebc8c98030
PASS new upstream needs no manual model; 132 paginated IDs are searchable and selectable; saving creates shared-source members and the existing automatic pool without inference
PASS reload retains the automatic pool; users can manually choose a member, see its credential source and receive only upstream-declared reasoning strengths
PASS an existing member opens its original upstream without exposing or re-entering a key and adds another model with an explicit manual default
PASS upstream error retains request ID and original reason with one GET; empty directory and cancellation preserve members, and a later explicit refresh recovers
PASS 390px search, filtered bulk selection, source feedback and save action fit and complete the same real API flow
PASS manual compatibility entry remains usable; every upstream call was a directory GET with zero model attempts
{"directory":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\catalog-electron-fb5944e4-de27-4333-b3e0-46ebc8c98030","success":true,"appRoot":"D:\\Ai-project\\REA逆向\\Agent-Workbench","actualElectron":true,"checks":["new upstream needs no manual model; 132 paginated IDs are searchable and selectable; saving creates shared-source members and the existing automatic pool without inference","reload retains the automatic pool; users can manually choose a member, see its credential source and receive only upstream-declared reasoning strengths","an existing member opens its original upstream without exposing or re-entering a key and adds another model with an explicit manual default","upstream error retains request ID and original reason with one GET; empty directory and cancellation preserve members, and a later explicit refresh recovers","390px search, filtered bulk selection, source feedback and save action fit and complete the same real API flow","manual compatibility entry remains usable; every upstream call was a directory GET with zero model attempts"],"screenshots":[{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\catalog-electron-fb5944e4-de27-4333-b3e0-46ebc8c98030\\catalog-multiselect-1440.png","sha256":"105974be74ee53127fb915b681b502b5ecc1f2b289915e37672041ce46cb6b97"},{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\catalog-electron-fb5944e4-de27-4333-b3e0-46ebc8c98030\\catalog-upstream-error.png","sha256":"6b7a111c2aa6d60d12fe2b1100f01a6a0a24205796166256c3781567cdcbfa9b"},{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\catalog-electron-fb5944e4-de27-4333-b3e0-46ebc8c98030\\catalog-list-390.png","sha256":"a5e12c74fd180ede1c8ef93d4e370870108ae5c64e86fa99bd497ddfc3719f0c"},{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\catalog-electron-fb5944e4-de27-4333-b3e0-46ebc8c98030\\catalog-saved-390.png","sha256":"3158850bd371dc4b7fcf6c4c82be2ca17cafe3d2e7e4413f31af0f5cfe17ec44"}],"errors":[],"directoryGets":9,"realModelRequests":0,"officialServiceTouched":false}
@@ -0,0 +1,10 @@
EVIDENCE D:\Ai-project\REA逆向\Agent-Workbench\.qa\reasoning-controls\catalog-electron-033081ae-e0da-4e98-9b04-043757a6c2b2
PASS new upstream needs no manual model; 132 paginated IDs are searchable and selectable; saving creates shared-source members and the existing automatic pool without inference
PASS reload retains the automatic pool; users can manually choose a member, see its credential source and receive only upstream-declared reasoning strengths
PASS an existing member opens its original upstream without exposing or re-entering a key and adds another model with an explicit manual default
PASS upstream error retains request ID and original reason with one GET; empty directory and cancellation preserve members, and a later explicit refresh recovers
PASS 390px search, filtered bulk selection, source feedback and save action fit and complete the same real API flow
PASS Anthropic brand pairs its endpoint, protocol and x-api-key pagination; upstream thinking modes restrict the actual controls
PASS DeepSeek brand pairs Chat and bearer discovery with a separate explicitly entered source key and no inference
PASS manual compatibility entry remains usable; every upstream call was a directory GET with zero model attempts
{"directory":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\catalog-electron-033081ae-e0da-4e98-9b04-043757a6c2b2","success":true,"version":"1.25.3","appRoot":"D:\\Ai-project\\REA逆向\\Agent-Workbench","actualElectron":true,"checks":["new upstream needs no manual model; 132 paginated IDs are searchable and selectable; saving creates shared-source members and the existing automatic pool without inference","reload retains the automatic pool; users can manually choose a member, see its credential source and receive only upstream-declared reasoning strengths","an existing member opens its original upstream without exposing or re-entering a key and adds another model with an explicit manual default","upstream error retains request ID and original reason with one GET; empty directory and cancellation preserve members, and a later explicit refresh recovers","390px search, filtered bulk selection, source feedback and save action fit and complete the same real API flow","Anthropic brand pairs its endpoint, protocol and x-api-key pagination; upstream thinking modes restrict the actual controls","DeepSeek brand pairs Chat and bearer discovery with a separate explicitly entered source key and no inference","manual compatibility entry remains usable; every upstream call was a directory GET with zero model attempts"],"screenshots":[{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\catalog-electron-033081ae-e0da-4e98-9b04-043757a6c2b2\\catalog-multiselect-1440.png","sha256":"105974be74ee53127fb915b681b502b5ecc1f2b289915e37672041ce46cb6b97"},{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\catalog-electron-033081ae-e0da-4e98-9b04-043757a6c2b2\\catalog-upstream-error.png","sha256":"ae70da6b3f1e2c67abd78351d16fe1623286bc017c6e9841a6abe52a410345b5"},{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\catalog-electron-033081ae-e0da-4e98-9b04-043757a6c2b2\\catalog-list-390.png","sha256":"a5e12c74fd180ede1c8ef93d4e370870108ae5c64e86fa99bd497ddfc3719f0c"},{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\catalog-electron-033081ae-e0da-4e98-9b04-043757a6c2b2\\catalog-saved-390.png","sha256":"39047a1fcffc660de102854f2de82214c0837836840e2ed8954f7aa82605b2fb"}],"errors":[],"directoryGets":13,"realModelRequests":0,"officialServiceTouched":false}
@@ -0,0 +1,12 @@
✔ hot upgrade adopts a retained virtual data root without reopening SQLite or changing rows, Provider or routing (1334.3286ms)
✔ failed adoption restores the original virtual data-root name and original open database (579.7528ms)
✔ hot upgrade refreshes retained model and importer methods while preserving credential and routing objects (626.688ms)
✔ hot upgrade refuses active probes and restores retained model methods if installation fails (730.109ms)
ℹ tests 4
ℹ suites 0
ℹ pass 4
ℹ fail 0
ℹ cancelled 0
ℹ skipped 0
ℹ todo 0
ℹ duration_ms 3576.7444
@@ -0,0 +1,56 @@
✔ a Windows virtual data-directory name binds new storage and credentials to its actual physical root (2495.784ms)
✔ physical data-root adoption still rejects explicit junctions at the directory or any ancestor (728.4078ms)
✔ actual Windows CurrentUser DPAPI roundtrips owned UTF-8 and rejects damaged bytes and wrong entropy (2244.5222ms)
✔ only explicitly supplied keys persist; new provider restores original revisions and public views contain no key (2518.6887ms)
✔ registry and cipher commit failure restores original routing, keys, block and every database row (2025.9052ms)
✔ a two-key import with a protection failure rolls back all registry and cipher entries and retains its preview (2369.0426ms)
✔ connection revision changes retain the original saved block and report a source mismatch on restart (2567.9211ms)
✔ disable skips real restart loading, explicit enable restores, delete preserves memory and forget removes both (2527.2289ms)
✔ corrupt saved block and another SID do not load; failures are visible without returning ciphertext or plaintext (2816.6017ms)
✔ legacy persistence requires a new explicit key and binds config and credential revisions; stale requests cannot save (2914.6037ms)
✔ owned plaintext keys never reach SQLite pages, WAL, config or public execution exports (1692.3592ms)
✔ upstream declarations include brand metadata and remain distinct from adapter support or verification (1.5406ms)
✔ new upstream retrieves more than 100 models without a manual model; choosing several stores one key and frozen metadata (1305.5615ms)
✔ Responses and DeepSeek list use bearer; Messages root and v1 endpoints paginate with selected authentication (751.6201ms)
✔ refresh preserves existing key objects, encrypted block, parameters and user capability overrides without re-encryption (1868.7089ms)
✔ shared credentials restore from the original saved block; source disable, endpoint changes and key replacement cannot revive frozen runs (3209.679ms)
✔ automatic selection reuses capability and health filtering, manual selection sends the exact chosen model and shared auth (814.1125ms)
✔ unknown catalog capabilities require explicit declarations; upstream negative and retired models cannot be overridden (982.3367ms)
✔ model reasoning controls use declared effort and mode ranges without silently enabling unsupported options (577.3494ms)
✔ cancelled, expired, changed-source and failed catalogs cannot save or overwrite a later selection (851.3064ms)
✔ pagination loops, conflicting models and upstream errors fail once without creating members or model attempts (858.696ms)
✔ limits and transaction failures preserve original registry, config, cipher and preview for explicit correction (1481.9399ms)
✔ a model that switches from a shared to an explicit credential receives a new revision for opaque-history isolation (886.558ms)
✔ compiled HTTP directory flow persists selection and references, gates changes during fetch and permits explicit cancel (2314.8799ms)
✔ large directories declare partial coverage and a batch crossing the local connection limit leaves no partial members (4150.1461ms)
✔ model registry keeps original credentials/endpoint intact, rejects stale edits, and forgets only volatile keys on restart (1240.6498ms)
✔ config import previews exact revisions, keeps secrets volatile, syncs only when enabled and preserves manual edits on conflicts (837.2492ms)
✔ real HTTP pool fails over 429 to another model and never repeats completed tools; attempts seal actual request bytes and Run identity (3983.9481ms)
✔ pool rotation and weights select members on new Runs; disabled and missing credentials are excluded without requests (7659.6501ms)
✔ 401 blocks the credential revision; updating the key restores availability without leaking error echoes (2739.0951ms)
✔ queued request cancellation and in-flight disable release only owned leases and do not dispatch another HTTP request (1104.7025ms)
✔ compiled model-center HTTP API persists public profiles, blocks forged origins, and local operations never invoke a model (2720.9882ms)
✔ model registry disk failures restore routing, pool profiles and credential references without partial updates (894.5153ms)
✔ model selection and saved configuration commit together and failures preserve the original mode and endpoint (912.7321ms)
✔ configuration imports roll back the entire batch and retain one usable preview after validation or commit failure (864.8288ms)
✔ capability routing skips text-only members before acquiring a lease and records exact refusal in the original Run (2716.2589ms)
✔ all incompatible members reject with no network request, no lease, preserved Task and explicit non-dispatched ledger (1763.767ms)
✔ explicit synthetic tool probe verifies only actual matching call, never executes it and loses verification on revision changes (1124.2248ms)
✔ plain text probe cannot verify tools; wrong tool nonce and client supplied evidence cannot claim verified support (1101.1036ms)
✔ a disabled connection invalidates an in-flight probe even when toggle mutates its original registry object (956.4407ms)
✔ brand and custom-compatible profiles send native reasoning controls per protocol without changing output budgets (2.0655ms)
✔ opaque Chat reasoning returns only to its original endpoint, model, connection and credential revision; bytes count against the actual request (0.7508ms)
✔ saved profile changes preserve credential references and old frozen Run settings, including the legacy connection after restart (1508.6634ms)
✔ DeepSeek JSON reasoning survives real HTTP tool rounds and restart without entering visible assistant text (5794.2788ms)
✔ DeepSeek SSE reasoning survives real HTTP tool rounds and restart without entering visible assistant text (4916.9995ms)
✔ malformed or post-terminal Chat reasoning is rejected with the raw stream retained (1.4278ms)
✔ connection import synchronizes profile, native fields and declared abilities without editing the source or manufacturing verification (817.528ms)
✔ compiled HTTP persists profile and reasoning choices for every protocol and rejects incompatible/stale edits atomically (2405.3444ms)
ℹ tests 48
ℹ suites 0
ℹ pass 48
ℹ fail 0
ℹ cancelled 0
ℹ skipped 0
ℹ todo 0
ℹ duration_ms 48535.3769
@@ -0,0 +1,76 @@
✔ a Windows virtual data-directory name binds new storage and credentials to its actual physical root (2311.0609ms)
✔ physical data-root adoption still rejects explicit junctions at the directory or any ancestor (659.6269ms)
✔ actual Windows CurrentUser DPAPI roundtrips owned UTF-8 and rejects damaged bytes and wrong entropy (1578.6055ms)
✔ only explicitly supplied keys persist; new provider restores original revisions and public views contain no key (2564.618ms)
✔ registry and cipher commit failure restores original routing, keys, block and every database row (2020.0751ms)
✔ a two-key import with a protection failure rolls back all registry and cipher entries and retains its preview (2333.4619ms)
✔ connection revision changes retain the original saved block and report a source mismatch on restart (2532.5859ms)
✔ disable skips real restart loading, explicit enable restores, delete preserves memory and forget removes both (2807.4052ms)
✔ corrupt saved block and another SID do not load; failures are visible without returning ciphertext or plaintext (2755.1485ms)
✔ legacy persistence requires a new explicit key and binds config and credential revisions; stale requests cannot save (3177.2069ms)
✔ owned plaintext keys never reach SQLite pages, WAL, config or public execution exports (1831.2138ms)
✔ upstream declarations include brand metadata and remain distinct from adapter support or verification (1.3517ms)
✔ new upstream retrieves more than 100 models without a manual model; choosing several stores one key and frozen metadata (1125.1682ms)
✔ Responses and DeepSeek list use bearer; Messages root and v1 endpoints paginate with selected authentication (723.1287ms)
✔ refresh preserves existing key objects, encrypted block, parameters and user capability overrides without re-encryption (1749.1142ms)
✔ shared credentials restore from the original saved block; source disable, endpoint changes and key replacement cannot revive frozen runs (2869.7182ms)
✔ automatic selection reuses capability and health filtering, manual selection sends the exact chosen model and shared auth (1031.656ms)
✔ unknown catalog capabilities require explicit declarations; upstream negative and retired models cannot be overridden (746.8593ms)
✔ model reasoning controls use declared effort and mode ranges without silently enabling unsupported options (606.5743ms)
✔ cancelled, expired, changed-source and failed catalogs cannot save or overwrite a later selection (1021.1965ms)
✔ pagination loops, conflicting models and upstream errors fail once without creating members or model attempts (504.7702ms)
✔ limits and transaction failures preserve original registry, config, cipher and preview for explicit correction (1599.1864ms)
✔ a model that switches from a shared to an explicit credential receives a new revision for opaque-history isolation (780.5976ms)
✔ compiled HTTP directory flow persists selection and references, gates changes during fetch and permits explicit cancel (2482.5708ms)
✔ large directories declare partial coverage and a batch crossing the local connection limit leaves no partial members (4426.6942ms)
✔ a distinct upstream account at the same endpoint and model never silently reuses another account (1126.2161ms)
✔ source changes invalidate an in-flight child probe and host shutdown cancels directory work and drops transient previews (812.9105ms)
✔ model registry keeps original credentials/endpoint intact, rejects stale edits, and forgets only volatile keys on restart (1034.414ms)
✔ config import previews exact revisions, keeps secrets volatile, syncs only when enabled and preserves manual edits on conflicts (1036.3585ms)
✔ real HTTP pool fails over 429 to another model and never repeats completed tools; attempts seal actual request bytes and Run identity (4826.1445ms)
✔ pool rotation and weights select members on new Runs; disabled and missing credentials are excluded without requests (7587.4325ms)
✔ 401 blocks the credential revision; updating the key restores availability without leaking error echoes (2037.6933ms)
✔ queued request cancellation and in-flight disable release only owned leases and do not dispatch another HTTP request (1060.4963ms)
✔ compiled model-center HTTP API persists public profiles, blocks forged origins, and local operations never invoke a model (2469.3297ms)
✔ model registry disk failures restore routing, pool profiles and credential references without partial updates (948.4446ms)
✔ model selection and saved configuration commit together and failures preserve the original mode and endpoint (804.6812ms)
✔ configuration imports roll back the entire batch and retain one usable preview after validation or commit failure (1024.7339ms)
✔ capability routing skips text-only members before acquiring a lease and records exact refusal in the original Run (2584.118ms)
✔ all incompatible members reject with no network request, no lease, preserved Task and explicit non-dispatched ledger (1626.527ms)
✔ explicit synthetic tool probe verifies only actual matching call, never executes it and loses verification on revision changes (1257.2826ms)
✔ plain text probe cannot verify tools; wrong tool nonce and client supplied evidence cannot claim verified support (1143.9835ms)
✔ a disabled connection invalidates an in-flight probe even when toggle mutates its original registry object (1098.8162ms)
✔ brand and custom-compatible profiles send native reasoning controls per protocol without changing output budgets (1.8031ms)
✔ opaque Chat reasoning returns only to its original endpoint, model, connection and credential revision; bytes count against the actual request (0.8382ms)
✔ saved profile changes preserve credential references and old frozen Run settings, including the legacy connection after restart (1193.6698ms)
✔ DeepSeek JSON reasoning survives real HTTP tool rounds and restart without entering visible assistant text (5753.5252ms)
✔ DeepSeek SSE reasoning survives real HTTP tool rounds and restart without entering visible assistant text (5457.5109ms)
✔ malformed or post-terminal Chat reasoning is rejected with the raw stream retained (1.6173ms)
✔ connection import synchronizes profile, native fields and declared abilities without editing the source or manufacturing verification (929.9267ms)
✔ compiled HTTP persists profile and reasoning choices for every protocol and rejects incompatible/stale edits atomically (2840.5624ms)
✔ runtime guides map archived evidence while all original reports and successful/failed source bytes remain independently verifiable (772.0537ms)
✔ missing required research reports fail packaging rather than creating a successful placeholder (900.7289ms)
✔ full verification reports only verified bytes, complete declared scope and actual directory inventory; changed/deleted files still fail (942.0458ms)
✔ preflight and activation continue one monotonic operation bound to installer hash; terminal replay and changed origin are rejected (8419.8243ms)
✔ preflight failures and fixed-candidate mismatches retain the actual reason without publishing a pointer (4374.6172ms)
✔ cancel finishes only the original ready preparation, preserves data and pointer and refuses replay (6397.659ms)
✔ the installation lock rejects concurrent same-id operations before they can overwrite original progress (3714.845ms)
✔ original post-publication failure is linked to its rollback receipt and every original byte survives (3379.594ms)
✔ a candidate mutation at the publication boundary still fails even after an earlier successful full validation (6568.4973ms)
✔ public progress marks a lost process unknown without editing its original record or claiming completion (4662.7171ms)
✔ throwing console/UI progress observers cannot decide the installation result (3572.3264ms)
✔ a progress-only write failure after committed publication returns the authoritative receipt and a warning (3570.0664ms)
✔ browser and MCP approvals use actual tool categories; mobile exposes only category while keeping original fingerprint (605.6872ms)
✔ the compiled runtime version constant tracks package.json so a release bump cannot ship a stale served version (9.1867ms)
✔ hot upgrade adopts a retained virtual data root without reopening SQLite or changing rows, Provider or routing (1268.4364ms)
✔ failed adoption restores the original virtual data-root name and original open database (519.3982ms)
✔ hot upgrade refreshes retained model and importer methods while preserving credential and routing objects (604.1607ms)
✔ hot upgrade refuses active probes and restores retained model methods if installation fails (1144.8847ms)
ℹ tests 68
ℹ suites 0
ℹ pass 68
ℹ fail 0
ℹ cancelled 0
ℹ skipped 0
ℹ todo 0
ℹ duration_ms 92783.4976
@@ -0,0 +1,8 @@
EVIDENCE D:\Ai-project\REA逆向\Agent-Workbench\.qa\reasoning-controls\catalog-web-b487f3fd-7997-4055-9d67-9b4e944c5bf7
PASS new upstream needs no manual model; 132 paginated IDs are searchable and selectable; saving creates shared-source members and the existing automatic pool without inference
PASS reload retains the automatic pool; users can manually choose a member, see its credential source and receive only upstream-declared reasoning strengths
PASS an existing member opens its original upstream without exposing or re-entering a key and adds another model with an explicit manual default
PASS upstream error retains request ID and original reason with one GET; empty directory and cancellation preserve members, and a later explicit refresh recovers
PASS 390px search, filtered bulk selection, source feedback and save action fit and complete the same real API flow
PASS manual compatibility entry remains usable; every upstream call was a directory GET with zero model attempts
{"directory":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\catalog-web-b487f3fd-7997-4055-9d67-9b4e944c5bf7","success":true,"appRoot":"D:\\Ai-project\\REA逆向\\Agent-Workbench","actualElectron":false,"checks":["new upstream needs no manual model; 132 paginated IDs are searchable and selectable; saving creates shared-source members and the existing automatic pool without inference","reload retains the automatic pool; users can manually choose a member, see its credential source and receive only upstream-declared reasoning strengths","an existing member opens its original upstream without exposing or re-entering a key and adds another model with an explicit manual default","upstream error retains request ID and original reason with one GET; empty directory and cancellation preserve members, and a later explicit refresh recovers","390px search, filtered bulk selection, source feedback and save action fit and complete the same real API flow","manual compatibility entry remains usable; every upstream call was a directory GET with zero model attempts"],"screenshots":[{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\catalog-web-b487f3fd-7997-4055-9d67-9b4e944c5bf7\\catalog-multiselect-1440.png","sha256":"33a22c11a9d73e4d7c9b5c678156b025926640642b687e6b708b3de2824d6e8c"},{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\catalog-web-b487f3fd-7997-4055-9d67-9b4e944c5bf7\\catalog-upstream-error.png","sha256":"8484855c5e2ea6209b4e9f757b0ff153bc020cd1dcbf6868ab4114a5787941ca"},{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\catalog-web-b487f3fd-7997-4055-9d67-9b4e944c5bf7\\catalog-list-390.png","sha256":"35484851d3c764dc1d5c433e78abba221e8a21e4b8622aaa0afb17be7657ada4"},{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\catalog-web-b487f3fd-7997-4055-9d67-9b4e944c5bf7\\catalog-saved-390.png","sha256":"05df7a9b23a4059ec0336a39d80dd31779bc92103f437791fa3677e140556692"}],"errors":[],"directoryGets":9,"realModelRequests":0,"officialServiceTouched":false}
@@ -0,0 +1,10 @@
EVIDENCE D:\Ai-project\REA逆向\Agent-Workbench\.qa\reasoning-controls\catalog-web-be6a2069-da68-46f1-9675-f8423ad6815b
PASS new upstream needs no manual model; 132 paginated IDs are searchable and selectable; saving creates shared-source members and the existing automatic pool without inference
PASS reload retains the automatic pool; users can manually choose a member, see its credential source and receive only upstream-declared reasoning strengths
PASS an existing member opens its original upstream without exposing or re-entering a key and adds another model with an explicit manual default
PASS upstream error retains request ID and original reason with one GET; empty directory and cancellation preserve members, and a later explicit refresh recovers
PASS 390px search, filtered bulk selection, source feedback and save action fit and complete the same real API flow
PASS Anthropic brand pairs its endpoint, protocol and x-api-key pagination; upstream thinking modes restrict the actual controls
PASS DeepSeek brand pairs Chat and bearer discovery with a separate explicitly entered source key and no inference
PASS manual compatibility entry remains usable; every upstream call was a directory GET with zero model attempts
{"directory":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\catalog-web-be6a2069-da68-46f1-9675-f8423ad6815b","success":true,"version":"1.25.3","appRoot":"D:\\Ai-project\\REA逆向\\Agent-Workbench","actualElectron":false,"checks":["new upstream needs no manual model; 132 paginated IDs are searchable and selectable; saving creates shared-source members and the existing automatic pool without inference","reload retains the automatic pool; users can manually choose a member, see its credential source and receive only upstream-declared reasoning strengths","an existing member opens its original upstream without exposing or re-entering a key and adds another model with an explicit manual default","upstream error retains request ID and original reason with one GET; empty directory and cancellation preserve members, and a later explicit refresh recovers","390px search, filtered bulk selection, source feedback and save action fit and complete the same real API flow","Anthropic brand pairs its endpoint, protocol and x-api-key pagination; upstream thinking modes restrict the actual controls","DeepSeek brand pairs Chat and bearer discovery with a separate explicitly entered source key and no inference","manual compatibility entry remains usable; every upstream call was a directory GET with zero model attempts"],"screenshots":[{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\catalog-web-be6a2069-da68-46f1-9675-f8423ad6815b\\catalog-multiselect-1440.png","sha256":"69ec5a2397ef17d3deb7387ec2851d066e0684f750a5c00752be963007ea0a94"},{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\catalog-web-be6a2069-da68-46f1-9675-f8423ad6815b\\catalog-upstream-error.png","sha256":"a70aaf7e0e372c1aa624ec0af244b1de634ac6de825ac0a78719781a97a29dfd"},{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\catalog-web-be6a2069-da68-46f1-9675-f8423ad6815b\\catalog-list-390.png","sha256":"35484851d3c764dc1d5c433e78abba221e8a21e4b8622aaa0afb17be7657ada4"},{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\catalog-web-be6a2069-da68-46f1-9675-f8423ad6815b\\catalog-saved-390.png","sha256":"c08f4ae73e5b1ae8aab3ca3499c62ec0f23ca503adafb1a107bdc4b6d02e83a6"}],"errors":[],"directoryGets":13,"realModelRequests":0,"officialServiceTouched":false}
@@ -0,0 +1,7 @@
EVIDENCE D:\Ai-project\REA逆向\Agent-Workbench\.qa\reasoning-controls\catalog-web-2b42b01f-5b2c-40ce-b4ad-0464af9e288d
PASS new upstream needs no manual model; 132 paginated IDs are searchable and selectable; saving creates shared-source members and the existing automatic pool without inference
PASS reload retains the automatic pool; users can manually choose a member, see its credential source and receive only upstream-declared reasoning strengths
PASS an existing member opens its original upstream without exposing or re-entering a key and adds another model with an explicit manual default
PASS upstream error retains request ID and original reason with one GET; empty directory and cancellation preserve members, and a later explicit refresh recovers
PASS 390px search, filtered bulk selection, source feedback and save action fit and complete the same real API flow
{"directory":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\catalog-web-2b42b01f-5b2c-40ce-b4ad-0464af9e288d","success":false,"appRoot":"D:\\Ai-project\\REA逆向\\Agent-Workbench","actualElectron":false,"checks":["new upstream needs no manual model; 132 paginated IDs are searchable and selectable; saving creates shared-source members and the existing automatic pool without inference","reload retains the automatic pool; users can manually choose a member, see its credential source and receive only upstream-declared reasoning strengths","an existing member opens its original upstream without exposing or re-entering a key and adds another model with an explicit manual default","upstream error retains request ID and original reason with one GET; empty directory and cancellation preserve members, and a later explicit refresh recovers","390px search, filtered bulk selection, source feedback and save action fit and complete the same real API flow"],"screenshots":[{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\catalog-web-2b42b01f-5b2c-40ce-b4ad-0464af9e288d\\catalog-multiselect-1440.png","sha256":"69ec5a2397ef17d3deb7387ec2851d066e0684f750a5c00752be963007ea0a94"},{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\catalog-web-2b42b01f-5b2c-40ce-b4ad-0464af9e288d\\catalog-upstream-error.png","sha256":"67fc6225032bfeaaa18b00fa5d9553d679fa7efc54b600914cef675280d9791e"},{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\catalog-web-2b42b01f-5b2c-40ce-b4ad-0464af9e288d\\catalog-list-390.png","sha256":"35484851d3c764dc1d5c433e78abba221e8a21e4b8622aaa0afb17be7657ada4"},{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\catalog-web-2b42b01f-5b2c-40ce-b4ad-0464af9e288d\\catalog-saved-390.png","sha256":"5a197b79ed69d39552c79a5205229e40d7755faacc0df53abe747bd34ee9e10f"},{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\catalog-web-2b42b01f-5b2c-40ce-b4ad-0464af9e288d\\failure.png","sha256":"c11a42e53ebfbeda9e496b95761008f56df640246af15bafdeec3eee2cb8f0b4"}],"errors":[],"error":"Error: expect(locator).toBeDisabled() failed\n\nLocator: getByRole('dialog', { name: '模型中心', exact: true }).getByRole('form', { name: '原生请求参数', exact: true }).getByLabel('推理模式', { exact: true }).locator('option[value=\"enabled\"]')\nExpected: disabled\nReceived: enabled\nTimeout: 5000ms\n\nCall log:\n - Expect \"toBeDisabled\" getByRole('dialog', { name: '模型中心', exact: true }).getByRole('form', { name: '原生请求参数', exact: true }).getByLabel('推理模式', { exact: true }).locator('option[value=\"enabled\"]') with timeout 5000ms\n - waiting for getByRole('dialog', { name: '模型中心', exact: true }).getByRole('form', { name: '原生请求参数', exact: true }).getByLabel('推理模式', { exact: true }).locator('option[value=\"enabled\"]')\n 14 × locator resolved to <option disabled value=\"enabled\">开启 · 指定推理预算</option>\n - unexpected value \"enabled\"\n\n at captureRawStack (D:\\Ai-project\\REA逆向\\Agent-Workbench\\node_modules\\playwright-core\\lib\\coreBundle.js:9282:17)\n at callMatcherAsStep (D:\\Ai-project\\REA逆向\\Agent-Workbench\\node_modules\\playwright\\lib\\matchers\\expect.js:13316:57)\n at Object.toBeDisabled (D:\\Ai-project\\REA逆向\\Agent-Workbench\\node_modules\\playwright\\lib\\matchers\\expect.js:13308:23)\n at D:\\Ai-project\\REA逆向\\Agent-Workbench\\scripts\\verify-model-catalog-ui.mjs:125:97","realModelRequests":0}
@@ -0,0 +1,95 @@
✔ Messages builds native system/tool history, bounds images, and seals a distinct mixed-pool variant (7.6706ms)
✔ Messages URL and native parameter contracts preserve other protocol paths and hard output budget (0.5922ms)
✔ native usage includes both cache buckets once, retains raw and rejects invalid counters (0.1714ms)
✔ real HTTP Messages JSON executes one tool, retains opaque history, actual URL and cache accounting (4374.5903ms)
✔ real HTTP Messages SSE executes one tool, retains opaque history, actual URL and cache accounting (3826.868ms)
✔ Messages api-key authentication is explicit; mismatched model remains blocked with original usage (6.6939ms)
✔ Messages bearer authentication is explicit; mismatched model remains blocked with original usage (4.354ms)
✔ Messages redirect is not followed, either credential stays at the original server (3.6699ms)
✔ strict native SSE reads the nested start model and retains mismatch source without retries (3.1668ms)
✔ a closed native tool block cannot execute before message_stop, then executes exactly once (4810.9681ms)
✔ mixed Chat/Messages pool seals both variants and preserves one completed tool on failover (4931.9047ms)
✔ native JSON rejects incomplete stops, duplicate tools and hosted content (0.2854ms)
✔ Messages rejects empty, whitespace and thinking-only answers without rejecting valid text or tool calls (0.2337ms)
✔ empty Messages JSON fails once with original response and 6692/0 usage, before verification or tools (2205.1266ms)
✔ empty Messages SSE fails once with original response and 6692/0 usage, before verification or tools (2941.9462ms)
▶ strict SSE validates all blocks, terminal, event names, UTF8 and tool JSON before dispatch
✔ reject native stream 0 (1.1827ms)
✔ reject native stream 1 (0.2338ms)
✔ reject native stream 2 (0.2415ms)
✔ reject native stream 3 (0.1981ms)
✔ reject native stream 4 (0.1741ms)
✔ reject native stream 5 (0.1716ms)
✔ reject native stream 6 (0.2182ms)
✔ reject native stream 7 (0.1652ms)
✔ reject native stream 8 (0.2068ms)
✔ strict SSE validates all blocks, terminal, event names, UTF8 and tool JSON before dispatch (3.4907ms)
✔ upstream SSE error retains original type/request ID and initial usage, never retries (3.9173ms)
✔ stop cancels the owned stream and retains partial consumption without running tool or replaying (59.3891ms)
✔ saved Messages import preserves auth choice, rejects endpoint duplication, and keeps source file untouched (798.1673ms)
✔ model registry keeps original credentials/endpoint intact, rejects stale edits, and forgets only volatile keys on restart (1477.04ms)
✔ config import previews exact revisions, keeps secrets volatile, syncs only when enabled and preserves manual edits on conflicts (1128.5644ms)
✔ real HTTP pool fails over 429 to another model and never repeats completed tools; attempts seal actual request bytes and Run identity (4406.1958ms)
✔ pool rotation and weights select members on new Runs; disabled and missing credentials are excluded without requests (7375.8046ms)
✔ 401 blocks the credential revision; updating the key restores availability without leaking error echoes (2501.3432ms)
✔ queued request cancellation and in-flight disable release only owned leases and do not dispatch another HTTP request (1145.0139ms)
✔ compiled model-center HTTP API persists public profiles, blocks forged origins, and local operations never invoke a model (2872.0722ms)
✔ model registry disk failures restore routing, pool profiles and credential references without partial updates (867.1473ms)
✔ model selection and saved configuration commit together and failures preserve the original mode and endpoint (1181.377ms)
✔ configuration imports roll back the entire batch and retain one usable preview after validation or commit failure (947.9502ms)
✔ HTTP quota retains native code/type/param/request id, bounded original and allowlisted headers, never repeats that connection (2100.4473ms)
✔ Retry-After above 90 seconds remains exact and a short local budget cannot trigger an early retry (1.7616ms)
✔ Retry-After date, expired date, huge value and invalid header retain explicit semantics (1.3698ms)
✔ unclassified 429 remains unspecified and every bounded actual retry retains its own original error (2302.9525ms)
✔ oversized HTTP error saves exactly the bounded prefix and labels its incomplete range (2032.5665ms)
✔ Chat SSE error preserves upstream fields, original failed event, partial tool and reported usage without executing or retrying (2833.3008ms)
✔ Responses failed SSE retains original event/response identity and usage (2373.9204ms)
✔ Responses incomplete SSE retains original event/response identity and usage (2765.4583ms)
✔ unsupported successful Responses output retains original JSON and usage with a local protocol origin (3098.6156ms)
✔ HTTP 200 native error is still an upstream failure rather than a fabricated successful answer (0.8005ms)
✔ JSON-escaped credential echoes are redacted in the original preview and capture as well as parsed fields (0.5817ms)
✔ quota failover uses another independent pool member and never replays a failed stream (3155.1647ms)
✔ native parameters are isolated by pool member, preserve Responses state and bind the original output reservation (0.3588ms)
✔ unsupported native fields and invalid values reject before dispatch instead of silently dropping or overriding constraints (0.7353ms)
✔ native settings for legacy and registered profiles persist without changing credentials or old records and reject stale edits (1667.9968ms)
✔ compiled native settings API saves/restores exact parameters and rejects reserved fields without mutating the profile (2747.8406ms)
✔ legacy connection test sends its saved native fields and selected output limit field without changing the configuration (3969.1194ms)
✔ HTTP download preserves original malformed SSE bytes and artifact SHA instead of decoding invalid UTF8 (2708.0402ms)
✔ terminated SSE adapts empty optional tool arguments, retains the exact wire bytes and executes the validated batch once (5158.0943ms)
✔ reported SSE usage survives rejected required arguments "" without dispatching any tool (2533.2522ms)
✔ reported SSE usage survives rejected required arguments "{\"path\":" without dispatching any tool (2470.7254ms)
✔ reported SSE usage survives rejected required arguments "null" without dispatching any tool (3058.2477ms)
✔ a stream without DONE retains reported usage but cannot adapt or execute even an optional tool (2803.5224ms)
✔ a timeout after reasoning-only fragments creates one unknown receipt and never invents reported tokens (2903.7231ms)
✔ real Chat SSE carries UTF8 and split tool JSON; Runtime executes only after DONE, stores partial progress and original usage (4014.6995ms)
✔ Responses typed SSE validates original item identity, terminal data, raw usage and tool result pair (6011.8252ms)
✔ incomplete network stream retains unknown attempt and never automatically switches pool or dispatches fragment tools (5167.2474ms)
✔ stopping an active HTTP stream closes only original request and keeps partial text without executing tools (3443.0124ms)
✔ explicit streaming probe uses actual SSE despite unknown declaration and binds its own original attempt (1070.5037ms)
✔ coalesced initial SSE events publish the latest partial state even when the upstream stays idle (2885.2758ms)
✔ stream timeout closes the original HTTP response, preserves redacted capture and never retries (1387.1712ms)
✔ stream idle timeout preserves incomplete tool arguments, records unknown use and never fails over (2242.4583ms)
✔ SSE accepts CR-only final event boundaries and retains exact invalid UTF8 bytes without dispatch (0.9561ms)
▶ stream parser rejects truncation, finish length, malformed JSON, duplicate sequence, changed final item and oversized data
✔ reject 0 (0.3139ms)
✔ reject 1 (0.1738ms)
✔ reject 2 (0.1382ms)
✔ reject 3 (5.0314ms)
✔ reject 4 (0.254ms)
✔ reject 5 (0.4498ms)
✔ stream parser rejects truncation, finish length, malformed JSON, duplicate sequence, changed final item and oversized data (6.8629ms)
✔ brand and custom-compatible profiles send native reasoning controls per protocol without changing output budgets (1.8164ms)
✔ opaque Chat reasoning returns only to its original endpoint, model, connection and credential revision; bytes count against the actual request (0.6821ms)
✔ DeepSeek JSON reasoning survives real HTTP tool rounds and restart without entering visible assistant text (10156.1791ms)
✔ DeepSeek SSE reasoning survives real HTTP tool rounds and restart without entering visible assistant text (7459.5587ms)
✔ malformed or post-terminal Chat reasoning is rejected with the raw stream retained (2.4721ms)
✔ connection import synchronizes profile, native fields and declared abilities without editing the source or manufacturing verification (1158.073ms)
✔ compiled HTTP persists profile and reasoning choices for every protocol and rejects incompatible/stale edits atomically (2704.5206ms)
ℹ tests 85
ℹ suites 0
ℹ pass 85
ℹ fail 0
ℹ cancelled 0
ℹ skipped 0
ℹ todo 0
ℹ duration_ms 75016.2225
@@ -0,0 +1,80 @@
✔ Messages builds native system/tool history, bounds images, and seals a distinct mixed-pool variant (7.1296ms)
✔ Messages URL and native parameter contracts preserve other protocol paths and hard output budget (0.9244ms)
✔ native usage includes both cache buckets once, retains raw and rejects invalid counters (0.1834ms)
✔ real HTTP Messages JSON executes one tool, retains opaque history, actual URL and cache accounting (5266.6751ms)
✔ real HTTP Messages SSE executes one tool, retains opaque history, actual URL and cache accounting (4833.9744ms)
✔ Messages api-key authentication is explicit; mismatched model remains blocked with original usage (5.942ms)
✔ Messages bearer authentication is explicit; mismatched model remains blocked with original usage (4.694ms)
✔ Messages redirect is not followed, either credential stays at the original server (4.1228ms)
✔ strict native SSE reads the nested start model and retains mismatch source without retries (4.0836ms)
✔ a closed native tool block cannot execute before message_stop, then executes exactly once (4584.6086ms)
✔ mixed Chat/Messages pool seals both variants and preserves one completed tool on failover (5555.6561ms)
✔ native JSON rejects incomplete stops, duplicate tools and hosted content (0.2714ms)
✔ Messages rejects empty, whitespace and thinking-only answers without rejecting valid text or tool calls (0.207ms)
✔ empty Messages JSON fails once with original response and 6692/0 usage, before verification or tools (2447.6918ms)
✔ empty Messages SSE fails once with original response and 6692/0 usage, before verification or tools (3044.5991ms)
▶ strict SSE validates all blocks, terminal, event names, UTF8 and tool JSON before dispatch
✔ reject native stream 0 (1.8356ms)
✔ reject native stream 1 (0.3292ms)
✔ reject native stream 2 (0.2386ms)
✔ reject native stream 3 (0.2114ms)
✔ reject native stream 4 (0.1813ms)
✔ reject native stream 5 (0.175ms)
✔ reject native stream 6 (0.2167ms)
✔ reject native stream 7 (0.5289ms)
✔ reject native stream 8 (0.2287ms)
✔ strict SSE validates all blocks, terminal, event names, UTF8 and tool JSON before dispatch (5.0206ms)
✔ upstream SSE error retains original type/request ID and initial usage, never retries (3.7454ms)
✔ stop cancels the owned stream and retains partial consumption without running tool or replaying (58.5325ms)
✔ saved Messages import preserves auth choice, rejects endpoint duplication, and keeps source file untouched (1451.991ms)
✔ model registry keeps original credentials/endpoint intact, rejects stale edits, and forgets only volatile keys on restart (1749.3674ms)
✔ config import previews exact revisions, keeps secrets volatile, syncs only when enabled and preserves manual edits on conflicts (1216.7556ms)
✔ real HTTP pool fails over 429 to another model and never repeats completed tools; attempts seal actual request bytes and Run identity (5705.4303ms)
✔ pool rotation and weights select members on new Runs; disabled and missing credentials are excluded without requests (7812.7464ms)
✔ 401 blocks the credential revision; updating the key restores availability without leaking error echoes (2735.1836ms)
✔ queued request cancellation and in-flight disable release only owned leases and do not dispatch another HTTP request (1264.1974ms)
✔ compiled model-center HTTP API persists public profiles, blocks forged origins, and local operations never invoke a model (3230.067ms)
✔ model registry disk failures restore routing, pool profiles and credential references without partial updates (1129.5317ms)
✔ model selection and saved configuration commit together and failures preserve the original mode and endpoint (947.3742ms)
✔ configuration imports roll back the entire batch and retain one usable preview after validation or commit failure (1491.7596ms)
✔ HTTP quota retains native code/type/param/request id, bounded original and allowlisted headers, never repeats that connection (2393.6026ms)
✔ Retry-After above 90 seconds remains exact and a short local budget cannot trigger an early retry (1.7751ms)
✔ Retry-After date, expired date, huge value and invalid header retain explicit semantics (1.09ms)
✔ unclassified 429 remains unspecified and every bounded actual retry retains its own original error (2682.4273ms)
✔ oversized HTTP error saves exactly the bounded prefix and labels its incomplete range (2572.5909ms)
✔ Chat SSE error preserves upstream fields, original failed event, partial tool and reported usage without executing or retrying (2934.3728ms)
✔ Responses failed SSE retains original event/response identity and usage (3690.287ms)
✔ Responses incomplete SSE retains original event/response identity and usage (2786.337ms)
✔ unsupported successful Responses output retains original JSON and usage with a local protocol origin (2917.6914ms)
✔ HTTP 200 native error is still an upstream failure rather than a fabricated successful answer (0.7849ms)
✔ JSON-escaped credential echoes are redacted in the original preview and capture as well as parsed fields (0.5591ms)
✔ quota failover uses another independent pool member and never replays a failed stream (3528.8719ms)
✔ native parameters are isolated by pool member, preserve Responses state and bind the original output reservation (0.3091ms)
✔ unsupported native fields and invalid values reject before dispatch instead of silently dropping or overriding constraints (0.3074ms)
✔ native settings for legacy and registered profiles persist without changing credentials or old records and reject stale edits (1577.2667ms)
✔ compiled native settings API saves/restores exact parameters and rejects reserved fields without mutating the profile (3067.493ms)
✔ legacy connection test sends its saved native fields and selected output limit field without changing the configuration (3050.9963ms)
✔ HTTP download preserves original malformed SSE bytes and artifact SHA instead of decoding invalid UTF8 (3005.7597ms)
✔ terminated SSE adapts empty optional tool arguments, retains the exact wire bytes and executes the validated batch once (5716.973ms)
✔ reported SSE usage survives rejected required arguments "" without dispatching any tool (3118.6902ms)
✔ reported SSE usage survives rejected required arguments "{\"path\":" without dispatching any tool (3727.617ms)
✔ reported SSE usage survives rejected required arguments "null" without dispatching any tool (2779.8954ms)
✔ a stream without DONE retains reported usage but cannot adapt or execute even an optional tool (2740.6565ms)
✔ a timeout after reasoning-only fragments creates one unknown receipt and never invents reported tokens (3063.2274ms)
✔ real Chat SSE carries UTF8 and split tool JSON; Runtime executes only after DONE, stores partial progress and original usage (4051.9701ms)
✔ Responses typed SSE validates original item identity, terminal data, raw usage and tool result pair (3829.8048ms)
✔ incomplete network stream retains unknown attempt and never automatically switches pool or dispatches fragment tools (3421.6605ms)
✔ stopping an active HTTP stream closes only original request and keeps partial text without executing tools (3039.5487ms)
✔ explicit streaming probe uses actual SSE despite unknown declaration and binds its own original attempt (1006.9865ms)
✔ coalesced initial SSE events publish the latest partial state even when the upstream stays idle (1696.3744ms)
✔ stream timeout closes the original HTTP response, preserves redacted capture and never retries (950.0963ms)
✔ stream idle timeout preserves incomplete tool arguments, records unknown use and never fails over (1043.9052ms)
✔ SSE accepts CR-only final event boundaries and retains exact invalid UTF8 bytes without dispatch (1.2534ms)
▶ stream parser rejects truncation, finish length, malformed JSON, duplicate sequence, changed final item and oversized data
✔ reject 0 (0.2815ms)
✔ reject 1 (0.2369ms)
✔ reject 2 (0.1547ms)
✔ reject 3 (5.5116ms)
✔ reject 4 (0.226ms)
✔ reject 5 (0.4891ms)
✔ stream parser rejects truncation, finish length, malformed JSON, duplicate sequence, changed final item and oversized data (7.5143ms)
@@ -0,0 +1,5 @@
import { Store } from '../../src/server/store.mjs';
import fs from 'node:fs';
const store = new Store(process.argv[2]);
try { console.log(JSON.stringify({ sessions: [...store.sessions.values()].map(s => ({ id: s.id, status: s.status, failure: s.failure, messages: s.messages.slice(-3), events: s.events.slice(-3) })), runs: [...store.core.run.values()].map(r => ({ status: r.status, stopReason: r.stopReason, usage: r.usage })) }, null, 2)); }
finally { store.close(); }
@@ -0,0 +1,8 @@
EVIDENCE D:\Ai-project\REA逆向\Agent-Workbench\.qa\reasoning-controls\electron-fbd6cfbe-ca37-4748-8148-a33220e12c1a
PASS custom-compatible profile and enabled/low controls save and reload without requests or losing unrelated parameters
PASS DeepSeek off/on/max/default reach actual HTTP with no stale fields and preserve the original output reservation
PASS JSON and controls edit one value; malformed JSON and protected fields fail visibly; changing custom dialect uses the matching wire format
PASS Responses strength and disable controls send reasoning.effort using their own protocol
PASS Messages adaptive/effort and manual budget persist; 390px controls, capability scope and save action remain readable and usable
PASS actual Messages empty SSE is diagnosed explicitly in the task at both widths with 6692/0 use, zero tools and one request
{"directory":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\electron-fbd6cfbe-ca37-4748-8148-a33220e12c1a","success":true,"version":"1.25.2","appRoot":"D:\\Ai-project\\REA逆向\\Agent-Workbench","actualElectron":true,"checks":["custom-compatible profile and enabled/low controls save and reload without requests or losing unrelated parameters","DeepSeek off/on/max/default reach actual HTTP with no stale fields and preserve the original output reservation","JSON and controls edit one value; malformed JSON and protected fields fail visibly; changing custom dialect uses the matching wire format","Responses strength and disable controls send reasoning.effort using their own protocol","Messages adaptive/effort and manual budget persist; 390px controls, capability scope and save action remain readable and usable","actual Messages empty SSE is diagnosed explicitly in the task at both widths with 6692/0 use, zero tools and one request"],"screenshots":[{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\electron-fbd6cfbe-ca37-4748-8148-a33220e12c1a\\deepseek-desktop.png","sha256":"342aa41e7e169079472361acfa2e13e0ceb18298001eac79d90a60b6dbf85c45"},{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\electron-fbd6cfbe-ca37-4748-8148-a33220e12c1a\\invalid-json.png","sha256":"3cae5d79753574ccc53c1dad92c4ee78507f7d54f59a7940e1dc6b2b6f0e1156"},{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\electron-fbd6cfbe-ca37-4748-8148-a33220e12c1a\\messages-390.png","sha256":"22ee39042c6540652e06f58cc6fc017c75f6c8fa70d2f69df4af232deaf9a6be"},{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\electron-fbd6cfbe-ca37-4748-8148-a33220e12c1a\\features-390.png","sha256":"d6d1b5a0a064827012f132eea54b4da0eb588eae8de9e3485d6f13438610bfc3"},{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\electron-fbd6cfbe-ca37-4748-8148-a33220e12c1a\\empty-response-desktop.png","sha256":"e1c92d4920d6564b5c9ec04ed7bb4456e8d307acad0afece216832da769436f5"},{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\electron-fbd6cfbe-ca37-4748-8148-a33220e12c1a\\empty-response-390.png","sha256":"e50d3de5c6e422b6908907c9a848f2bd8ab00b4aefc34e4920ca521a5a9cf0d8"}],"errors":[],"syntheticRequests":9,"realModelRequests":0,"officialServiceTouched":false}
@@ -0,0 +1,8 @@
EVIDENCE D:\Ai-project\REA逆向\Agent-Workbench\.qa\reasoning-controls\electron-4ceb3998-63f7-4ab5-99a8-a2c93e424f46
PASS custom-compatible profile and enabled/low controls save and reload without requests or losing unrelated parameters
PASS DeepSeek off/on/max/default reach actual HTTP with no stale fields and preserve the original output reservation
PASS JSON and controls edit one value; malformed JSON and protected fields fail visibly; changing custom dialect uses the matching wire format
PASS Responses strength and disable controls send reasoning.effort using their own protocol
PASS Messages adaptive/effort and manual budget persist; 390px controls, capability scope and save action remain readable and usable
PASS actual Messages empty SSE is diagnosed explicitly in the task at both widths with 6692/0 use, zero tools and one request
{"directory":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\electron-4ceb3998-63f7-4ab5-99a8-a2c93e424f46","success":true,"version":"1.25.2","appRoot":"D:\\Ai-project\\REA逆向\\Agent-Workbench","actualElectron":true,"checks":["custom-compatible profile and enabled/low controls save and reload without requests or losing unrelated parameters","DeepSeek off/on/max/default reach actual HTTP with no stale fields and preserve the original output reservation","JSON and controls edit one value; malformed JSON and protected fields fail visibly; changing custom dialect uses the matching wire format","Responses strength and disable controls send reasoning.effort using their own protocol","Messages adaptive/effort and manual budget persist; 390px controls, capability scope and save action remain readable and usable","actual Messages empty SSE is diagnosed explicitly in the task at both widths with 6692/0 use, zero tools and one request"],"screenshots":[{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\electron-4ceb3998-63f7-4ab5-99a8-a2c93e424f46\\deepseek-desktop.png","sha256":"19c71cfe68410a7422ee0ca3c2dfd64209a0384f06f3d10c4901b453b748b123"},{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\electron-4ceb3998-63f7-4ab5-99a8-a2c93e424f46\\invalid-json.png","sha256":"3cae5d79753574ccc53c1dad92c4ee78507f7d54f59a7940e1dc6b2b6f0e1156"},{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\electron-4ceb3998-63f7-4ab5-99a8-a2c93e424f46\\messages-390.png","sha256":"22ee39042c6540652e06f58cc6fc017c75f6c8fa70d2f69df4af232deaf9a6be"},{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\electron-4ceb3998-63f7-4ab5-99a8-a2c93e424f46\\features-390.png","sha256":"d6d1b5a0a064827012f132eea54b4da0eb588eae8de9e3485d6f13438610bfc3"},{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\electron-4ceb3998-63f7-4ab5-99a8-a2c93e424f46\\empty-response-desktop.png","sha256":"9b1a4977769b019e120e15344551c0e3afcfcd56f08bf63d77f101bdd7427a38"},{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\electron-4ceb3998-63f7-4ab5-99a8-a2c93e424f46\\empty-response-390.png","sha256":"8688c63735a49a60b03c2bc158dfff0b1f340f9ba47e49360206b6f5d692fee3"}],"errors":[],"syntheticRequests":9,"realModelRequests":0,"officialServiceTouched":false}
@@ -0,0 +1,7 @@
PASS native parameters persist after reload and the narrow form remains usable without calling a model
PASS unsupported field is rejected visibly before any dispatch
PASS actual HTTP error shows raw fields and original wait, keeps secrets hidden and downloads SHA-verified original at both widths
PASS physical request ledger preserves the same error origin and offers the sealed actual request
PASS Responses failed event retains original identity and reported use at both widths; partial tool is never dispatched or retried
PASS explicit synthetic probe also shows the API diagnostic and labels its preview-only retention without inventing a Task artifact
{"success":true,"directory":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\model-fidelity-ui-690a1968-48ee-4a52-a6d6-8bdc0ba06693","checks":6,"screenshots":10}
@@ -0,0 +1,9 @@
EVIDENCE D:\Ai-project\REA逆向\Agent-Workbench\.qa\messages-ui-m-490210e4-de43-48eb-89f9-492bdcf77b74
PASS root and /v1 preview agree, full endpoint rejects with input retained; save makes no model request
PASS saved protocol, auth and selected connection survive reload and remain operable at 390px
PASS native settings reject other-protocol fields and keep a single hard max_tokens output field
PASS actual local JSON text/tool and native SSE capability probes use the saved protocol and auth
PASS compiled native tool roundtrip completes with original call ID, cache-inclusive usage and observed request/response URLs
PASS strict mismatch is visible with original GPT identifier and one request, no retry or tool dispatch
PASS 429 preserves native type, raw message, request ID, URL and inputs without automatic replay
{"success":true,"checks":7,"directory":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\messages-ui-m-490210e4-de43-48eb-89f9-492bdcf77b74"}
@@ -0,0 +1,37 @@
✔ Messages builds native system/tool history, bounds images, and seals a distinct mixed-pool variant (7.0907ms)
✔ Messages URL and native parameter contracts preserve other protocol paths and hard output budget (0.5459ms)
✔ native usage includes both cache buckets once, retains raw and rejects invalid counters (0.1601ms)
✔ real HTTP Messages JSON executes one tool, retains opaque history, actual URL and cache accounting (1844.951ms)
✔ real HTTP Messages SSE executes one tool, retains opaque history, actual URL and cache accounting (1998.1126ms)
✔ Messages api-key authentication is explicit; mismatched model remains blocked with original usage (5.3043ms)
✔ Messages bearer authentication is explicit; mismatched model remains blocked with original usage (3.9637ms)
✔ Messages redirect is not followed, either credential stays at the original server (4.2348ms)
✔ strict native SSE reads the nested start model and retains mismatch source without retries (3.0718ms)
✔ a closed native tool block cannot execute before message_stop, then executes exactly once (2688.0387ms)
✔ mixed Chat/Messages pool seals both variants and preserves one completed tool on failover (2779.4277ms)
✔ native JSON rejects incomplete stops, duplicate tools and hosted content (0.2839ms)
✔ Messages rejects empty, whitespace and thinking-only answers without rejecting valid text or tool calls (0.2153ms)
✔ empty Messages JSON fails once with original response and 6692/0 usage, before verification or tools (1478.1953ms)
✔ empty Messages SSE fails once with original response and 6692/0 usage, before verification or tools (1477.923ms)
▶ strict SSE validates all blocks, terminal, event names, UTF8 and tool JSON before dispatch
✔ reject native stream 0 (1.0821ms)
✔ reject native stream 1 (0.2232ms)
✔ reject native stream 2 (0.1694ms)
✔ reject native stream 3 (0.162ms)
✔ reject native stream 4 (0.1476ms)
✔ reject native stream 5 (0.247ms)
✔ reject native stream 6 (0.2926ms)
✔ reject native stream 7 (0.2085ms)
✔ reject native stream 8 (0.3131ms)
✔ strict SSE validates all blocks, terminal, event names, UTF8 and tool JSON before dispatch (3.5047ms)
✔ upstream SSE error retains original type/request ID and initial usage, never retries (4.9154ms)
✔ stop cancels the owned stream and retains partial consumption without running tool or replaying (63.8547ms)
✔ saved Messages import preserves auth choice, rejects endpoint duplication, and keeps source file untouched (564.6921ms)
ℹ tests 28
ℹ suites 0
ℹ pass 28
ℹ fail 0
ℹ cancelled 0
ℹ skipped 0
ℹ todo 0
ℹ duration_ms 13253.1781
@@ -0,0 +1,9 @@
EVIDENCE D:\Ai-project\REA逆向\Agent-Workbench\.qa\messages-ui-m-0efd6320-6785-47b1-8dc2-0002118eb753
PASS root and /v1 preview agree, full endpoint rejects with input retained; save makes no model request
PASS saved protocol, auth and selected connection survive reload and remain operable at 390px
PASS native settings reject other-protocol fields and keep a single hard max_tokens output field
PASS actual local JSON text/tool and native SSE capability probes use the saved protocol and auth
PASS compiled native tool roundtrip completes with original call ID, cache-inclusive usage and observed request/response URLs
PASS strict mismatch is visible with original GPT identifier and one request, no retry or tool dispatch
PASS 429 preserves native type, raw message, request ID, URL and inputs without automatic replay
{"success":true,"checks":7,"directory":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\messages-ui-m-0efd6320-6785-47b1-8dc2-0002118eb753"}
@@ -0,0 +1,7 @@
PASS connection form stores only the volatile credential and renders configured status without exposing it
PASS pool editor saves weighted members and binds selection for new Runs
PASS schema-bound import preview, opt-in automatic sync, revision update and sync shutdown keep credentials hidden
PASS real historical pagination detects the owned corrupted older artifact, retains its identity and leaves original bytes unchanged
PASS local inspection shows real database and artifact checks and emits zero model requests
PASS 390px model-center modal and connection cards remain operable without horizontal page overflow
EVIDENCE D:\Ai-project\REA逆向\Agent-Workbench\.qa\model-center-ui-14b3a91a-e1ec-4dc8-86c7-f5bff306f7be
@@ -0,0 +1,6 @@
PASS actual form saves Responses and declared stream, refresh preserves both and narrow editor remains operable
PASS separate text, nonce tool and actual SSE probes bind only their requested capability and own attempts
PASS actual pooled task excludes text-only before dispatch, executes original tool once and downloads SHA-matched original SSE
PASS desktop and actual paired gateway show same bounded stream; mobile stops original HTTP with zero fragment tool dispatch or replay
PASS truncated output keeps explicit unknown ledger, partial evidence and no automatic failover or incomplete tool execution
EVIDENCE D:\Ai-project\REA逆向\Agent-Workbench\.qa\model-protocols-ui-530110ad-1e26-4d68-9d41-e8dd82c7adff
File diff suppressed because one or more lines are too long
@@ -0,0 +1,411 @@
import { APIResource } from '../core/resource';
import { APIPromise } from '../core/api-promise';
import { Page, type PageParams, PagePromise } from '../core/pagination';
import { buildHeaders } from '../internal/headers';
import { RequestOptions } from '../internal/request-options';
import { startApiCallSpan } from '../internal/tracing';
import { path } from '../internal/utils/path';
export class Models extends APIResource {
/**
* Get a specific model.
*
* The Models API response can be used to determine information about a specific
* model or resolve a model alias to a model ID.
*
* @example
* ```ts
* const modelInfo = await client.models.retrieve('model_id');
* ```
*/
retrieve(
modelID: string,
params: ModelRetrieveParams | null | undefined = {},
options?: RequestOptions,
): APIPromise<ModelInfo> {
const { workspace_id } = params ?? {};
return this._client.get(path`/v1/models/${modelID}`, {
...options,
headers: buildHeaders([
{ ...(workspace_id != null ? { 'anthropic-workspace-id': workspace_id } : undefined) },
options?.headers,
]),
__span: startApiCallSpan(this._client, 'anthropic.models.retrieve', options, {
route: '/v1/models/{model_id}',
attributes: { 'anthropic.model.id': modelID },
}),
});
}
/**
* List available models.
*
* The Models API response can be used to determine which models are available for
* use in the API. More recently released models are listed first.
*
* @example
* ```ts
* // Automatically fetches more pages as needed.
* for await (const modelInfo of client.models.list()) {
* // ...
* }
* ```
*/
list(
params: ModelListParams | null | undefined = {},
options?: RequestOptions,
): PagePromise<ModelInfosPage, ModelInfo> {
const { workspace_id, ...query } = params ?? {};
return this._client.getAPIList('/v1/models', Page<ModelInfo>, {
query,
...options,
headers: buildHeaders([
{ ...(workspace_id != null ? { 'anthropic-workspace-id': workspace_id } : undefined) },
options?.headers,
]),
__span: startApiCallSpan(this._client, 'anthropic.models.list', options, {
route: '/v1/models',
queryParams: ['after_id', 'before_id', 'lifecycle', 'limit'],
}),
});
}
}
export type ModelInfosPage = Page<ModelInfo>;
/**
* Indicates whether a capability is supported.
*/
export interface CapabilitySupport {
/**
* Whether this capability is supported by the model.
*/
supported: boolean;
}
/**
* Context management capability details.
*/
export interface ContextManagementCapability {
/**
* Whether the clear_thinking_20251015 strategy is supported.
*/
clear_thinking_20251015: CapabilitySupport | null;
/**
* Whether the clear_tool_uses_20250919 strategy is supported.
*/
clear_tool_uses_20250919: CapabilitySupport | null;
/**
* Whether the compact_20260112 strategy is supported.
*/
compact_20260112: CapabilitySupport | null;
/**
* Whether this capability is supported by the model.
*/
supported: boolean;
}
/**
* Effort (reasoning_effort) capability details.
*/
export interface EffortCapability {
/**
* Whether the model supports high effort level.
*/
high: CapabilitySupport;
/**
* Whether the model supports low effort level.
*/
low: CapabilitySupport;
/**
* Whether the model supports max effort level.
*/
max: CapabilitySupport;
/**
* Whether the model supports medium effort level.
*/
medium: CapabilitySupport;
/**
* Whether this capability is supported by the model.
*/
supported: boolean;
/**
* Whether the model supports xhigh effort level.
*/
xhigh: CapabilitySupport | null;
}
/**
* Model capability information.
*/
export interface ModelCapabilities {
/**
* Whether the model supports the Batch API.
*/
batch: CapabilitySupport;
/**
* Whether the model supports citation generation.
*/
citations: CapabilitySupport;
/**
* Whether code that the model runs in the code execution tool can call the
* request's other tools, as in programmatic tool calling and dynamic filtering for
* web search and web fetch. Support for the code execution tool itself is in
* `server_tools.code_execution`.
*/
code_execution: CapabilitySupport;
/**
* Context management support and available strategies.
*/
context_management: ContextManagementCapability;
/**
* Effort (reasoning_effort) support and available levels.
*/
effort: EffortCapability;
/**
* Whether the model accepts image content blocks.
*/
image_input: CapabilitySupport;
/**
* Whether the model accepts PDF content blocks.
*/
pdf_input: CapabilitySupport;
/**
* Whether this model supports the web search and code execution server tools.
* `supported` is true when the model supports at least one of the tools. A
* supported tool can still be rejected for your organization, for example when an
* admin has turned web search off.
*/
server_tools: ServerToolsCapability;
/**
* Whether the model supports structured output / JSON mode / strict tool schemas.
*/
structured_outputs: CapabilitySupport;
/**
* Thinking capability and supported type configurations.
*/
thinking: ThinkingCapability;
}
export interface ModelInfo {
/**
* Unique model identifier.
*/
id: string;
/**
* Object mapping capability names to their support details. Keys are always
* present for all known capabilities.
*/
capabilities: ModelCapabilities | null;
/**
* RFC 3339 datetime string representing the time at which the model was released.
* May be set to an epoch value if the release date is unknown.
*/
created_at: string;
/**
* RFC 3339 datetime string representing the time of the model's most recent
* deprecation. Populated for `deprecated` and `retired` models; `null` while the
* model is `active`.
*/
deprecated_at: string | null;
/**
* A human-readable name for the model.
*/
display_name: string;
/**
* The model's current lifecycle stage.
*
* - `active`: The model is available for use, open to new adopters, and not
* scheduled for retirement.
* - `deprecated`: The model remains callable for organizations with existing
* access, but is headed for retirement and closed to new adopters.
* - `retired`: The model is no longer available for use; inference requests naming
* it fail. It remains in the catalogue as the historical record of its
* retirement.
*/
lifecycle: 'active' | 'deprecated' | 'retired';
/**
* The model line this model belongs to, such as `opus` for both Claude Opus 4.5
* and Claude Opus 4.6. More lines may be added. `null` when the model belongs to
* no line; do not infer a line from the `id`.
*/
line: ModelLine | null;
/**
* Maximum input context window size in tokens for this model.
*/
max_input_tokens: number | null;
/**
* Maximum value for the `max_tokens` parameter when using this model.
*/
max_tokens: number | null;
/**
* RFC 3339 datetime string representing the model's currently scheduled retirement
* date. The schedule can be revised until retirement occurs; `null` while the
* model is `active` or while no retirement is scheduled. A past date on a
* `deprecated` model means retirement is overdue, not that it has occurred:
* `lifecycle` is the retirement signal.
*/
retires_at: string | null;
/**
* Object type.
*
* For Models, this is always `"model"`.
*/
type: 'model';
}
/**
* A Claude model line, such as `opus` or `sonnet`. More lines may be added as new
* values.
*/
export type ModelLine = 'haiku' | 'sonnet' | 'opus' | 'fable' | 'mythos';
/**
* Web search and code execution tool support, with one entry per tool.
*/
export interface ServerToolsCapability {
/**
* Whether the model supports the code execution tool: true when the model supports
* at least one version of the tool, not necessarily every version.
*/
code_execution: CapabilitySupport;
/**
* Whether this capability is supported by the model.
*/
supported: boolean;
/**
* Whether the model supports the web search tool: true when the model supports at
* least one version of the tool, not necessarily every version.
*/
web_search: CapabilitySupport;
}
/**
* Thinking capability details.
*/
export interface ThinkingCapability {
/**
* Whether this capability is supported by the model.
*/
supported: boolean;
/**
* Supported thinking type configurations.
*/
types: ThinkingTypes;
}
/**
* Which `thinking.type` values the model accepts on requests. Read each key on its
* own: for example, `enabled` can be false while `disabled` is true.
*/
export interface ThinkingTypes {
/**
* Whether the model accepts thinking with type 'adaptive' (the model decides
* whether and how much to think).
*/
adaptive: CapabilitySupport;
/**
* Whether the model accepts thinking with type 'between_tools' (thinking turned
* off, and any short progress updates the model writes between tool calls come
* back as thinking blocks). False exactly when a request that sends it gets a 400
* from this model. A request that sends 'between_tools' can still be rejected for
* another reason, such as an effort level the model does not allow with thinking
* off.
*/
between_tools: CapabilitySupport;
/**
* Whether the model accepts thinking with type 'disabled' (thinking turned off).
* False exactly when a request that sends it gets a 400 from this model. True on a
* model that does not support thinking. A request that sends 'disabled' can still
* be rejected for another reason, such as an effort level the model does not allow
* with thinking off.
*/
disabled: CapabilitySupport;
/**
* Whether the model accepts thinking with type 'enabled' (extended thinking with a
* caller-set `budget_tokens`).
*/
enabled: CapabilitySupport;
}
export interface ModelRetrieveParams {
/**
* Optional header to select the Workspace for this request. The value is a
* Workspace ID (for example, `wrkspc_011CZkZaBF1tNoB5wlCeusgy`).
*
* Only needed for credentials that can act on more than one Workspace. A
* credential that belongs to a specific Workspace may omit it; if sent, it must
* match that Workspace.
*/
workspace_id?: string;
}
export interface ModelListParams extends PageParams {
/**
* Query param: Filter the list to models in any of the given lifecycle stages
* (`active`, `deprecated`, or `retired`). Up to 3 values. When omitted, the list
* contains the `active` and `deprecated` models; `retired` models appear only when
* `retired` is requested explicitly.
*/
lifecycle?: Array<'active' | 'deprecated' | 'retired'>;
/**
* Header param: Optional header to select the Workspace for this request. The
* value is a Workspace ID (for example, `wrkspc_011CZkZaBF1tNoB5wlCeusgy`).
*
* Only needed for credentials that can act on more than one Workspace. A
* credential that belongs to a specific Workspace may omit it; if sent, it must
* match that Workspace.
*/
workspace_id?: string;
}
export declare namespace Models {
export {
type CapabilitySupport as CapabilitySupport,
type ContextManagementCapability as ContextManagementCapability,
type EffortCapability as EffortCapability,
type ModelCapabilities as ModelCapabilities,
type ModelInfo as ModelInfo,
type ModelLine as ModelLine,
type ServerToolsCapability as ServerToolsCapability,
type ThinkingCapability as ThinkingCapability,
type ThinkingTypes as ThinkingTypes,
type ModelInfosPage as ModelInfosPage,
type ModelRetrieveParams as ModelRetrieveParams,
type ModelListParams as ModelListParams,
};
}
@@ -0,0 +1,10 @@
[
{
"url": "https://developers.openai.com/api/docs/guides/reasoning#reasoning-effort",
"text": "## Reasoning effort\n\nThe `reasoning.effort` parameter guides the model on how much to think when performing a task.\n\nSupported values are model-dependent and can include `none`, `minimal`, `low`, `medium`, `high`, `xhigh`, and `max`. Lower effort favors speed and lower token usage, while at higher effort the model thinks more completely to provide higher quality responses. The models also reason adaptively across reasoning efforts, using fewer tokens for simpler tasks and thinking harder for complex tasks.\n\n[GPT-6 Astra](https://developers.openai.com/api/docs/models/gpt-6-astra) does not support `none` reasoning\n effort. Setting `reasoning.effort` (Responses) or `reasoning_effort` (Chat\n Completions) to `none` returns HTTP 400. [GPT-6.1 Sol](https://developers.openai.com/api/docs/guides/latest-model?model=gpt-6-astra#gpt-61-sol)\n does not support `none` or `minimal` and defaults to `medium`.\n\nUse the [Responses API](https://developers.openai.com/api/docs/guides/migrate-to-responses) for function\ncalling. Chat Completions does not support function calling with GPT-6 Astra or\nGPT-6.1 Sol.\n\nDefaults are also model-dependent rather than universal. `gpt-5.5` defaults to `medium` reasoning effort. This is the best starting point for `gpt-5.5`’s full balance of quality, reliability and performance.\n\n| Effort | Best for |\n| -------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |\n| `none` | Latency-critical tasks that do not benefit from any reasoning or multi-chained tool calls. For latency-sensitive use cases with `gpt-5.5`, we recommend trying `low` to begin with and then moving to `none` if required.<br /><br />Common use cases include voice, fast information retrieval, and classification. |\n| `low` | Efficient reasoning with a modest latency increase. Ideal for use cases requiring tool-use, planning, search, or multi-step decision making, while optimizing for speed and cost.<br /><br />Common use cases include data analysis, drafting, execution-oriented coding, and customer support / chat assistant workflows. |\n| `medium` | When quality and reliability matter, and the task involves planning, complex reasoning, and judgement. Default configuration for most workloads, and a well-balanced point on the pareto curve of latency, performance and cost.<br /><br />Common use cases include agentic coding, research, working with spreadsheets & slides, and delegating long-horizon work. |\n| `high` | Hard reasoning, complex debugging, deep planning, and high-value tasks where quality and intelligence matters more than latency. Recommended for complex workflows and agentic tasks.<br /><br />Common use cases include agentic coding, long-horizon research, and knowledge work. Depending on the complexity of the task, evaluate both `medium` and `high`. |\n| `xhigh` | Deep research, asynchronous workflows and agentic tasks that require long runs. Only use when your evals show a clear benefit that justifies the extra latency and cost.<br /><br />Common use cases include security and code review, enterprise productivity, deeper research tasks, and challenging coding workflows. |\n| `max` | Maximum reasoning for your most complex tasks. If you are currently using `xhigh`, evaluate if `max` results in stronger performance |\n\nFor faster time to first visible token in latency-sensitive applications, ask the model to generate a short preamble before continuing with deeper reasoning.\n\nSome models support only a subset of these values, so check the relevant [model page](https://developers.openai.com/api/docs/models) before choosing a setting."
},
{
"url": "https://developers.openai.com/api/reference/resources/chat/subresources/completions/methods/create",
"text": "- `reasoning_effort: optional ReasoningEffort or null (default: \"medium\")`\n\n Constrains effort on reasoning for reasoning models. Currently supported\n values are `none`, `minimal`, `low`, `medium`, `high`, `xhigh`, and `max`.\n Reducing reasoning effort can result in faster responses and fewer tokens\n used on reasoning in a response. Not all reasoning models support every\n value. See the\n [reasoning guide](/api/docs/guides/reasoning)\n for model-specific support.\n\n - `\"none\"`\n\n - `\"minimal\"`\n\n - `\"low\"`\n\n - `\"medium\"`\n\n - `\"high\"`\n\n - `\"xhigh\"`\n\n - `\"max\"`\n\n- `response_format: optional ResponseFormatText or ResponseFormatJSONSchema or ResponseFormatJSONObject`\n\n An object specifying the format that the model must output.\n\n Setting to `{ \"type\": \"json_schema\", \"json_schema\": {...} }` enables\n Structured Outputs which ensures the model will match your sup"
}
]
@@ -0,0 +1,8 @@
EVIDENCE D:\Ai-project\REA逆向\Agent-Workbench\.qa\reasoning-controls\electron-0562fc33-ceb5-4cb3-81e8-de1801dbd212
PASS custom-compatible profile and enabled/low controls save and reload without requests or losing unrelated parameters
PASS DeepSeek off/on/max/default reach actual HTTP with no stale fields and preserve the original output reservation
PASS JSON and controls edit one value; malformed JSON and protected fields fail visibly; changing custom dialect uses the matching wire format
PASS Responses strength and disable controls send reasoning.effort using their own protocol
PASS Messages adaptive/effort and manual budget persist; 390px controls, capability scope and save action remain readable and usable
PASS actual Messages empty SSE is diagnosed explicitly in the task at both widths with 6692/0 use, zero tools and one request
{"directory":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\electron-0562fc33-ceb5-4cb3-81e8-de1801dbd212","success":true,"version":"1.25.3","appRoot":"D:\\Ai-project\\REA逆向\\Agent-Workbench","actualElectron":true,"checks":["custom-compatible profile and enabled/low controls save and reload without requests or losing unrelated parameters","DeepSeek off/on/max/default reach actual HTTP with no stale fields and preserve the original output reservation","JSON and controls edit one value; malformed JSON and protected fields fail visibly; changing custom dialect uses the matching wire format","Responses strength and disable controls send reasoning.effort using their own protocol","Messages adaptive/effort and manual budget persist; 390px controls, capability scope and save action remain readable and usable","actual Messages empty SSE is diagnosed explicitly in the task at both widths with 6692/0 use, zero tools and one request"],"screenshots":[{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\electron-0562fc33-ceb5-4cb3-81e8-de1801dbd212\\deepseek-desktop.png","sha256":"ec87f8fba1f5b37031ed6aa7987db0a8a0cf8a365ac183fac0c10d0c3f0a3a8a"},{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\electron-0562fc33-ceb5-4cb3-81e8-de1801dbd212\\invalid-json.png","sha256":"c04c4811a9ce75ab260be0cbe924ae75cd3d09b72b93f0e702c192096e9ffc55"},{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\electron-0562fc33-ceb5-4cb3-81e8-de1801dbd212\\messages-390.png","sha256":"f2c75bb2cee87f65994760e7744376cc94884521672526724d8ef251e1852c72"},{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\electron-0562fc33-ceb5-4cb3-81e8-de1801dbd212\\features-390.png","sha256":"b99abe5f7fded70147e32836bd417194f42851a0c2d2152630e0aad9acdbe3ff"},{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\electron-0562fc33-ceb5-4cb3-81e8-de1801dbd212\\empty-response-desktop.png","sha256":"75476982de86e11dfc6bcb96fa510f2c41919fabe753233859fa5c049c37a207"},{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\electron-0562fc33-ceb5-4cb3-81e8-de1801dbd212\\empty-response-390.png","sha256":"8af384eaa4e2e0b43b39036ae627776a8e350ec7319bb388efee8029185dc4d4"}],"errors":[],"syntheticRequests":9,"realModelRequests":0,"officialServiceTouched":false}
@@ -0,0 +1,10 @@
✔ brand and custom-compatible profiles send native reasoning controls per protocol without changing output budgets (1.8129ms)
✔ opaque Chat reasoning returns only to its original endpoint, model, connection and credential revision; bytes count against the actual request (0.6896ms)
ℹ tests 2
ℹ suites 0
ℹ pass 2
ℹ fail 0
ℹ cancelled 0
ℹ skipped 0
ℹ todo 0
ℹ duration_ms 202.7673
@@ -0,0 +1 @@
✖ DeepSeek JSON reasoning survives real HTTP tool rounds and restart without entering visible assistant text (33916.588ms)
@@ -0,0 +1,15 @@
✔ brand and custom-compatible profiles send native reasoning controls per protocol without changing output budgets (1.8214ms)
✔ opaque Chat reasoning returns only to its original endpoint, model, connection and credential revision; bytes count against the actual request (0.6771ms)
✔ DeepSeek JSON reasoning survives real HTTP tool rounds and restart without entering visible assistant text (3514.3792ms)
✔ DeepSeek SSE reasoning survives real HTTP tool rounds and restart without entering visible assistant text (3247.9649ms)
✔ malformed or post-terminal Chat reasoning is rejected with the raw stream retained (1.3611ms)
✔ connection import synchronizes profile, native fields and declared abilities without editing the source or manufacturing verification (727.6219ms)
✔ compiled HTTP persists profile and reasoning choices for every protocol and rejects incompatible/stale edits atomically (2182.4779ms)
ℹ tests 7
ℹ suites 0
ℹ pass 7
ℹ fail 0
ℹ cancelled 0
ℹ skipped 0
ℹ todo 0
ℹ duration_ms 9868.9092
@@ -0,0 +1,16 @@
✔ brand and custom-compatible profiles send native reasoning controls per protocol without changing output budgets (1.8449ms)
✔ opaque Chat reasoning returns only to its original endpoint, model, connection and credential revision; bytes count against the actual request (0.6793ms)
✔ saved profile changes preserve credential references and old frozen Run settings, including the legacy connection after restart (1135.2747ms)
✔ DeepSeek JSON reasoning survives real HTTP tool rounds and restart without entering visible assistant text (3448.481ms)
✔ DeepSeek SSE reasoning survives real HTTP tool rounds and restart without entering visible assistant text (3530.8684ms)
✔ malformed or post-terminal Chat reasoning is rejected with the raw stream retained (1.351ms)
✔ connection import synchronizes profile, native fields and declared abilities without editing the source or manufacturing verification (652.8332ms)
✔ compiled HTTP persists profile and reasoning choices for every protocol and rejects incompatible/stale edits atomically (2199.807ms)
ℹ tests 8
ℹ suites 0
ℹ pass 8
ℹ fail 0
ℹ cancelled 0
ℹ skipped 0
ℹ todo 0
ℹ duration_ms 11168.0582
@@ -0,0 +1,8 @@
EVIDENCE D:\Ai-project\REA逆向\Agent-Workbench\.qa\reasoning-controls\web-2356bb69-973c-4067-aa24-c60f6156401e
PASS custom-compatible profile and enabled/low controls save and reload without requests or losing unrelated parameters
PASS DeepSeek off/on/max/default reach actual HTTP with no stale fields and preserve the original output reservation
PASS JSON and controls edit one value; malformed JSON and protected fields fail visibly; changing custom dialect uses the matching wire format
PASS Responses strength and disable controls send reasoning.effort using their own protocol
PASS Messages adaptive/effort and manual budget persist; 390px controls, capability scope and save action remain readable and usable
PASS actual Messages empty SSE is diagnosed explicitly in the task at both widths with 6692/0 use, zero tools and one request
{"directory":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\web-2356bb69-973c-4067-aa24-c60f6156401e","success":true,"version":"1.25.3","appRoot":"D:\\Ai-project\\REA逆向\\Agent-Workbench","actualElectron":false,"checks":["custom-compatible profile and enabled/low controls save and reload without requests or losing unrelated parameters","DeepSeek off/on/max/default reach actual HTTP with no stale fields and preserve the original output reservation","JSON and controls edit one value; malformed JSON and protected fields fail visibly; changing custom dialect uses the matching wire format","Responses strength and disable controls send reasoning.effort using their own protocol","Messages adaptive/effort and manual budget persist; 390px controls, capability scope and save action remain readable and usable","actual Messages empty SSE is diagnosed explicitly in the task at both widths with 6692/0 use, zero tools and one request"],"screenshots":[{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\web-2356bb69-973c-4067-aa24-c60f6156401e\\deepseek-desktop.png","sha256":"990ad31bf7fafedc0661031be3218806982e38e5f1729666a73e6c6f8b429521"},{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\web-2356bb69-973c-4067-aa24-c60f6156401e\\invalid-json.png","sha256":"9d1a1ae9f77c4c1c734d0e32fd42ea2586b808d3606a9b05ada6368feba91b9a"},{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\web-2356bb69-973c-4067-aa24-c60f6156401e\\messages-390.png","sha256":"7a47c8bb11538aeaec4f581c07f9bfd9cfff8a7af95b7da17e16dc7efb5188c2"},{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\web-2356bb69-973c-4067-aa24-c60f6156401e\\features-390.png","sha256":"69a49b72adcfb89aa2a8cc0a81123b2b00296c0fbe0ea03586d5f66426e5f934"},{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\web-2356bb69-973c-4067-aa24-c60f6156401e\\empty-response-desktop.png","sha256":"1880fb7d8bb9b3081ed1b86af5561b39a19c42c620fc2520e3c2c5d123a7189e"},{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\web-2356bb69-973c-4067-aa24-c60f6156401e\\empty-response-390.png","sha256":"41916c137133405c45ef543528e979003bc9a8c9d25065e011500d9f9ec36f60"}],"errors":[],"syntheticRequests":9,"realModelRequests":0,"officialServiceTouched":false}
@@ -0,0 +1,22 @@
✔ runtime guides map archived evidence while all original reports and successful/failed source bytes remain independently verifiable (789.4658ms)
✔ missing required research reports fail packaging rather than creating a successful placeholder (533.0526ms)
✔ full verification reports only verified bytes, complete declared scope and actual directory inventory; changed/deleted files still fail (632.6761ms)
✔ preflight and activation continue one monotonic operation bound to installer hash; terminal replay and changed origin are rejected (7683.8601ms)
✔ preflight failures and fixed-candidate mismatches retain the actual reason without publishing a pointer (4494.1616ms)
✔ cancel finishes only the original ready preparation, preserves data and pointer and refuses replay (6386.7809ms)
✔ the installation lock rejects concurrent same-id operations before they can overwrite original progress (3814.6476ms)
✔ original post-publication failure is linked to its rollback receipt and every original byte survives (3600.2929ms)
✔ a candidate mutation at the publication boundary still fails even after an earlier successful full validation (6423.2389ms)
✔ public progress marks a lost process unknown without editing its original record or claiming completion (4924.3992ms)
✔ throwing console/UI progress observers cannot decide the installation result (3391.6346ms)
✔ a progress-only write failure after committed publication returns the authoritative receipt and a warning (3537.1206ms)
✔ browser and MCP approvals use actual tool categories; mobile exposes only category while keeping original fingerprint (513.6766ms)
✔ the compiled runtime version constant tracks package.json so a release bump cannot ship a stale served version (8.4133ms)
ℹ tests 14
ℹ suites 0
ℹ pass 14
ℹ fail 0
ℹ cancelled 0
ℹ skipped 0
ℹ todo 0
ℹ duration_ms 46844.6927
@@ -0,0 +1,8 @@
EVIDENCE D:\Ai-project\REA逆向\Agent-Workbench\.qa\reasoning-controls\web-60c52f6a-d9a7-4c3c-9306-80fa620c75d3
PASS custom-compatible profile and enabled/low controls save and reload without requests or losing unrelated parameters
PASS DeepSeek off/on/max/default reach actual HTTP with no stale fields and preserve the original output reservation
PASS JSON and controls edit one value; malformed JSON and protected fields fail visibly; changing custom dialect uses the matching wire format
PASS Responses strength and disable controls send reasoning.effort using their own protocol
PASS Messages adaptive/effort and manual budget persist; 390px controls, capability scope and save action remain readable and usable
PASS actual Messages empty SSE is diagnosed explicitly in the task at both widths with 6692/0 use, zero tools and one request
{"directory":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\web-60c52f6a-d9a7-4c3c-9306-80fa620c75d3","success":true,"version":"1.25.2","appRoot":"D:\\Ai-project\\REA逆向\\Agent-Workbench","actualElectron":false,"checks":["custom-compatible profile and enabled/low controls save and reload without requests or losing unrelated parameters","DeepSeek off/on/max/default reach actual HTTP with no stale fields and preserve the original output reservation","JSON and controls edit one value; malformed JSON and protected fields fail visibly; changing custom dialect uses the matching wire format","Responses strength and disable controls send reasoning.effort using their own protocol","Messages adaptive/effort and manual budget persist; 390px controls, capability scope and save action remain readable and usable","actual Messages empty SSE is diagnosed explicitly in the task at both widths with 6692/0 use, zero tools and one request"],"screenshots":[{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\web-60c52f6a-d9a7-4c3c-9306-80fa620c75d3\\deepseek-desktop.png","sha256":"fc1accaabd8bb2d9cd38b1240dae5a98eb9558c774fbc53079d424bbd4981ffa"},{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\web-60c52f6a-d9a7-4c3c-9306-80fa620c75d3\\invalid-json.png","sha256":"075e49a587f1639db820b5a7422bfe192b6377551eb91557acf394c4cacd2adc"},{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\web-60c52f6a-d9a7-4c3c-9306-80fa620c75d3\\messages-390.png","sha256":"7a47c8bb11538aeaec4f581c07f9bfd9cfff8a7af95b7da17e16dc7efb5188c2"},{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\web-60c52f6a-d9a7-4c3c-9306-80fa620c75d3\\features-390.png","sha256":"16cf1a51186b15f799e98a06a0b1d751e21c4d4d6dbf154eade1345e2a75b056"},{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\web-60c52f6a-d9a7-4c3c-9306-80fa620c75d3\\empty-response-desktop.png","sha256":"116a50b05fbc177abfea913c7360924ada622092010af2b049cb4d2a114d1d97"},{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\web-60c52f6a-d9a7-4c3c-9306-80fa620c75d3\\empty-response-390.png","sha256":"321a600928ba68bd510248d50963986162a9f80ec882daecc29406a277958221"}],"errors":[],"syntheticRequests":9,"realModelRequests":0,"officialServiceTouched":false}
@@ -0,0 +1,8 @@
EVIDENCE D:\Ai-project\REA逆向\Agent-Workbench\.qa\reasoning-controls\web-93c26fdd-8a9a-479d-bef3-e2e7b59fcc8c
PASS custom-compatible profile and enabled/low controls save and reload without requests or losing unrelated parameters
PASS DeepSeek off/on/max/default reach actual HTTP with no stale fields and preserve the original output reservation
PASS JSON and controls edit one value; malformed JSON and protected fields fail visibly; changing custom dialect uses the matching wire format
PASS Responses strength and disable controls send reasoning.effort using their own protocol
PASS Messages adaptive/effort and manual budget persist; 390px controls, capability scope and save action remain readable and usable
PASS actual Messages empty SSE is diagnosed explicitly in the task at both widths with 6692/0 use, zero tools and one request
{"directory":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\web-93c26fdd-8a9a-479d-bef3-e2e7b59fcc8c","success":true,"version":"1.25.2","appRoot":"D:\\Ai-project\\REA逆向\\Agent-Workbench","actualElectron":false,"checks":["custom-compatible profile and enabled/low controls save and reload without requests or losing unrelated parameters","DeepSeek off/on/max/default reach actual HTTP with no stale fields and preserve the original output reservation","JSON and controls edit one value; malformed JSON and protected fields fail visibly; changing custom dialect uses the matching wire format","Responses strength and disable controls send reasoning.effort using their own protocol","Messages adaptive/effort and manual budget persist; 390px controls, capability scope and save action remain readable and usable","actual Messages empty SSE is diagnosed explicitly in the task at both widths with 6692/0 use, zero tools and one request"],"screenshots":[{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\web-93c26fdd-8a9a-479d-bef3-e2e7b59fcc8c\\deepseek-desktop.png","sha256":"e16a2377ef27c211150fe057fb24304a90e6dc0c7634a400a46747d5e28e3887"},{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\web-93c26fdd-8a9a-479d-bef3-e2e7b59fcc8c\\invalid-json.png","sha256":"075e49a587f1639db820b5a7422bfe192b6377551eb91557acf394c4cacd2adc"},{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\web-93c26fdd-8a9a-479d-bef3-e2e7b59fcc8c\\messages-390.png","sha256":"7a47c8bb11538aeaec4f581c07f9bfd9cfff8a7af95b7da17e16dc7efb5188c2"},{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\web-93c26fdd-8a9a-479d-bef3-e2e7b59fcc8c\\features-390.png","sha256":"16cf1a51186b15f799e98a06a0b1d751e21c4d4d6dbf154eade1345e2a75b056"},{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\web-93c26fdd-8a9a-479d-bef3-e2e7b59fcc8c\\empty-response-desktop.png","sha256":"6cad92fcc589471363d880b1c6f118076f115e1c5b4f3ae32201125c0eb4a5da"},{"path":"D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\reasoning-controls\\web-93c26fdd-8a9a-479d-bef3-e2e7b59fcc8c\\empty-response-390.png","sha256":"cfa3ba5e22fd23fb1b1192c9cd214d31b42ababff0f9f660df665f94ae3e88e6"}],"errors":[],"syntheticRequests":9,"realModelRequests":0,"officialServiceTouched":false}
@@ -0,0 +1,20 @@
import fs from 'node:fs';
import path from 'node:path';
import assert from 'node:assert/strict';
import { spawn } from 'node:child_process';
const root = '.qa/reasoning-1.25.3';
const read = file => JSON.parse(fs.readFileSync(root + '/' + file));
assert.equal(read('formal/upgrade.json').success, true);
const before = read('formal/before.json'), output = path.join(root, 'formal-native');
assert.equal(fs.existsSync(output), false); fs.mkdirSync(output);
const boot = await (await fetch('http://127.0.0.1:4318/api/bootstrap')).json();
const host = await (await fetch('http://127.0.0.1:4318/api/host/status', { headers: { 'X-Workbench-Token': boot.token } })).json();
assert.equal(boot.version, '1.25.3'); assert.equal(host.identity.pid, 36552);
assert.equal(host.activity.active, false); assert.equal(host.desktopWindows.count, 0);
const executable = path.join(before.installRoot, 'AgentWorkbench.exe'), startedAt = new Date().toISOString();
const child = spawn(executable, [], { windowsHide: false, stdio: 'ignore', cwd: before.installRoot });
const pid = child.pid;
const code = await new Promise((resolve, reject) => { child.once('error', reject); child.once('close', resolve); });
const result = { executable, pid, code, startedAt, finishedAt: new Date().toISOString(), scope: 'Launch the attested installed native UI for direct Windows review; retain the existing host and Provider.' };
fs.writeFileSync(path.join(output, 'launch.json'), JSON.stringify(result, null, 2) + '\n', { flag: 'wx' });
assert.equal(code, 0); console.log(JSON.stringify(result));
@@ -0,0 +1,78 @@
import fs from 'node:fs';
import path from 'node:path';
import assert from 'node:assert/strict';
import { chromium } from 'playwright';
const root = path.resolve('.qa/reasoning-1.25.3/formal-web'), origin = 'http://127.0.0.1:4318';
assert.equal(fs.existsSync(root), false); fs.mkdirSync(root);
const bootstrap = await (await fetch(origin + '/api/bootstrap')).json();
assert.equal(bootstrap.version, '1.25.3');
const projects = bootstrap.projects.filter(row => row.root.replaceAll('\\', '/').endsWith('/releases/Agent-Workbench-0.1.0-win-x64/examples/starter-project'));
assert.equal(projects.length, 1); const project = projects[0];
const sessions = bootstrap.sessions.filter(row => row.projectId === project.id && row.title.startsWith('创建一个HTML'));
assert.equal(sessions.length, 1); const session = sessions[0]; assert.equal(session.status, 'completed');
const browser = await chromium.launch({ channel: 'msedge', headless: true }), context = await browser.newContext({ viewport: { width: 1440, height: 1080 } });
const rejectedWrites = [], errors = [], checks = [], views = []; let result;
await context.route('**/*', async route => {
const request = route.request();
if (request.url().startsWith(origin + '/api/') && !['GET', 'HEAD'].includes(request.method())) {
rejectedWrites.push({ method: request.method(), url: request.url() }); await route.abort();
} else await route.continue();
});
const page = await context.newPage(); page.on('pageerror', error => errors.push(error.message));
async function capture(label) {
assert.equal(await page.evaluate(() => document.documentElement.scrollWidth <= innerWidth + 1), true);
await page.screenshot({ path: path.join(root, label + '.png'), scale: 'css' });
fs.writeFileSync(path.join(root, label + '.txt'), await page.locator('body').innerText());
}
try {
await page.goto(origin + '/?release=1.25.3'); await page.locator('.connection.connected').waitFor();
await page.getByLabel('当前项目', { exact: true }).selectOption(project.id);
await page.locator('.session-list button').filter({ hasText: session.title }).click();
await page.getByRole('button', { name: '查看任务记录', exact: true }).click();
const panel = page.getByLabel('要求与证据', { exact: true }); await panel.waitFor();
await panel.getByText('尚未逐项记录要求。模型的完成答复保留在原会话中。', { exact: true }).waitFor();
checks.push('Original completed Task retains its original unconfigured requirement view');
for (const [label, width, height] of [['desktop', 1440, 1080], ['narrow', 390, 844]]) {
await page.setViewportSize({ width, height });
if (!await page.getByLabel('选择运行记录', { exact: true }).isVisible()) await page.getByRole('button', { name: '查看任务记录', exact: true }).click();
await page.getByLabel('当前缺口与下一步', { exact: true }).getByText('当前文件已交付,按约定未测试', { exact: true }).waitFor();
await panel.scrollIntoViewIfNeeded();
assert.equal(await page.getByLabel('单轮观察', { exact: true }).count(), 0);
await capture('task-' + label);
views.push({ label, width, height, requirementItems: await panel.locator('.requirement-list li').count(), legacyShadowPanels: 0 });
}
await page.setViewportSize({ width: 1440, height: 1080 });
await page.getByRole('button', { name: '打开模型中心', exact: true }).click();
const dialog = page.getByRole('dialog', { name: '模型中心', exact: true }); await dialog.waitFor();
await dialog.getByRole('button', { name: '从上游添加模型', exact: true }).waitFor();
await dialog.getByRole('button', { name: '手动填写(兼容)', exact: true }).waitFor();
const reasoning = dialog.getByRole('button', { name: '推理与能力', exact: true });
assert.ok(await reasoning.count() > 0); await reasoning.first().click();
const form = page.getByRole('form', { name: '原生请求参数', exact: true }); await form.waitFor();
await form.getByLabel('参数配置档', { exact: true }).waitFor();
await form.getByLabel('推理模式', { exact: true }).waitFor();
for (const [label, width, height] of [['desktop', 1440, 1080], ['narrow', 390, 844]]) {
await page.setViewportSize({ width, height }); await form.getByLabel('推理模式', { exact: true }).scrollIntoViewIfNeeded();
await capture('reasoning-' + label);
}
await form.getByRole('button', { name: '取消', exact: true }).click();
checks.push('Retained connection opens profile and reasoning controls without saving or probing');
await dialog.getByRole('button', { name: '从上游添加模型', exact: true }).click();
const catalog = page.getByLabel('上游模型目录', { exact: true }); await catalog.waitFor();
assert.equal(await catalog.getByLabel('目录上游连接', { exact: true }).inputValue(), '');
for (const [label, width, height] of [['desktop', 1440, 1080], ['narrow', 390, 844]]) {
await page.setViewportSize({ width, height }); await catalog.getByRole('heading', { name: '从上游选择模型', exact: true }).scrollIntoViewIfNeeded();
await capture('catalog-' + label);
}
await catalog.getByRole('button', { name: '关闭目录', exact: true }).click();
checks.push('Upstream catalog primary entry opens at both viewports without fetching real upstreams');
assert.deepEqual(errors, []); assert.deepEqual(rejectedWrites, []);
const health = await (await fetch(origin + '/api/health')).json(); assert.equal(health.pid, 36552); assert.equal(health.version, '1.25.3');
result = { success: true, version: health.version, pid: health.pid, projectId: project.id, sessionId: session.id, checks, views, errors, rejectedWrites,
realModelRequests: 0, realCatalogRequests: 0, scope: 'Read-only retained formal Task, reasoning and model catalog UI; no save, probe or real upstream request. New discovery, routing and persistence verified separately on isolated source, portable and installed fixtures.' };
} catch (error) {
await page.screenshot({ path: path.join(root, 'failure.png') }).catch(() => {});
fs.writeFileSync(path.join(root, 'failure.txt'), await page.locator('body').innerText().catch(() => 'Unavailable'));
result = { success: false, error: error.stack, checks, views, errors, rejectedWrites, realModelRequests: 0, realCatalogRequests: 0 }; process.exitCode = 1;
} finally { await browser.close(); }
fs.writeFileSync(path.join(root, 'result.json'), JSON.stringify(result, null, 2) + '\n', { flag: 'wx' }); console.log(JSON.stringify(result));
@@ -0,0 +1,955 @@
> local-agent-workbench@1.25.3 test
> node --test --test-concurrency=2 tests/*.test.mjs
✔ Messages builds native system/tool history, bounds images, and seals a distinct mixed-pool variant (7.6339ms)
✔ Messages URL and native parameter contracts preserve other protocol paths and hard output budget (0.5753ms)
✔ native usage includes both cache buckets once, retains raw and rejects invalid counters (0.167ms)
✔ real HTTP Messages JSON executes one tool, retains opaque history, actual URL and cache accounting (3317.2402ms)
✔ real HTTP Messages SSE executes one tool, retains opaque history, actual URL and cache accounting (3115.2975ms)
✔ Messages api-key authentication is explicit; mismatched model remains blocked with original usage (5.5492ms)
✔ Messages bearer authentication is explicit; mismatched model remains blocked with original usage (4.7039ms)
✔ Messages redirect is not followed, either credential stays at the original server (3.7695ms)
✔ strict native SSE reads the nested start model and retains mismatch source without retries (2.8326ms)
✔ a closed native tool block cannot execute before message_stop, then executes exactly once (3005.0365ms)
✔ mixed Chat/Messages pool seals both variants and preserves one completed tool on failover (3875.4202ms)
✔ native JSON rejects incomplete stops, duplicate tools and hosted content (0.4259ms)
✔ Messages rejects empty, whitespace and thinking-only answers without rejecting valid text or tool calls (0.3587ms)
✔ empty Messages JSON fails once with original response and 6692/0 usage, before verification or tools (2584.1119ms)
✔ empty Messages SSE fails once with original response and 6692/0 usage, before verification or tools (2290.3735ms)
▶ strict SSE validates all blocks, terminal, event names, UTF8 and tool JSON before dispatch
✔ reject native stream 0 (1.2283ms)
✔ reject native stream 1 (0.2524ms)
✔ reject native stream 2 (0.1788ms)
✔ reject native stream 3 (0.1635ms)
✔ reject native stream 4 (0.1472ms)
✔ reject native stream 5 (0.1548ms)
✔ reject native stream 6 (0.1183ms)
✔ reject native stream 7 (0.1771ms)
✔ reject native stream 8 (0.2856ms)
✔ strict SSE validates all blocks, terminal, event names, UTF8 and tool JSON before dispatch (3.3834ms)
✔ upstream SSE error retains original type/request ID and initial usage, never retries (3.8196ms)
✔ stop cancels the owned stream and retains partial consumption without running tool or replaying (57.1007ms)
✔ saved Messages import preserves auth choice, rejects endpoint duplication, and keeps source file untouched (563.3622ms)
✔ large source receipts use cursor-preserving invalidation without filling the event stream or changing the session (5.44ms)
✔ small event updates retain the current session and actual cursor (0.1152ms)
✔ the bundled local service enforces host, origin, token and JSON object contracts (2442.53ms)
✔ an in-flight connection probe prevents configuration changes and another probe until its actual result returns (2629.7621ms)
✔ API review applies verified content, detects conflicts, exports sessions and keeps keys in memory (4994.2485ms)
✔ a real bundle restart preserves proposals, interrupts old approvals and requests fresh consent on resume (11692.6575ms)
✔ runtime diagnostics and model-response validation use actual records and preserve the saved connection (2498.8476ms)
✔ persistent event replay is read-only and keeps run/artifact identity across reconnects (8285.269ms)
✔ real browser HTTP plans, original field pages and PNG bytes preserve ownership and survive host restart (7760.7758ms)
✔ self harness browser tools require approval and feed verified actual page fields back to the original Task (13564.7788ms)
✔ browser host recovery retains unknown root outcome, blocks replay, and requires an explicit owned acknowledgement (3475.9019ms)
✔ late HTTP stop of a finished browser job cannot interrupt a later job in the same Task (7870.9049ms)
✔ actual failed page blocks source-passed delivery; a new passing page permits a bound commit and preserves both original records (20900.8753ms)
✔ a build-only change after Git review invalidates the actual browser evidence and leaves HEAD and the protected index intact (10660.3985ms)
✔ actual DOM controls include validated unique CSS selectors across repeated parent buttons and escaped labels (7638.9153ms)
✔ browser field coverage distinguishes sections and Run windows while retaining original source identity (0.4554ms)
✔ repeated original browser pages pause a live Task without re-opening the browser or breaking tool pairing (6121.9121ms)
✔ browser plan rejects outside builds, invalid interaction scope and unbounded steps before execution (1.3841ms)
✔ real owned Edge interactions, assertions, raw PNG and stopped process survive exact result reuse and host reopen (4176.7285ms)
✔ failed real page assertion preserves actual value and forbids silent retry of the original plan (8525.8187ms)
✔ build/source mutation and sealed step tampering reject before browser dispatch; earlier task cannot execute plan (1699.3029ms)
✔ outside network never reaches another local service, and original JavaScript/console failures remain failures (3176.2853ms)
✔ stopping a real in-flight browser preserves finished steps, closes its owned root and never replays (2924.0161ms)
✔ 原PNG损坏使原结果校验失败,不能以已有通过状态复用 (3380.1942ms)
✔ fixed browsing contracts reject local host APIs, credential URLs, protocols, cross-origin and arbitrary execution (0.9884ms)
✔ attested desktop queue preserves same-tab observations, original PNG bytes and approval-bound one-shot actions (1032.0589ms)
✔ page data cannot grant domains, interaction or other Task scope; nonce and ownership remain fixed (1183.1207ms)
✔ unknown dispatched submission blocks even after a new observation, rejects late success and retains explicit acknowledgement (1948.5855ms)
✔ downloads require exact approved address, remain original binary artifacts, and cannot reuse consumed observation (1185.8018ms)
✔ revoking shared persistent profile closes all its tabs and clears only that profile after original desktop acknowledgement (1573.4062ms)
✔ service recovery never reopens old tabs or replays dispatched actions and keeps unknown original receipts (1006.4597ms)
✔ agent browser flow waits for original Task/Run approval, dispatches once, and returns paired original results (4517.9954ms)
✔ changed approved input and stopped dispatched agent action cannot be accepted or replayed (1459.5586ms)
✔ browser inventory gives bounded metadata and original references instead of repeating full website histories (1135.9491ms)
✔ receipt storage failure settles the original dispatched action as unknown and never hangs or accepts late success (1196.1338ms)
✔ input that fits the engine window but exceeds remaining credit diagnoses the credit limit (0.4892ms)
✔ window overflow and historical pauses retain distinct and honest diagnoses (0.2808ms)
✔ R observed reservation shortage keeps the window-fit diagnosis separate from Run/Task credit (0.1087ms)
✔ a budget pause persists concrete text/tool/history/image breakdown and never dispatches an oversized request (3042.1748ms)
✔ explicit frozen preview resumes the same Task in a new Run, preserves completed tools and does not change global config or selected model (4472.6836ms)
✔ insufficient budgets, silently clamped capacity and arbitrary config fields reject without a physical request or mutating task budget (3130.1715ms)
✔ changed session, changed registry, expiry and altered frozen preview all require a new explicit preview (3298.4696ms)
✔ preview preserves unknown invocation status and increasing Task allowance is committed only with confirmed new Run (4487.8459ms)
✔ batch APIs enforce ownership and review fingerprints, apply once, and persist the completed batch across restart (6770.9957ms)
✔ delivery reports bind actual batch identity and reject corrupted snapshot evidence without changing project files (4513.415ms)
✔ a compatible model can seal proposals through the real HTTP loop while application stays behind user review (9161.5631ms)
✔ a sealed batch at the step limit hands off and continues the same Task into actual checks without replaying saved tools (19270.8754ms)
✔ applied batches cannot turn cancellation, other budgets, restart or unknown commands into a review continuation (2412.2856ms)
▶ complete sealed batch hands off after all tools settle with no next model request
✔ complete (5668.7612ms)
✔ later-error (5522.8385ms)
✔ complete sealed batch hands off after all tools settle with no next model request (11192.3968ms)
▶ partial, changed, failed or cancelled batches cannot claim complete review handoff
✔ partial (4743.7783ms)
✔ new-proposal (5073.1071ms)
✔ failed (4904.3697ms)
✔ conflict (5719.9563ms)
✔ cancelled (4433.9481ms)
✔ unknown (5413.1958ms)
✔ tampered (5231.0573ms)
✔ partial, changed, failed or cancelled batches cannot claim complete review handoff (35520.4988ms)
✔ actual process exit at file-batch prepared preserves durable identity and never replays writes (2156.9303ms)
✔ actual process exit at file-batch application_started preserves durable identity and never replays writes (2322.5597ms)
✔ actual process exit at file-batch write_intent preserves durable identity and never replays writes (2058.7775ms)
✔ actual process exit at file-batch staged preserves durable identity and never replays writes (2784.161ms)
✔ actual process exit at file-batch replaced preserves durable identity and never replays writes (3082.4599ms)
✔ actual process exit at file-batch file_recorded preserves durable identity and never replays writes (2578.0587ms)
✔ actual process exit at file-batch finished preserves durable identity and never replays writes (2994.4564ms)
✔ actual process exit at file-batch rollback_started preserves durable identity and never replays writes (3320.3231ms)
✔ actual process exit at file-batch restore_intent preserves durable identity and never replays writes (3518.8311ms)
✔ actual process exit at file-batch restore_staged preserves durable identity and never replays writes (3239.5402ms)
✔ actual process exit at file-batch restored preserves durable identity and never replays writes (3519.8756ms)
✔ actual process exit at file-batch restore_recorded preserves durable identity and never replays writes (3599.9942ms)
✔ actual process exit at file-batch rollback_finished preserves durable identity and never replays writes (2724.4882ms)
warning: in the working copy of 'README.md', LF will be replaced by CRLF the next time Git touches it
warning: in the working copy of 'src/two.mjs', LF will be replaced by CRLF the next time Git touches it
✔ sealed batches preserve bytes until review, then apply and explicitly restore BOM/CRLF/new files (2793.0256ms)
Switched to a new branch 'owner-branch'
✔ a conflict in the last file or Git branch prevents writing every file in the batch (3835.1175ms)
✔ duplicate canonical paths, ancestor collisions, links, dependencies and Windows aliases never form a batch (1288.7747ms)
✔ fingerprints, changed proposals and corrupt snapshot artifacts cannot write or restore any target (2798.8365ms)
✔ a failure after replacement restores only batch bytes and never retries an old application (2017.7979ms)
✔ rollback preserves owner edits, blocks other project writers, and explicit acknowledgement releases recovery (2114.028ms)
✔ an unattempted owner-created file identical to a suggestion remains external and is never removed (2166.0662ms)
✔ active batch requests are serialized and cancellation leaves a verifiable ended batch (1640.1138ms)
✔ batch application and rollback preserve unrelated Git index and worktree bytes (3548.407ms)
✔ recovery reuses exact staged bytes but retains a conflicting temporary file for owner inspection (2693.6165ms)
✔ one Task can seal pending proposals from earlier Runs while snapshots retain each source Run (1439.3998ms)
✔ engineering inspection preserves a dirty Git baseline and treats project rules as data (2841.732ms)
✔ real npm tests and build produce counted results, log artifacts and persistent verification (3204.4331ms)
✔ an actual failing test retains the nonzero exit code and failed count (2706.8909ms)
✔ unchanged automatic checks return original failure evidence, while explicit user reruns and changed source still execute (5478.2204ms)
✔ agent loops cannot hide identical failing checks behind new plan IDs or unrelated reads (11235.1484ms)
✔ zero tests, skipped tests and unrecognized stdout cannot become verification passes (9372.2586ms)
✔ a successful build with no selected tests does not verify the task (2323.8658ms)
✔ stale check plans and forged script names execute no project commands (1620.3087ms)
✔ applying a later proposal invalidates the prior passed engineering snapshot (3208.1711ms)
✔ stopping an actual check terminates its child process and blocks concurrent jobs (2372.3881ms)
✔ check timeout records a failure with its actual termination result (2782.967ms)
✔ source edits during a successful check make the result stale (3542.7962ms)
✔ bounded large logs retain the test footer and readable artifact instead of inventing counts (2385.6413ms)
✔ model-selected checks require a run-bound decision and denial creates no check job (3993.7595ms)
✔ check APIs expose owned plans and jobs and preserve the authoritative results across restart (6380.7083ms)
✔ reviewed changes continue the original task without reopening a generic completed run (3085.3934ms)
✔ concurrent review-continuation API requests accept one run and retain the goal after restart (9736.8866ms)
✔ the model reads owned failure logs without running the check again (3698.2482ms)
✔ process exit at check queued preserves results and never replays commands (1489.4733ms)
✔ process exit at check result_committed preserves results and never replays commands (2182.9802ms)
✔ model code tools keep Task/Run ownership, index evidence and exact proposals across storage recovery (5680.4872ms)
✔ bundled code APIs isolate project results, validate queries and create review-only patches with conflicts (3796.7548ms)
✔ code tool arguments reject malformed ownership-independent inputs before side effects (0.3738ms)
✔ observe-only device authority cannot read, configure or clear project parsing caches (932.3143ms)
✔ pairing is short lived, single-use, generation-bound and validates exact permissions (716.3264ms)
✔ device transaction failure neither consumes the ticket nor leaves a credential (718.2535ms)
✔ devices have project and action boundaries; rotation, expiration, revocation and host changes reject old authority (796.4603ms)
✔ persistent registry stores hashes and a bounded audit, never raw credentials (684.5999ms)
✔ real TLS verifies IP SAN, CA chain, encrypted restart and public fingerprint (956.5917ms)
✔ TLS sealing failure leaves listener closed and no plaintext or successful TLS registration (657.7632ms)
✔ HTTP pairing is concurrent single-use; secure cookie and strict origin/host checks (791.2819ms)
✔ gateway never exposes management routes, arbitrary assets or unpaired project data (1041.8906ms)
✔ authorized project events paginate from original seq without leaking other project events or payloads (1584.1921ms)
✔ remote task projection excludes raw model history, host configuration, tool input and credentials (934.6913ms)
✔ permission checks apply to direct remote mutation, not only UI controls (1032.963ms)
✔ approval fingerprint and exact task/revision are bound; duplicate requests never execute twice (832.682ms)
✔ unknown original operation is recorded and never replayed after response failure (1074.6226ms)
✔ revocation and gateway closure reject old access without invoking runtime stop (989.627ms)
✔ rate limits reject pairing brute force and do not persist candidate secrets (848.21ms)
✔ real Windows CurrentUser DPAPI persists only a sealed CA and reopens it (2222.8306ms)
✔ mobile approval executes the original Runtime command exactly once and rejects the later desktop decision (3075.6991ms)
✔ desktop rejection wins a concurrent mobile request and original command remains unexecuted (2433.9168ms)
✔ phone stop waits for the original process exit and prevents later heartbeats and command replay (3416.1061ms)
✔ Run replacement during parent-stop wait is rejected rather than stopping the new Run (625.9254ms)
✔ diff pages preserve source SHA; report reads verify blobs and exclude unrelated artifact kinds (1261.44ms)
✔ hot upgrade refuses an active gateway and restores closed gateway ownership on failure (534.2537ms)
✔ damaged device permissions or expiry fail closed and never overwrite the original registry (757.871ms)
✔ device revocation destroys its actual authenticated TLS connection (1269.457ms)
✔ companion observes only current Task screenshots, verifies original PNG bytes and loses access immediately on revoke (1087.2087ms)
✔ mobile browser observation is current Task/project scoped, preserves original PNG and exposes no browser control or profile data (1604.0029ms)
✔ actual config HTTP failure preserves the committed selection, config and volatile credential before a successful retry (885.4489ms)
✔ controlled installation pins the native kernel and bridge independently of the existing connection-probe identity (777.9589ms)
✔ controlled profiles disable every native project/terminal provider and native model/retry route using the actual include patch contract (0.6106ms)
✔ owned tool catalogues seal full schemas canonically and reject duplicate or malformed declarations (4.9653ms)
✔ bridge authentication, current Run, process, SDK session and browser-origin checks reject before dispatch (1041.1815ms)
✔ bridge duplicates, unknown tools, malformed arguments and source changes are refused before a second side effect (668.4472ms)
✔ bridge model requests bind the exact catalogue, route and output bound and preserve raw request/result identity (608.0129ms)
✔ stopped and oversized bridge requests never reach a local model or tool callback (804.0804ms)
✔ a Windows virtual data-directory name binds new storage and credentials to its actual physical root (2040.2004ms)
✔ physical data-root adoption still rejects explicit junctions at the directory or any ancestor (758.221ms)
✔ actual Windows CurrentUser DPAPI roundtrips owned UTF-8 and rejects damaged bytes and wrong entropy (1290.8274ms)
✔ only explicitly supplied keys persist; new provider restores original revisions and public views contain no key (2471.3713ms)
✔ registry and cipher commit failure restores original routing, keys, block and every database row (1799.0383ms)
✔ a two-key import with a protection failure rolls back all registry and cipher entries and retains its preview (1936.016ms)
✔ connection revision changes retain the original saved block and report a source mismatch on restart (2793.6467ms)
✔ disable skips real restart loading, explicit enable restores, delete preserves memory and forget removes both (2950.147ms)
✔ corrupt saved block and another SID do not load; failures are visible without returning ciphertext or plaintext (2904.4714ms)
✔ legacy persistence requires a new explicit key and binds config and credential revisions; stale requests cannot save (2783.0931ms)
✔ owned plaintext keys never reach SQLite pages, WAL, config or public execution exports (1451.0129ms)
✔ reviewed application invalidates old checks, names current SHA and advances through a new approved check to Git preview (9072.9421ms)
✔ skip-tests, foreign task, unknown operations, diagnostic and parent review retain their boundaries (4100.7219ms)
✔ before the first change, real baseline failures persist across later reads; passed baselines and changed sources cannot certify delivery (7442.7343ms)
✔ actual process exit at Git queued never replays a commit and can reconcile the durable result (12443.2937ms)
✔ actual process exit at Git preparing never replays a commit and can reconcile the durable result (11887.1892ms)
✔ actual process exit at Git index_locked never replays a commit and can reconcile the durable result (11648.5678ms)
✔ actual process exit at Git prepared never replays a commit and can reconcile the durable result (12893.4971ms)
✔ actual process exit at Git ref_prepared never replays a commit and can reconcile the durable result (13220.5997ms)
✔ actual process exit at Git publishing never replays a commit and can reconcile the durable result (12730.1526ms)
✔ actual process exit at Git committed never replays a commit and can reconcile the durable result (10507.5493ms)
✔ actual process exit at Git index_published never replays a commit and can reconcile the durable result (9699.3664ms)
✔ actual process exit at Git finished never replays a commit and can reconcile the durable result (9872.2096ms)
✔ an unknown Git result cannot overwrite an externally changed index or delete an unrelated lock (8980.0849ms)
✔ known publication with a later source edit retains the commit and marks the delivery for inspection (8978.848ms)
✔ CRLF normalization, multiple applied revisions and unrelated special index flags preserve exact bytes and ownership (10129.9927ms)
✔ adding new durable record kinds to an existing hot Store preserves all existing Maps and records (8767.3081ms)
✔ an unrelated owner deletion is fingerprinted and preserved while linked task paths are rejected (13403.1373ms)
✔ actual failed tests become a reviewed local commit and evidence report while preserving unrelated index and files (12467.2974ms)
✔ selected preexisting unstaged, staged and untracked owner edits cannot enter a task commit (15212.6226ms)
Switched to a new branch 'owner-branch'
✔ source, unrelated stage and branch changes after review invalidate a commit plan without changing HEAD (18927.8929ms)
Switched to a new branch 'delivery-branch'
✔ a later Git metadata change keeps matching source evidence valid but a real source edit requires checks (9275.6086ms)
✔ a corrupt passed-check log and a pending proposal block Git delivery instead of inventing evidence (5289.0425ms)
HEAD is now at 60e5601 Sealed delivery fixture baseline
✔ Git filters, merge state, detached HEAD and old task ownership are explicitly rejected (20683.8637ms)
✔ local snapshot commits disable repository hooks and signing and isolate the Git child environment (8664.4923ms)
✔ stopping before publication cancels safely; stopping after publication preserves the actual completed commit (16272.0301ms)
✔ delivery API confirms exactly one owned plan, serves verified reports and preserves results across restart (12681.3208ms)
✔ the model can prepare an owned Git review and evidence report while publishing remains a separate user action (10393.8815ms)
✔ HTTP Git review pairs deferred calls, survives reload and never requests more work until the saved preview is resolved (9718.5094ms)
✔ a reviewed Git preview can be revised or withdrawn without publishing or losing the current Task (8739.0906ms)
✔ actual Electron debug argument ordering retains exactly the compiled entry and rejects extra or repeated actions (1.1463ms)
✔ native bridge refuses other origins, subframes, other windows and unsupported inputs before connecting (567.6742ms)
✔ desktop preview opening requires a currently registered project preview and rejects iframe callers (655.5613ms)
✔ file reveal binds the current project, task, path, UTF-8 bytes and original SHA without executing the file (551.9996ms)
✔ file changed during asynchronous task recheck is not revealed as the old bytes (645.0377ms)
✔ desktop browser profile and aliases stay outside project inventory and file access (682.0177ms)
✔ native picker cancel and origin change after dialog publish no project registration (549.4353ms)
✔ pending native action is not dispatched twice, and current task review returns its original seq (535.7096ms)
✔ native stop cancellation and changed original host never stop a replacement process (727.9385ms)
✔ window register rejects host PID, stale installation, extra fields and missing creation identity (0.5369ms)
✔ a rebuilt source archive binds actual generated client assets and retains its original running identity after disk changes (2033.9198ms)
✔ host challenge proof rejects a stale challenge, changed identity and changed signature (0.711ms)
✔ stop confirmation binds the original instance, creation and loaded runtime (0.1225ms)
✔ state paths reject escaping, linked ancestors, ambiguous hardlinks and damaged JSON (1148.1999ms)
✔ failed registration publishes no initialized memory and can complete after its actual inputs are supplied (2755.4258ms)
✔ host shutdown cancels an un-signalled HTTP model request and refuses later requests until recovered (1.386ms)
✔ host shutdown cancels model-pool acquisition and releases queue ownership without inventing attempts (717.3441ms)
✔ full candidate validation returns the exact loaded-input host identity and rejects undeclared or altered files (1236.489ms)
✔ fast per-file scan still rejects a junctioned subdirectory, a hard-linked file and a stream path with the original bytes intact (942.4754ms)
✔ installation requires separate absolute program/data directories and a candidate in its own versions namespace (743.8679ms)
✔ inventory rejects a new hardlink on an optional file after its full-manifest scan and accepts the untouched candidate after removal (739.1508ms)
✔ staging retains the original data and pointer without querying or restarting a host (2991.9501ms)
✔ a post-pointer failure restores every original pointer byte and persistent row; a first-install failure removes only its published pointer (10319.0435ms)
✔ changed active manifest and concurrent pointer changes are rejected without overwriting the unknown record (4479.5978ms)
✔ an unregistered non-HTTP listener blocks preflight even when service-info is absent (2752.3205ms)
✔ uninstall archives its original pointer and preserves every data row/artifact instead of recursively deleting data (6584.5577ms)
✔ preflight rejects an unrelated program root or changed native bootstrap while preserving its original bytes (4485.0288ms)
✔ a changed owned bootstrap follows its candidate and a post-pointer fault restores both original bytes (17562.0755ms)
✔ an external bootstrap change after preflight is preserved and never overwritten as an update or rollback (7708.7991ms)
✔ owned startup retries only read-only observation after a slow health response and attests the same original child (3861.2288ms)
✔ unverified startup expires with its original live child and never treats timeout or an unrelated health PID as proof (5941.7772ms)
✔ diagnostic failures and repeated questions retain the original execution error, task, checks and read-only authority (2920.9883ms)
✔ diagnostic binding survives reopen, rejects foreign Task references and detects original error or artifact tampering (957.739ms)
✔ a new Task never adopts a previous Task error from session.lastFailure (706.3494ms)
✔ verified research and explicit no-test scopes do not demand an unrelated test job or claim engineering success (3034.5551ms)
✔ unknown scope and unknown operations remain actionable even after a diagnostic answer or read proof (1128.1203ms)
✔ delivery status HTTP is read-only, selection-bound and equal to the saved report after host restart (4758.9994ms)
✔ routine engineering facts remove duplicate success excerpts while complete diagnostics and errors remain available (706.6179ms)
✔ a question about the previous workbench error receives local error context and current change states (2487.4301ms)
✔ an unknown model request stops, and explicit resume never replays the approved command (4013.4843ms)
✔ exhausted refused connections preserve tool results, and explicit resume adds no new user turn or replay (3256.2825ms)
✔ stopping during model backoff cancels further requests and clears the running request state (1485.3298ms)
✔ HTTP authentication errors are classified, redacted and never automatically retried (17.059ms)
✔ temporary HTTP service errors retry and recover with recorded attempt counts (7.9185ms)
✔ invalid and oversized protocol responses fail once with a useful response-format diagnosis (5.4401ms)
✔ HTTP rate limits expose their source and obey the service retry interval before recovery (1019.9198ms)
✔ Windows commands preserve Chinese through a PowerShell pipeline into Node (1005.7614ms)
✔ Windows shell syntax errors stay readable and native command exit codes are retained (1127.7744ms)
✔ review and completed states do not turn a generic continue click into a new model request (763.8939ms)
✔ old session data receives a truthful historical connection diagnosis without rewriting history (729.5738ms)
✔ engine HTTP checks bind the detected installation and current task, page original bytes and preserve them across host restart (5312.5855ms)
✔ HTTP stop seals a pending SDK call, rejects automatic replay and keeps failures attached to the original session (4613.7829ms)
✔ sealed r1-off failure fits its original 16K config with authentic latest status and readable source (12.6751ms)
✔ sealed r1-on failure fits its original 16K config with authentic latest status and readable source (9.5495ms)
✔ sealed r2-on failure fits its original 16K config with authentic latest status and readable source (8.0372ms)
✔ sealed r2-off failure fits its original 16K config with authentic latest status and readable source (7.4741ms)
✔ archived native call pairs require exact saved arguments/results rather than model-written summaries (8.1951ms)
✔ full system/user requirements and mandatory current failure can refuse rather than be trimmed (12.276ms)
✔ small native input is byte-preserving; developer instructions remain mandatory after compaction (0.1692ms)
✔ older failing checks and pending proposals remain visible after a later large file read (14.8525ms)
✔ bounded result keeps every check status and cuts Unicode only on whole characters (0.4553ms)
✔ controlled engine HTTP freezes one child, pages original owned input/report, rejects foreign reads and prevents changed-source reuse (14613.7483ms)
✔ a sealed whole-batch review pauses the owned native session without certifying engine completion or losing the saved host result (15856.7489ms)
✔ a changed review seal still closes the owned native process and cannot claim a verified review handoff (15011.1703ms)
✔ native text and call IDs preserve original result pairing while unsupported content fails closed (0.4484ms)
✔ one engine child freezes real installation identity before creating tasks and rejects a changed source (8664.6541ms)
✔ owned native tools return saved results, use real approvals/checks and account each actual request once across new engine Runs (28471.3222ms)
✔ forged native scope cannot change a parent file and a final completion claim cannot replace worker proof (13547.9268ms)
✔ duplicate native IDs are rejected without replaying a saved tool (12290.6248ms)
✔ native input exceeding the context budget is retained and sends no model request (11491.9288ms)
✔ bounded native sends verify original tool provenance and reject forged views or missing versions (14013.5138ms)
✔ stop retains a late reported model usage once and executes no tool from that output (11674.9891ms)
✔ a source change after the original model intent stops before native tools execute (11059.151ms)
✔ an engine source change while approval is pending invalidates the old permission before any check starts (14542.3449ms)
✔ missing or noncompleted turn end cannot be adopted even with an owned receipt and assistant text (0.5467ms)
✔ closed root with an original live descendant remains unknown and retains its process tree (0.2846ms)
✔ host recovery reserves original pending attempts exactly once and explicit review preserves unknown outcomes (15708.4285ms)
✔ corrupted model reservation cannot reduce recovered budget or permit a new engine request (12564.4552ms)
✔ engine detection seals the actual CLI, SDK, profile and Node and rejects mixed versions without launch (1416.4239ms)
✔ owned engine probe waits for real ready identity, verifies six native responses and saves immutable task evidence (1601.2789ms)
✔ an invalid startup nonce is rejected before initialize and preserves the process close and partial report (1687.058ms)
✔ SDK identity and session directory mismatches retain raw responses and never become verified (4773.186ms)
✔ cancel during a pending initialize closes only the owned process without sending an unsupported MCP cancellation (1612.6598ms)
✔ cancel before readiness sends no protocol request and a user retry creates a different process and record (2027.7956ms)
✔ source or scratch changes and unexpected model traffic fail after preserving the actual observed result (4779.6312ms)
✔ unowned idle notifications stay quarantined and server requests receive a rejection without granting operations (2791.7697ms)
✔ engine probe environment excludes parent secrets and NODE_OPTIONS while preserving operating system paths (1754.0769ms)
✔ foreign sessions and altered reports or task bindings cannot reuse verified engine evidence (1392.8969ms)
✔ restart never replays an active engine check or stops a historical PID, and keeps prior task state (1384.8549ms)
✔ model argument validation and runtime activity do not grant external engine prompt or project execution (584.9585ms)
✔ a follow-up question receives the actual failed engine check from its original Task and keeps its report unchanged (2354.6418ms)
✔ unfixed dependency URLs and invalid integrity are reported without exposing URL credentials or allowing a plan (970.2117ms)
✔ hot upgrade refuses an active dependency operation or evaluation before changing runtime objects (0.6146ms)
✔ read-only environment diagnosis identifies actual runtime, missing dependencies and sealed inputs without installing or running scripts (6440.8037ms)
✔ diagnosis rejects unsupported managers, lock integrity and credential configuration without reading config content (5255.3544ms)
✔ approved real npm ci runs only in owned worktree, skips scripts, retains logs and binds later actual tests (12109.2822ms)
✔ explicitly allowing lifecycle scripts executes sealed root script and preserves the separate preparation result (8516.2152ms)
✔ stale lockfile, tampered plan/config, cross-session identities and mismatched download sources never dispatch npm (7013.165ms)
✔ registry failure keeps original npm exit/log and performs no automatic fallback or replay (8690.2833ms)
✔ stopping a hanging dependency download records only its owned process and preserves stopped logs (8503.2652ms)
✔ restart retains unknown preparation, blocks writes and checks, and requires explicit read-only reconciliation (7222.9085ms)
✔ HTTP environment flow survives refresh, requires explicit approval and preserves task and artifact ownership (13598.1262ms)
✔ Python environment diagnosis reads declarations and runtime identity without executing project code or installing packages (877.5173ms)
✔ empty build cache directories keep dependency identity while added executable files invalidate it (1175.9152ms)
✔ dependency snapshots reject external links and incomplete budgets instead of claiming environment equivalence (738.475ms)
✔ model-requested environment preparation uses the original approval boundary and never treats preparation as passed engineering checks (13103.4071ms)
✔ timeout and offline cache miss retain actual failures and new attempts require newly sealed plans (13172.4712ms)
✔ both evaluation worktrees can approve fixed dependency preparation and different actual dependencies reject candidate execution (30613.8257ms)
✔ evaluation denominator distinguishes attempted unsealed candidates from unexecuted samples (0.3284ms)
✔ evaluation summary retains sealed outcomes and stopped or failed original attempts (0.4341ms)
✔ evaluation contract fixes bounded samples, source paths, checks and aggregate budget (520.9411ms)
✔ same-source workspaces seal real commands before execution and preserve original staged and unstaged bytes (14548.4094ms)
✔ actual original failures and candidate success are both preserved, then adoption requires parent review, checks and Git (59730.9345ms)
✔ candidate cannot shrink tests and changing original test bytes rejects execution before any model request (24617.7589ms)
✔ missing upstream usage remains unknown and does not become a free run or invented cost (37862.1556ms)
✔ missing raw check logs preserve unknown result and original sample denominator (40760.2225ms)
✔ source and model changes never silently replace an existing batch input (3791.0988ms)
✔ model revision change before preparing a fixed sample cannot silently use the replacement connection (4009.3538ms)
✔ HTTP evaluations require host token, original fingerprint and session ownership; worker API cannot bypass evaluation flow (18641.0947ms)
✔ sample queue cannot skip or reset attempted samples, and reports freeze further writes (29217.7581ms)
✔ stop cancels only current candidate and preserves partial results instead of replaying a request (23371.9878ms)
✔ restart recovers interrupted evaluations as unknown and keeps completed checks and plan identity (21343.7956ms)
✔ JSON migration survives process exit before commit and retains exact legacy backups (2260.7326ms)
✔ process exit after_state rolls back state, event and outbox together (1681.142ms)
✔ process exit after_event rolls back state, event and outbox together (1772.751ms)
✔ crash recovery at accepted retains execution identity and never replays side effects (1742.0504ms)
✔ crash recovery at queued retains execution identity and never replays side effects (2841.8739ms)
✔ crash recovery at approval_wait retains execution identity and never replays side effects (2887.0753ms)
✔ crash recovery at approved retains execution identity and never replays side effects (3277.4834ms)
✔ crash recovery at effect_unrecorded retains execution identity and never replays side effects (3197.9401ms)
✔ crash recovery at result_committed retains execution identity and never replays side effects (3484.745ms)
✔ crash recovery at before_next_model retains execution identity and never replays side effects (3471.3909ms)
✔ state and event failures publish no success and preserve the last committed memory snapshot (557.2873ms)
✔ tool result artifacts and event cursors retain project ownership across restart (1991.3526ms)
✔ token budgets pause before tools execute and continuation keeps original task and accumulated usage (2809.5838ms)
✔ a time budget interrupts an unresponsive model and persists a resumable run (1109.7012ms)
✔ a corrupted large result artifact cannot become a recovered success (1045.0376ms)
✔ hard-link aliases cannot expose protected bytes through reads, proposals, inventory or automatic project rules (859.9567ms)
✔ ambiguous project hard links are explicit and ordinary independent file copies remain readable (563.5942ms)
✔ current-file HTTP receipts bind Task/Run and bytes, persist on restart and end without altering original runs or user Git data (4734.4782ms)
✔ file targets exclude referenced inputs and prohibited subsequent modifications (1.5873ms)
✔ full actual read and existing receipt complete a no-test file task without fabricating changes, checks or commits (3083.9199ms)
✔ partial pages, partial lines, wrong hashes and failed reads cannot be adopted; contiguous full paging can (1264.0329ms)
✔ user edits and new requirements invalidate receipts, and a second task must explicitly adopt a source in its own project/session (1093.9175ms)
✔ adopting an unchanged file never satisfies an explicit modification or a new modification instruction (1079.7306ms)
✔ required checks and stale passing flags cannot be bypassed by current file receipts (1254.4474ms)
✔ results plan proofs bind current file receipts while changes proofs still demand real applied change sets (814.2043ms)
✔ duplicate receipts cannot create progress indefinitely or reset no-progress reads (3030.7375ms)
✔ manual completion preserves the failed original run and refuses unknown operations (663.5766ms)
✔ completion refuses active work, files changed during inspection and a newly selected task (876.5196ms)
✔ unrelated files cannot stand in for named output paths, nor can a late apply satisfy newer instructions with an old proposal (952.0469ms)
✔ delivery and mobile projections distinguish adopted bytes, new changes, no tests and no commit; mutations cannot rewrite receipts (1088.6113ms)
✔ proposals preserve project bytes until review; apply backs up the exact original bytes (721.8496ms)
✔ new nested files are created only on apply and unchanged content creates no proposal (569.0239ms)
✔ stale reads and changed baselines are rejected without overwriting external edits (565.2661ms)
✔ file paths reject traversal, absolute paths, credentials and outside junctions (923.9816ms)
✔ read and search expose bounded text scopes and reject non UTF-8 / oversized files (1285.0853ms)
✔ project-local junctions cannot expose or create protected Git and credential files (538.4753ms)
✔ Windows trailing-dot and trailing-space path aliases cannot bypass protected paths (639.0665ms)
✔ a starter is copied into user data once; restarting from another program version preserves existing projects and editable bytes (1038.5408ms)
✔ startup does not insert a demo or rewrite/reorder an existing user project registry (605.8176ms)
✔ PNG bytes are fully decoded, metadata-free and bounded; JPEG EXIF orientation is applied to the sent PNG (104.4062ms)
✔ forged type, short headers, CRC damage, incomplete chunks, trailing payloads, animation and dimension bombs reject (23.9268ms)
✔ draft image ownership, withdrawal, reload and atomic Task binding retain original SHA and original bytes (1223.6749ms)
✔ a failed session transaction restores draft status and cannot leave a Run consuming the attachment (845.6106ms)
✔ draft limits, duplicate selection, source changes and cross-Task association reject before dispatch (1109.9509ms)
✔ demo and legacy paths refuse image dispatch without creating a Task or consuming the draft (603.713ms)
✔ chat-completions sends one owned image per actual request, pairs original tools and verifies exact sealed bytes (5663.3988ms)
✔ responses sends one owned image per actual request, pairs original tools and verifies exact sealed bytes (2805.742ms)
✔ mixed pool excludes unknown vision before lease and sends zero HTTP to the incompatible member (1982.4898ms)
✔ vision probe requires correct randomized pixel order and evidence expires when the connection revision changes (1077.3518ms)
✔ HTTP upload/read/withdraw uses session authorization, returns byte SHA and refuses unknown ownership (1013.2817ms)
✔ image accounting excludes Base64 as text but seals every original request byte and reserves image input separately (1.7609ms)
✔ actual model comes from owned consumption, excludes other Tasks and deduplicates the current Run (1.6719ms)
✔ mixed routes retain each original model/protocol/outcome and unknown use is never reported as zero (0.1348ms)
✔ zero dispatch, missing ledger, missing model and truncated Run history never borrow configured model identity (0.2058ms)
✔ repeated attempt identities across Runs and contradictory usage stay explicit instead of inflating successes (0.1059ms)
✔ observed route never certifies returned identity and retains mismatches and original source (0.1509ms)
✔ directory picker reads one level, preserves empty folders, skips credential and linked folders (800.9969ms)
✔ HTML preview serves encoded names and relative modules on an isolated loopback origin; closing revokes it (607.7677ms)
✔ HTML preview refuses credentials, hidden inventories, traversal, external junctions, hardlinks and unsupported file types (775.0086ms)
✔ runtime shutdown closes local preview ports and pending starts cannot escape shutdown (612.8921ms)
✔ idle file previews expire and release their loopback servers (1169.7192ms)
✔ directory and preview HTTP controls require workbench authentication; project pages cannot read bootstrap (2340.0939ms)
✔ MCP registration stays disabled until an explicit real stdio probe and schema-based tool selection, with actual process cleanup (2148.9203ms)
✔ MCP probe holds the project operation lock until its owned process and report are settled (1654.3688ms)
✔ MCP validates the declared JSON schema subset and rejects unsupported schemas before external dispatch (0.6117ms)
✔ actual model MCP call approves long schema-valid arguments, validates the sent catalog and stores one original result without leaking environment secrets (5025.3734ms)
✔ denying an MCP call never starts another process or executes the external tool (3322.2307ms)
✔ actual handshake drift rejects tools/call even if program entry hash is unchanged (4553.3356ms)
✔ cancelled dispatched external side effect stays unknown, closes its owned process and cannot replay on reopen (3928.4946ms)
✔ changed executable input, foreign project IDs and corrupted original contracts block enable and dispatch (1744.2249ms)
✔ registered MCP HTTP API uses fixed execution contracts and project-owned probe records with no model requests (3326.8985ms)
✔ unknown MCP pauses later tool calls and all project writers; fingerprinted acknowledgement clears only the review gates and explicit resume does not replay (6617.2403ms)
✔ MCP acknowledgement persistence failure restores both job and tool review gates; corrupt evidence cannot be acknowledged (4755.8421ms)
✔ actual host exit after an external effect restores unknown jobs without a fabricated report, allows source-bound review and never replays the call (3533.8928ms)
✔ changed enabled MCP programs can be explicitly disabled without launching or trusting their changed source (1706.3952ms)
✔ stdio preserves split UTF-8 frames and final replies drained after process exit (2195.174ms)
✔ server-initiated model requests are refused instead of inheriting provider privileges (994.6094ms)
✔ invalid or over-budget protocol invalid closes only its owned process and is not retried (946.6131ms)
✔ invalid or over-budget protocol oversized closes only its owned process and is not retried (944.1028ms)
✔ cancel and timeout do not resend a dispatched request and pre-cancel has no request trace (1025.5516ms)
✔ startup failure remains undispatched and has no process PID to terminate (485.2674ms)
✔ REA launch filters a fixture credential canary and rejects changed entry identity before a new connection (676.3986ms)
✔ schema1 connections remain readable without inventing observed capabilities or migrating original state (646.4926ms)
✔ protocol, unsupported capabilities and stream choice persist; client claims cannot forge verified evidence (1059.733ms)
✔ ability evidence belongs to exact connection and credential revision; unknown and unsupported reject required routing (0.3011ms)
✔ invalid declarations and inconsistent stream settings reject before registry, revision and credentials mutate (487.2829ms)
✔ upstream declarations include brand metadata and remain distinct from adapter support or verification (1.5544ms)
✔ new upstream retrieves more than 100 models without a manual model; choosing several stores one key and frozen metadata (771.1171ms)
✔ Responses and DeepSeek list use bearer; Messages root and v1 endpoints paginate with selected authentication (709.1238ms)
✔ refresh preserves existing key objects, encrypted block, parameters and user capability overrides without re-encryption (1949.5777ms)
✔ shared credentials restore from the original saved block; source disable, endpoint changes and key replacement cannot revive frozen runs (3653.8355ms)
✔ automatic selection reuses capability and health filtering, manual selection sends the exact chosen model and shared auth (981.9325ms)
✔ unknown catalog capabilities require explicit declarations; upstream negative and retired models cannot be overridden (817.7953ms)
✔ model reasoning controls use declared effort and mode ranges without silently enabling unsupported options (661.2865ms)
✔ cancelled, expired, changed-source and failed catalogs cannot save or overwrite a later selection (823.939ms)
✔ pagination loops, conflicting models and upstream errors fail once without creating members or model attempts (810.2492ms)
✔ limits and transaction failures preserve original registry, config, cipher and preview for explicit correction (1864.7219ms)
✔ a model that switches from a shared to an explicit credential receives a new revision for opaque-history isolation (819.8612ms)
✔ compiled HTTP directory flow persists selection and references, gates changes during fetch and permits explicit cancel (2618.1523ms)
✔ large directories declare partial coverage and a batch crossing the local connection limit leaves no partial members (4558.5212ms)
✔ a distinct upstream account at the same endpoint and model never silently reuses another account (823.9648ms)
✔ source changes invalidate an in-flight child probe and host shutdown cancels directory work and drops transient previews (830.7536ms)
✔ model registry keeps original credentials/endpoint intact, rejects stale edits, and forgets only volatile keys on restart (1296.593ms)
✔ config import previews exact revisions, keeps secrets volatile, syncs only when enabled and preserves manual edits on conflicts (954.4239ms)
✔ real HTTP pool fails over 429 to another model and never repeats completed tools; attempts seal actual request bytes and Run identity (3258.4979ms)
✔ pool rotation and weights select members on new Runs; disabled and missing credentials are excluded without requests (7176.3849ms)
✔ 401 blocks the credential revision; updating the key restores availability without leaking error echoes (2571.3836ms)
✔ queued request cancellation and in-flight disable release only owned leases and do not dispatch another HTTP request (1296.524ms)
✔ compiled model-center HTTP API persists public profiles, blocks forged origins, and local operations never invoke a model (2699.687ms)
✔ model registry disk failures restore routing, pool profiles and credential references without partial updates (750.9182ms)
✔ model selection and saved configuration commit together and failures preserve the original mode and endpoint (888.1601ms)
✔ configuration imports roll back the entire batch and retain one usable preview after validation or commit failure (792.847ms)
✔ HTTP quota retains native code/type/param/request id, bounded original and allowlisted headers, never repeats that connection (2231.2039ms)
✔ Retry-After above 90 seconds remains exact and a short local budget cannot trigger an early retry (1.9014ms)
✔ Retry-After date, expired date, huge value and invalid header retain explicit semantics (1.1314ms)
✔ unclassified 429 remains unspecified and every bounded actual retry retains its own original error (2073.7475ms)
✔ oversized HTTP error saves exactly the bounded prefix and labels its incomplete range (1884.9314ms)
✔ Chat SSE error preserves upstream fields, original failed event, partial tool and reported usage without executing or retrying (2076.4266ms)
✔ Responses failed SSE retains original event/response identity and usage (1915.1818ms)
✔ Responses incomplete SSE retains original event/response identity and usage (2733.2182ms)
✔ unsupported successful Responses output retains original JSON and usage with a local protocol origin (2483.5281ms)
✔ HTTP 200 native error is still an upstream failure rather than a fabricated successful answer (0.8808ms)
✔ JSON-escaped credential echoes are redacted in the original preview and capture as well as parsed fields (0.6678ms)
✔ quota failover uses another independent pool member and never replays a failed stream (2181.3185ms)
✔ native parameters are isolated by pool member, preserve Responses state and bind the original output reservation (0.3146ms)
✔ unsupported native fields and invalid values reject before dispatch instead of silently dropping or overriding constraints (0.3186ms)
✔ native settings for legacy and registered profiles persist without changing credentials or old records and reject stale edits (1166.1339ms)
✔ compiled native settings API saves/restores exact parameters and rejects reserved fields without mutating the profile (2551.8966ms)
✔ legacy connection test sends its saved native fields and selected output limit field without changing the configuration (2433.2619ms)
✔ identity distinguishes missing, invalid, mixed, exact and explicitly declared aliases without guessing (1.2289ms)
✔ chat-completions JSON rejects a different model before tool dispatch, retaining source and reported usage (1982.1313ms)
✔ chat-completions SSE rejects a different model before tool dispatch, retaining source and reported usage (2222.615ms)
✔ responses JSON rejects a different model before tool dispatch, retaining source and reported usage (2095.6033ms)
✔ responses SSE rejects a different model before tool dispatch, retaining source and reported usage (1570.9742ms)
✔ record policy accepts missing identity but keeps it explicit, and strict declared mapping is frozen in the consumption (1.1942ms)
✔ mixed streamed identities cannot pass strict mode, and interrupted streams retain observed identifiers without authenticating them (1.1987ms)
✔ identity refusal never fails over to another pool member and raw response model cannot leak known credentials (1820.0459ms)
✔ strict refusal remains visible even when the same original response exhausts the budget (1705.088ms)
✔ strict identity crosses the controlled engine bridge with its original source and zero tool dispatch (11655.9138ms)
✔ frozen direct connection policy and delivery reports preserve response identity without filling historical gaps (2308.5435ms)
✔ saved engine budgets verify all Chat/Responses route variants and reject a mismatched output allowance (820.8909ms)
✔ owned CC Switch SQLite imports the selected TOML Responses wire_api and leaves unknown clients untouched (927.7771ms)
✔ Responses executes one original tool through Runtime, carries encrypted Items, reports raw usage and seals exact wire bytes (2544.4875ms)
✔ mixed-protocol pool keeps both actual variants and changes route on 429 without re-executing completed tools (2955.5739ms)
▶ Responses incomplete, failed, missing call identity and unsupported hosted outputs never dispatch a tool
✔ Reject {"body":{"id":"resp_owned","status":"incomplete","output":[{"type":"function_call","id":"fc_partial" (1548.4997ms)
✔ Reject {"body":{"id":"resp_owned","status":"failed","output":[{"type":"function_call","id":"fc_failed","cal (1484.4608ms)
✔ Reject {"body":{"id":"resp_owned","status":"completed","output":[{"type":"function_call","id":"fc_bad","cal (1697.5483ms)
✔ Reject {"body":{"id":"resp_owned","status":"completed","output":[{"type":"computer_call","id":"not-supporte (1694.0593ms)
✔ Responses incomplete, failed, missing call identity and unsupported hosted outputs never dispatch a tool (6425.3417ms)
▶ invalid JSON in a nonstream Chat or Responses tool group prevents every tool and pool replay
✔ chat-completions (2380.7507ms)
✔ responses (1628.7034ms)
✔ invalid JSON in a nonstream Chat or Responses tool group prevents every tool and pool replay (4009.8367ms)
✔ wire adapter preserves tool IDs but never passes Responses-private Items to Chat or another endpoint (0.1745ms)
✔ connection JSON import accepts explicit Responses protocol and continues to reject unknown protocols without reading client sessions (635.4258ms)
✔ reservation uses actual known usage from the same route, never lowers estimates or rewrites original receipts (1.7616ms)
✔ reservation verification binds original Task/Run consumption and rejects changed usage or foreign provenance (0.6391ms)
✔ known high input calibrates the next request and stops before dispatch while preserving the completed read (2018.2593ms)
✔ one reported overrun stops legacy requests and keeps actual use and excess (1489.9762ms)
✔ one reported overrun stops pool requests and keeps actual use and excess (2010.425ms)
✔ one reported overrun stops connection requests and keeps actual use and excess (1757.0162ms)
✔ reported usage below the limit prevents an unaffordable legacy retry before another dispatch (1476.4626ms)
✔ reported usage below the limit prevents an unaffordable pool retry before another dispatch (1683.6779ms)
✔ native bridge seals calibrated refusal with original consumption and preserves versioned request verification (13959.2324ms)
✔ original R v1/v2 stay verifiable while v3 excludes the isolated input outlier and keeps the ledger intact (4.4554ms)
✔ v3 never excludes a repeated high ratio, the latest receipt, or a short history (3.0852ms)
✔ reservation proofs reject reordered latest high receipts, altered history and duplicates while old snapshots survive later consumption (3.234ms)
✔ input-ratio exclusion preserves independently reported output maxima without rewriting historical v2 (1.7571ms)
✔ capability routing skips text-only members before acquiring a lease and records exact refusal in the original Run (1818.6447ms)
✔ all incompatible members reject with no network request, no lease, preserved Task and explicit non-dispatched ledger (1398.9034ms)
✔ explicit synthetic tool probe verifies only actual matching call, never executes it and loses verification on revision changes (857.7748ms)
✔ plain text probe cannot verify tools; wrong tool nonce and client supplied evidence cannot claim verified support (1025.6032ms)
✔ a disabled connection invalidates an in-flight probe even when toggle mutates its original registry object (769.5207ms)
✔ selected connection owns the actual endpoint, credential, wire model and mismatch even when legacy is configured for GPT (3110.2214ms)
✔ budget-only API and demo/real selection preserve routes, reject injected endpoint fields and publish authoritative selection without keys (2727.9677ms)
✔ another session cannot change default model, mode or budgets while a request is active (5051.8835ms)
✔ HTTP download preserves original malformed SSE bytes and artifact SHA instead of decoding invalid UTF8 (2364.3664ms)
✔ terminated SSE adapts empty optional tool arguments, retains the exact wire bytes and executes the validated batch once (3337.3605ms)
✔ reported SSE usage survives rejected required arguments "" without dispatching any tool (2404.2812ms)
✔ reported SSE usage survives rejected required arguments "{\"path\":" without dispatching any tool (2403.8382ms)
✔ reported SSE usage survives rejected required arguments "null" without dispatching any tool (2028.8623ms)
✔ a stream without DONE retains reported usage but cannot adapt or execute even an optional tool (2487.8544ms)
✔ a timeout after reasoning-only fragments creates one unknown receipt and never invents reported tokens (2868.8506ms)
✔ real Chat SSE carries UTF8 and split tool JSON; Runtime executes only after DONE, stores partial progress and original usage (3587.8633ms)
✔ Responses typed SSE validates original item identity, terminal data, raw usage and tool result pair (3810.7055ms)
✔ incomplete network stream retains unknown attempt and never automatically switches pool or dispatches fragment tools (2760.5516ms)
✔ stopping an active HTTP stream closes only original request and keeps partial text without executing tools (2203.6925ms)
✔ explicit streaming probe uses actual SSE despite unknown declaration and binds its own original attempt (1065.6427ms)
✔ coalesced initial SSE events publish the latest partial state even when the upstream stays idle (2615.4659ms)
✔ stream timeout closes the original HTTP response, preserves redacted capture and never retries (1429.2899ms)
✔ stream idle timeout preserves incomplete tool arguments, records unknown use and never fails over (1928.6138ms)
✔ SSE accepts CR-only final event boundaries and retains exact invalid UTF8 bytes without dispatch (0.8158ms)
▶ stream parser rejects truncation, finish length, malformed JSON, duplicate sequence, changed final item and oversized data
✔ reject 0 (0.2684ms)
✔ reject 1 (0.194ms)
✔ reject 2 (0.1509ms)
✔ reject 3 (4.9808ms)
✔ reject 4 (0.548ms)
✔ reject 5 (0.1875ms)
✔ stream parser rejects truncation, finish length, malformed JSON, duplicate sequence, changed final item and oversized data (6.873ms)
✔ compiled host retains interrupted HTTP 200 and explicitly resumes after the recorded cooldown (10621.0815ms)
✔ compiled host retains interrupted HTTP 503 and explicitly resumes after the recorded cooldown (10484.4624ms)
✔ owned IPv4 service works after a failed IPv6 candidate without a second model POST or global DNS change (26.0255ms)
✔ environment HTTP proxy tunnels the owned request and NO_PROXY sends a separate direct request (10.4572ms)
✔ original connect timeout has a bounded nested cause and does not masquerade as response generation timeout (0.7159ms)
✔ POST accepted then reset is unknown, is charged once and never automatically retried (3.5823ms)
✔ partial HTTP bodies retain original request id and bytes, and an unknown pool attempt never switches model (2883.0906ms)
✔ response wait timeout after an accepted POST is unknown; cancellation closes the owned socket and permits explicit reuse (138.1707ms)
✔ older configurations receive bounded wait defaults without changing the original object (1.0145ms)
✔ active stream survives sixty seconds; byte activity resets idle but cannot extend total deadline (0.9587ms)
✔ silent stream has its own accurate deadline; caller cancellation and cleanup preserve their cause (1.367ms)
✔ waiting for headers is bounded independently of stream-idle time (0.2339ms)
✔ wait settings persist, pin new runs and leave model pool, credentials and old run snapshots intact (1488.822ms)
✔ plain endpoint checks only TCP, records family coverage and never reads model credentials or writes routing/health/attempts (1762.2537ms)
✔ local DNS failure is separate from TCP/TLS and arbitrary input, stale revision and TLS overrides are rejected (1519.2769ms)
✔ refused TCP socket preserves its stage and cannot masquerade as a TLS or API response (1182.1552ms)
✔ TLS validation succeeds for a trusted owned certificate and rejects the same untrusted certificate (1878.0193ms)
✔ a TCP peer that never answers TLS is bounded, and cancellation waits for owned sockets to close (1945.7878ms)
✔ a stalled lookup is cancellable and the overall deadline bounds DNS without late socket work (944.6705ms)
✔ dual-stack fallback and explicit family results remain separate and never change global DNS defaults (1276.1888ms)
✔ proxy settings are reported as scope only; diagnostics do not contact the proxy or leak its credentials (1019.007ms)
✔ history is bounded, survives reopening, marks interrupted work and reports changed revisions without rewriting frozen targets (1672.547ms)
✔ HTTP API, host activity, configuration edits, model start and upgrade honor the same active diagnostic and stop operation (1473.8538ms)
✔ provider shutdown cancels diagnostics and permits explicit reuse after a failed host stop (1202.4673ms)
✔ unchanged proposals cannot hide repeated original-log reads or identical session inspections (9603.3029ms)
✔ reusing a verified proof never resets identical-state detection or creates a new proof (7800.9969ms)
✔ coverage merges adjacent pages, separates changed sources and never claims old Run coverage in a continuation (0.9799ms)
✔ paged inspection finds a corrupted old artifact outside the recent fifty and retains original evidence without model/tool execution (5525.5927ms)
✔ inspection rejects ambiguous/out-of-range pagination and reports unknown tool/task identity without resolving it (748.9288ms)
✔ original source locks all 60 byte-identical files including modules outside the nine-file prompt experiment (1030.7071ms)
✔ a new original executable or missing license cannot silently join the fixed trusted-source profile (1173.9303ms)
✔ original mode has a separate source identity and does not accept retry or privilege fields (837.552ms)
✔ actual original model input requires the complete postPrompt once in system context and cannot borrow a user or tool copy (0.2942ms)
✔ Windows directory permissions normalize overlapping grants without allowing an unrelated sibling (687.8097ms)
✔ durability broker rejects foreign authentication, outside paths and wrong file identity before real fsync (1046.1579ms)
✔ exact patches preserve BOM/CRLF, unicode and all unselected bytes and back up the original on apply (943.2982ms)
✔ several patches are matched against one original snapshot in any input order and keep LF (821.3406ms)
✔ mixed line endings require exact anchors and preserve untouched formatting (917.4361ms)
✔ ambiguous, overlapping, missing and malformed patch anchors never create a proposal or write (741.1001ms)
✔ stale original hashes and changes before apply preserve owner edits and mark conflicts (522.9519ms)
✔ no-op BOM patches/full content produce no change and replacements cannot split Unicode bytes (479.4044ms)
✔ real owned Vite starts only after approval, records OS ownership and live changes, and stops its tree (9190.7375ms)
✔ occupied ports and approval-time conflicts never take over an unrelated listener (6615.282ms)
✔ background preview final-log failure retains unknown evidence without an unhandled startup promise (9102.5318ms)
✔ stale source, tampered plans, wrong ownership and arbitrary scripts cannot dispatch a preview (7515.3829ms)
✔ restart retains original unknown preview, blocks mutation, reconciles absent identity without replay (9040.0643ms)
✔ an actual shell command keeps Node and UTF-8 output while omitting host-only credentials (824.5488ms)
✔ real npm tests and build do not inherit arbitrary host credentials or Node injection settings (2101.339ms)
✔ real Python tests and compilation keep their sealed inputs without host credentials (2592.3658ms)
✔ a process exiting between the Windows snapshot and owner query is absent only after a fresh query (1883.9117ms)
✔ owner observation errors remain errors while a process is present, replaced, or cannot be rechecked (1907.4118ms)
✔ cache controls require desktop authority, exact inputs and owned project identity while preserving task evidence (2749.4035ms)
✔ cache enablement and derived reuse survive a real host restart without becoming task recovery state (4272.612ms)
✔ persistent parsing is project/SHA/config bound and produces identical semantic indexes after restart (1397.6621ms)
warning: in the working copy of 'README.md', LF will be replaced by CRLF the next time Git touches it
warning: in the working copy of 'main.ts', LF will be replaced by CRLF the next time Git touches it
warning: in the working copy of 'value.ts', LF will be replaced by CRLF the next time Git touches it
✔ same-size restored-mtime changes, additions, renames and deletions cannot reuse stale entries or import targets (1281.0945ms)
Switched to a new branch 'other'
✔ branch and commit changes invalidate even byte-identical sources (2131.8191ms)
✔ corrupt, malformed and schema-invalid records fall back to fresh parsing (991.7545ms)
✔ disable and clear persist without touching sources, sessions, configuration or original artifacts (1247.8051ms)
✔ cache capacity bounds disk and memory by bytes and LRU project count (960.6788ms)
✔ incomplete scans do not load or save cached entries; unavailable files leave the current result scope (966.7218ms)
✔ project and cache junctions cannot expose or delete an outside directory (857.9659ms)
✔ concurrent file changes refuse a snapshot and clear during pending work cannot repopulate it (911.2711ms)
✔ file access mode changes invalidate parsed material even when bytes are unchanged (1089.3038ms)
✔ tracked directory replaced by a junction invalidates access scope and never uses its old cache (4031.9208ms)
✔ JS/TS/React AST indexes actual declarations, exports, imports and verified source lines (813.1067ms)
✔ default expressions, named default exports and declared functions retain their AST evidence (927.6374ms)
✔ incremental refresh invalidates changed/deleted symbols and recalculates reused import targets (1174.3305ms)
✔ Python candidates exclude comments and multiline strings and declare limited evidence (639.1904ms)
✔ indexes isolate project identity and root and evict bounded caches without sharing results (902.406ms)
✔ Git ignore, credential paths and links stay outside the index and no project files are executed (1532.3066ms)
✔ malformed, binary, non-UTF8 and bounded files expose partial coverage while good files remain searchable (817.2492ms)
✔ index query rejects invalid limits/types and does not imply reference or alias resolution (960.7187ms)
✔ read-only engineering inspection disables configured Git clean filters and preserves index bytes (2193.6ms)
✔ Python multiline declarations retain candidate scope without interpreting expressions as declarations (888.7361ms)
✔ TypeScript module candidates support import-equals and MTS/CTS while opaque imports stay explicit (994.5987ms)
✔ actual HTTP requests reuse only this project source-checked memory as redacted data and preserve sent evidence after archive/restart (2956.4394ms)
✔ actual HTTP request excludes archived, timed-out and foreign-project memories (4527.0282ms)
✔ actual next HTTP request removes expired memory after a tool observes a source change without replaying that tool (4242.4636ms)
✔ incomplete inventory and excluded-file SHA changes cannot enter actual HTTP memory context or inspect_session snapshots (4178.8482ms)
✔ corrupt original artifact expires its memory and cannot reach an HTTP model request (2917.7688ms)
✔ memory JSON is bounded in escaped UTF-8 bytes and entry count, skips oversized entries and retains complete multilingual values (4587.7163ms)
✔ optional memory yields to the complete goal and original code evidence under the whole-request budget (1580.5525ms)
✔ changed sealed memory artifacts or binding fields fail verification while the original request remains immutable (2614.8236ms)
✔ a malformed project declaration omits optional memory while the repair Task still reaches the owned HTTP service (2153.756ms)
✔ an expiration persistence failure prevents dispatch and is not hidden as an optional scan failure (1485.6608ms)
✔ project memories persist across restart, keep source provenance, and are visible across tasks (1616.1092ms)
✔ source changes expire current project memories and keep the original evidence (1388.4775ms)
✔ incomplete source scans mark current memories unverified without expiring or reusing them (1461.7013ms)
✔ explicit file provenance is rechecked even when the inventory excludes that path (860.9776ms)
✔ execution snapshots do not expose project memories before their source scope is validated (1221.4074ms)
✔ project memory records stay isolated and archive requires the owning session (1372.5332ms)
✔ project memory HTTP API exposes read-only source status and explicit writes (3497.7577ms)
✔ two Purge source versions lock nine original inputs without executing JavaScript or postinstall and remain disabled by default (1131.7475ms)
✔ incomplete, unsupported, relative and oversized sources cannot enable the experiment (1123.1367ms)
✔ invalid source fingerprints and invented privilege or retry fields fail before any parent Task is created (2443.8844ms)
✔ source edits between detection and plan are rejected without creating a parent Task (2937.1555ms)
✔ source edits after sealing cannot launch a child and do not reset its cumulative budget (9709.4042ms)
✔ literal experiment text is sealed into the parent, worker and each new Run while actual host approvals and saved results remain required (29153.1816ms)
✔ source changes after a model response stop before any native tool is accepted, retaining the actual usage once (11726.2714ms)
✔ source changes while approval is pending invalidate the old permission before an engineering check starts (16104.0071ms)
✔ altered text, interpolation, section order and policy flags cannot reuse a sealed experiment (2187.0368ms)
✔ the observed SDK system-message layout and the explicit system field both require exactly one effective section, never a user or tool copy (2038.7317ms)
✔ a missing or duplicated experimental section in the actual input is rejected before any model dispatch (21796.6302ms)
✔ experimental prompt overhead still participates in the original complete-input budget guard (11696.1214ms)
✔ source detection HTTP remains local-authenticated and read-only with no configured model requests (1938.8074ms)
✔ Python HTTP plans keep npm defaults, require owned approval and preserve actual results across restart (10335.1039ms)
✔ an isolated three-module Python Task repairs, checks and delivers a source-bound Git report without extra model calls (40211.2175ms)
✔ Python static inspection does not import project code or probe a project interpreter before approval (981.8976ms)
✔ a real three-module Python failure, reviewed repairs, tests and compilation retain exact structured evidence (4984.1481ms)
✔ unittest counts repeated execution ordinals and subtest skips rather than deduplicating display names (3099.1939ms)
✔ unittest empty, skipped, expected failure and class initialization errors never verify a task (8386.1957ms)
✔ stdout claims cannot replace actual unittest results and an early exit cannot forge a successful collector (3220.9802ms)
✔ compilation reads actual Python source bytes without executing top-level project code (3108.3279ms)
✔ Node and Python mixed projects retain the existing npm default and require an explicit Python selection (2711.7725ms)
✔ changed project interpreters and forged Python plan bindings execute no check (1781.2467ms)
✔ malformed TOML and invalid Python test paths are reported without expanding npm check scope (1461.0125ms)
✔ stopping a real Python check terminates its child and keeps the completed result as cancelled (1981.0244ms)
✔ real pytest pass, failure, skip, collection errors and teardown errors use collected case outcomes (15074.5171ms)
✔ a real project virtual environment is detected statically and its actual version is recorded only during the approved check (8734.6176ms)
✔ a missing interpreter and repeated pytest item objects cannot create a false verification pass (3006.8506ms)
✔ a mixed turn with new bytes and three repeated modules advances, while repeat-only and independent no-op guards still pause (1.0463ms)
✔ runtime sends the next real HTTP request after a W-shaped mixed-progress turn instead of stopping on its trailing repeats (6100.0632ms)
✔ continuation reads advance through historical pages without a false pause while repeated current-Run pages still pause (1.1701ms)
✔ a new tool page replaces old latest-feedback space and still obeys its escaped protocol budget (0.4189ms)
✔ budgeted artifact pages count escaped protocol bytes and reconstruct original multilingual JSON without losing characters (8.4146ms)
✔ known artifact coverage detects semantic repeats despite changing limits, and advancing pages or direct evidence restore progress (0.5137ms)
✔ actual large artifact reads are present in the next bounded HTTP request and next_offset progresses without redoing tools (5004.1754ms)
✔ a real repeated artifact loop pauses the same Task after recorded results, survives reload and only resumes explicitly (8702.8304ms)
✔ actual HTTP continuation after reload sends distinct historically covered pages without false no_progress or replay (9732.7711ms)
✔ continuations distinguish unread source, later original line ranges and already returned result wrappers (638.2924ms)
✔ pure-file batch returns unused small-page space while mixed and historical allocation stays conservative (0.3145ms)
✔ v3 pools mixed file/log reads, serves small sources first and leaves historical/mutating batches unchanged (0.2646ms)
✔ v3 W-shaped mixed reading returns complete small modules and substantially larger first test pages in the next bounded request (5034.5152ms)
✔ observed upstream high input pauses before a new request, preserves reported usage and distinguishes the limiting ledger (967.4406ms)
✔ chat-completions: complete actual Unicode/CRLF pages and small source survive the next bounded wire request (4275.5664ms)
✔ responses: complete actual Unicode/CRLF pages and small source survive the next bounded wire request (5232.4504ms)
✔ anthropic-messages: complete actual Unicode/CRLF pages and small source survive the next bounded wire request (4642.2975ms)
✔ brand and custom-compatible profiles send native reasoning controls per protocol without changing output budgets (1.8902ms)
✔ opaque Chat reasoning returns only to its original endpoint, model, connection and credential revision; bytes count against the actual request (0.6709ms)
✔ saved profile changes preserve credential references and old frozen Run settings, including the legacy connection after restart (1520.2862ms)
✔ DeepSeek JSON reasoning survives real HTTP tool rounds and restart without entering visible assistant text (6086.4515ms)
✔ DeepSeek SSE reasoning survives real HTTP tool rounds and restart without entering visible assistant text (5237.8316ms)
✔ malformed or post-terminal Chat reasoning is rejected with the raw stream retained (1.4438ms)
✔ connection import synchronizes profile, native fields and declared abilities without editing the source or manufacturing verification (607.8496ms)
✔ compiled HTTP persists profile and reasoning choices for every protocol and rejects incompatible/stale edits atomically (3128.2343ms)
✔ an installed app outside the source drive resolves registered SDK and Purge through actual input verification (1093.294ms)
✔ standalone registration resolves the upstream metadata directory and missing upstream falls back to the other registered source (1017.6648ms)
✔ registered hints keep explicit roots, environment and existing app-relative candidates in their original priority (960.6104ms)
✔ unknown IDs, duplicate registered IDs, wrong versions and relative or inconsistent metadata are not lookup candidates (771.2337ms)
✔ oversized, malformed, invalid UTF-8 and non-file registry inputs cannot silently enable a source (744.3287ms)
✔ a linked registry directory is rejected without reading or deleting its external source (673.0575ms)
✔ matching research declarations do not replace actual installed version and critical-source checks (880.6931ms)
✔ runtime guides map archived evidence while all original reports and successful/failed source bytes remain independently verifiable (978.4858ms)
✔ missing required research reports fail packaging rather than creating a successful placeholder (1071.8505ms)
✔ full verification reports only verified bytes, complete declared scope and actual directory inventory; changed/deleted files still fail (868.7285ms)
✔ preflight and activation continue one monotonic operation bound to installer hash; terminal replay and changed origin are rejected (9335.7999ms)
✔ preflight failures and fixed-candidate mismatches retain the actual reason without publishing a pointer (4997.6715ms)
✔ cancel finishes only the original ready preparation, preserves data and pointer and refuses replay (7187.9481ms)
✔ the installation lock rejects concurrent same-id operations before they can overwrite original progress (4314.7812ms)
✔ original post-publication failure is linked to its rollback receipt and every original byte survives (6194.8463ms)
✔ a candidate mutation at the publication boundary still fails even after an earlier successful full validation (9430.3199ms)
✔ public progress marks a lost process unknown without editing its original record or claiming completion (6336.5333ms)
✔ throwing console/UI progress observers cannot decide the installation result (6130.5743ms)
✔ a progress-only write failure after committed publication returns the authoritative receipt and a warning (5351.053ms)
✔ browser and MCP approvals use actual tool categories; mobile exposes only category while keeping original fingerprint (955.7861ms)
✔ the compiled runtime version constant tracks package.json so a release bump cannot ship a stale served version (10.3873ms)
✔ context APIs retain exact request identity, paginate archived groups and reject foreign sessions/runs (3296.6159ms)
✔ delivery reports verify request body, sealed plan and original archived bytes, and show corruption as needing verification (1780.8434ms)
✔ native seed framing reserves capacity before optional history without changing the selected window (811.2941ms)
✔ direct code evidence remains selected while recent historical-index browsing is archived within the complete request budget (2066.7269ms)
✔ one long user turn becomes bounded complete groups with exact UTF-8 history and recent JSON arguments (3476.2223ms)
✔ tool definitions, exact system bytes, output reserve and remaining Run/Task budget are accounted before requests (774.2937ms)
✔ oversized required user instructions pause intact and can be explicitly resumed after increasing the window (2592.0088ms)
✔ all task instructions, constraints and acceptance remain in the runtime system instead of an eight-item or byte prefix (1638.4231ms)
✔ a newest manageable multi-tool group stays exact even beyond the working target; oversized earlier groups are archived (727.9379ms)
✔ invalid orphan result is rejected before archive or network preparation (535.4588ms)
✔ invalid missing result is rejected before archive or network preparation (816.051ms)
✔ invalid wrong result is rejected before archive or network preparation (1108.8982ms)
✔ invalid duplicate call is rejected before archive or network preparation (568.5246ms)
✔ a group larger than the artifact limit uses verified Unicode-safe chunks without replacing tool JSON (2155.7496ms)
✔ changed or missing archived evidence stops reuse and is never silently recreated (1179.7545ms)
✔ transaction failure prevents publishing a plan and any corresponding model request (1929.357ms)
✔ the Provider rejects body mutation after a sealed plan before calling HTTP (1416.7889ms)
✔ HTTP failure and explicit same-Task resume reuse exact completed results and request hashes without repeating a command (11582.9088ms)
✔ a real process exit after plan persistence marks interruption and never issues an automatic model request (1604.2216ms)
✔ local draft preserves exact original ranges and duplicate source instances without inferring outputs or accepting facts (1481.8978ms)
✔ review keeps unknown blocking, references distinct from output and boundaries distinct from compliance evidence (1319.9485ms)
✔ draft adoption must account for every original range and does not fill paths from a reference (1427.0002ms)
✔ source and requirement changes, other Tasks and tampered draft bytes are rejected without losing originals (2450.406ms)
✔ partial explicit source review leaves the other source unreviewed while old file evidence remains valid (2277.1343ms)
✔ bounded draft does not claim full source coverage and cannot acknowledge an omitted remainder (917.1287ms)
✔ draft HTTP preserves its original artifact across restart and rejects stale scope and source versions (5552.2514ms)
✔ requirements HTTP binds selection, enforces optimistic revisions, restores original proofs and leaves provider history untouched (5917.9725ms)
✔ research evidence binds actual original bytes; model completion is not user acceptance (1640.0851ms)
✔ same text with distinct user source instances stays distinct; an appended requirement only invalidates its own revision (1808.8808ms)
✔ foreign project, old Task and fabricated evidence strings are rejected with preserved failed proofs (4089.785ms)
✔ source changes invalidate prior proofs and cannot certify a read that was already stale at binding (3246.5198ms)
✔ missing or tampered artifacts never become evidence and original failures remain after recovery (2508.22ms)
✔ checks omitted by explicit agreement are reported as not executed and never passed (1385.4657ms)
✔ user confirmation stays separate from engineering evidence, is bounded to current inputs and cannot be silently converted (1644.67ms)
✔ existing multi-file results keep provenance and invalidate only evidence whose files changed (2129.2275ms)
✔ unrelated additions preserve the checked file facts but remain visible as unreviewed user sources (2289.7043ms)
✔ actual Git preview and publication recheck changed requirement bindings without touching HEAD or the owner index (8908.4265ms)
✔ no-test file application and manual completion keep pending user requirements for explicit review (2601.3185ms)
✔ revision races, forged source ranges, evidence fields and deletion cannot overwrite sealed requirements (1684.5076ms)
✔ requirement binding prevents completion and Git acceptance after a review changes; legacy tasks stay explicit (4379.6524ms)
✔ deep Windows data directories keep REA output within the external path allowance and retain owned evidence (3526.6472ms)
✔ user research registration creates an owned Task without model requests and seals the exact artifact identity (1528.6136ms)
✔ complete research binds original Evidence and Bundle, pages exact fields and reuses verified results after reload (3403.8573ms)
✔ artifact changes before analysis refuse dispatch and changes during analysis preserve unverified original responses (4162.3815ms)
✔ owned cancellation closes the connection, concurrent and automatic retries are refused, explicit new analysis preserves history (2967.6732ms)
✔ a dispatched transport timeout stays unknown and reload never launches or kills a saved PID (2577.8037ms)
✔ corrupt Evidence parts and changed binding records are refused instead of silently re-analysed (2558.2557ms)
✔ research documents over the single-artifact limit preserve exact Unicode bytes with owned verifiable parts (1476.5327ms)
✔ research tools are only sent for registered Task artifacts and actual HTTP uses the sealed tool contract without replay (8491.1532ms)
✔ file identity rejects directory and invalid formats without running a target (1225.7628ms)
✔ a stop from the event loop interrupts immediate model replies before later tools (2408.6032ms)
✔ a stop after the first tool preserves a complete interrupted multi-call protocol group (4089.2981ms)
✔ the compatible HTTP loop preserves call IDs, validates errors and waits before executing commands (9456.0311ms)
✔ denial prevents command execution and produces a paired denied result (3525.2713ms)
✔ stopping during approval interrupts pending calls without executing or replaying them (3139.1906ms)
✔ stopping a running fixture command terminates its own process tree and retains uncertain state (2952.2852ms)
✔ command output decodes split UTF-8 and excludes the workbench key from child environment (1493.4406ms)
✔ model-list testing uses a tentative key without changing the saved connection and redacts service errors (5.9852ms)
✔ truncated or repeated model calls fail before execution; tool input validation rejects unknown capabilities (5335.4355ms)
✔ request projection keeps exact tool pairs and complete recent user instructions while preserving disk history (3371.94ms)
✔ local skill revisions require explicit enable, retain original text, survive restart and stay within the selected project (1540.47ms)
✔ registration rejects non-SKILL paths, oversized/non-UTF8 files, duplicate sources and wrong field shapes without running project code (1447.2893ms)
✔ actual HTTP models receive only enabled current-project metadata, then a redacted owned skill page with complete tool pairing (4748.3306ms)
✔ source edits after a catalog request prevent the read and appear as changed on the next request without replay (4144.5243ms)
✔ disabled and foreign skills cannot be read by guessing their IDs from another project (5013.6676ms)
✔ skill catalog takes remaining escaped JSON budget and cannot evict the user goal or current code (1522.0259ms)
✔ multilingual skill body pages fit actual tool JSON bytes and every page uses the same immutable artifact (6475.4309ms)
✔ registry failures roll back enable and refresh state; corrupted snapshots block requests before network (2200.5913ms)
✔ saved skill catalogs retain SHA-bound historical evidence and reject corrupt catalog blobs (3480.4403ms)
✔ HTTP skill administration enforces project scope, revision checks and idle mutation without model calls (4647.2143ms)
✔ JSON state keeps an atomic backup and preserves corrupt primary files (943.6976ms)
✔ tool-pair recovery fills missing results and drops orphan or duplicate results (0.324ms)
✔ restart pauses unfinished work, retains proposals and never replays a command (1030.5307ms)
✔ public snapshots carry increasing revisions without the private provider transcript or config key (853.8269ms)
✔ a checkpoint seals all task requirements and original results without a model request and explicit continuation keeps the same Task without replay (7125.501ms)
✔ checkpoint sources, requirements, plan and original tool changes block continuation and leave original checkpoint bytes intact (4743.3261ms)
✔ unknown tool effects and pending proposals stay blocked in recovery packages rather than being changed into completed work (5398.008ms)
✔ checkpoint and referenced result corruption are rejected before dispatch and historical blobs are never rebuilt (5125.5647ms)
✔ checkpoint ownership, exact confirmation and concurrent operations reject foreign or stale resumes (4621.9263ms)
✔ Git identity changes and altered plan proofs cannot be accepted as the sealed recovery state (5833.2066ms)
✔ checkpoint records survive SQLite reopen and do not trigger automatic model or tool replay (5128.2942ms)
✔ incomplete scans and in-progress project activity refuse checkpoint creation while private credentials are redacted from saved requirements (4735.8724ms)
✔ oversized recovery evidence is refused before reading or publishing the checkpoint (5076.8485ms)
✔ compiled checkpoint HTTP APIs bind current Task/Run, isolate sessions and only resume on an explicit matching confirmation (11570.4235ms)
✔ actual process exit before checkpoint commit retains only durable records without replay (4458.0397ms)
✔ actual process exit after checkpoint commit retains only durable records without replay (5204.9014ms)
✔ a real two-reply failure replay reaches request three at 16K with complete goal, original sources and sealed tool evidence (12514.4387ms)
✔ request deduplication retains long original intent, extra requirements, constraints and acceptance without changing the stored contract (3529.6022ms)
✔ only the routine context avoids plan echoes; diagnostic API and requests without an included plan keep original observations (1146.9332ms)
✔ operational questions are diagnostic while explicit repair requests remain execution instructions (1.1111ms)
✔ an explicit prohibition on running engineering checks is not interpreted as a requirement to run them (1.7154ms)
✔ file delivery requires a positive instruction, not saved evidence or a prohibited write (0.2935ms)
✔ an unsubstantiated text claim cannot complete a requested file delivery (2323.8192ms)
✔ HTML without tests ends at reviewed file delivery and never offers automatic rechecking (5062.6919ms)
✔ diagnostic answers retain the unfinished task, limits and check result, and enforce read-only tools (6869.292ms)
✔ review continuation is a system action and adds no testing demand or fake user instruction (4586.7929ms)
✔ a provider cannot execute tests against an explicit no-test task even if it ignores advertised tools (3245.8472ms)
✔ a budget pause becomes visible only after its input snapshot can be sealed with the original Run (2320.0342ms)
✔ alternating duplicate file reads and listing files still pauses without treating new pages as duplicates (25209.5386ms)
✔ task-plan HTTP mutations bind current Task/Run and retain original evidence through real host restart (6250.3988ms)
✔ concurrent stale HTTP plan revisions cannot replace conditions or attach proofs to a different plan (3724.2952ms)
✔ a real escaped long-file tool message keeps its proof identity within a tight request budget (983.7189ms)
✔ task plans reject fabricated completion, invalid scopes and cyclic or missing dependencies (0.5236ms)
✔ sealed plans retain goal, contract hash and original steps through SQLite reopen without executing tools (1834.0928ms)
✔ proofs require original owned reads, honor dependencies and reject altered invocation arguments and bytes (1241.0794ms)
✔ new user requirements invalidate current proofs and reseal the same original operations in a new revision (2151.0003ms)
✔ applied change-set evidence binds reviewed current files and invalidates dependent steps after outside edits (2253.8758ms)
✔ actual npm checks are verified by original counted logs and current source, with failure history retained (9667.7314ms)
✔ real desktop and mobile browser proofs bind required actions and PNGs, rejecting source changes, a later failed plan and changed builds (13800.9971ms)
✔ Git delivery blocks incomplete plans and changed preview bindings, preserving HEAD and owner index before a fresh verified commit (15822.6016ms)
✔ the runtime supplies owned read IDs and completes a plan only after original step evidence verification (4512.9519ms)
✔ premature model completion pauses the original Task and resume preserves the existing plan without replay (6590.3169ms)
✔ plan operations participate in project activity and orderly shutdown (1371.3689ms)
✔ actual process exit before plan save restores only committed plan evidence without replay (1603.7939ms)
✔ actual process exit after plan save restores only committed plan evidence without replay (1584.7824ms)
✔ actual process exit before proof save restores only committed plan evidence without replay (1900.7788ms)
✔ actual process exit after proof save restores only committed plan evidence without replay (1866.6275ms)
✔ the latest oversized session result reaches the next request together with already read source modules (3673.4841ms)
✔ every call in a large latest group keeps its actual outcome, including denied and unknown results (1801.723ms)
✔ a small complete latest tool group is sent once and needs no duplicate feedback view (1469.5469ms)
✔ real HTTP repeated state inspection pauses after three identical engineering outcomes and preserves tool pairing (5631.9414ms)
✔ fixed normal exploration, pagination, SHA changes and repeated ranges are external observations, not a score (1.3531ms)
✔ read-only research is valid work; model summaries and unchanged timestamped listings do not become engineering evidence (0.4652ms)
✔ new permissions, network and broken tools are failure evidence; repeated invalid inputs remain candidates (0.3743ms)
✔ waiting for original long checks, review and approvals stays waiting across repeated observations (0.4983ms)
✔ baseline facts, real repairs, current-source check results and requirement evidence have distinct growth (0.2239ms)
✔ bounded memory keeps unknown instead of claiming an unretained range is repeated; Task/Run instances cannot mix (12.4364ms)
✔ actual runtime records growing pages then a repeated page without changing model rounds, stop rules or original tool pairs (5917.3747ms)
✔ actual invalid tool calls repeat under the original step budget; observation never adds an early stop (4972.3245ms)
✔ actual command approval is observed as waiting and denial stays a permission failure (4423.2216ms)
✔ network errors remain model failures with shadow failure evidence, never a fabricated no-progress cause (4218.6167ms)
✔ observer receipt failure cannot change tool results, model budget or completion (2441.238ms)
✔ a later receipt recovers after an observation write failure without misaligning retained source receipts (3765.3736ms)
✔ real applied files, counted checks and current requirement proofs grow after an existing-facts baseline (5727.5811ms)
✔ receipt retention stays aligned at the cap and a foreign step cannot enter the Run (4475.5739ms)
✔ an actual approved long check records its owned running job as waiting before the original counted result (8749.7947ms)
✔ a live SQLite host with the actual 1.9 artifact object upgrades, keeps its original blobs and verifies real browser PNG bytes (6443.8278ms)
✔ failed upgrade restores the original artifact object, provider and request handler (2292.8264ms)
✔ claiming the same runtime owner twice preserves its original metadata and every database row (1087.2528ms)
✔ sessions without controlled-engine history retain their prior export shape and never expose another session record (961.1761ms)
✔ live workers, controlled drivers, or MCP operations refuse an upgrade before changing handlers, managers or database rows (1855.9018ms)
✔ an idle upgrade loads the newly introduced engine record map, resets obsolete managers and preserves original rows, sessions and provider objects (910.1803ms)
✔ an installation failure restores old managers, original record maps, handlers and all existing database rows (2257.2965ms)
✔ the owned controlled driver remains an active session and project even after outer task activity has ended (817.85ms)
✔ release and upgrade help or invalid arguments have no file or network setup side effects in an independent directory (1582.3089ms)
✔ hot upgrade adopts a retained virtual data root without reopening SQLite or changing rows, Provider or routing (1917.5231ms)
✔ failed adoption restores the original virtual data-root name and original open database (1159.5456ms)
✔ hot upgrade refreshes retained model and importer methods while preserving credential and routing objects (910.4976ms)
✔ hot upgrade refuses active probes and restores retained model methods if installation fails (751.3238ms)
✔ the upgrade compatibility gate includes the served current version and the previous formal patch (0.4427ms)
✔ a directly started compiled host upgrades twice with the same PID, sessions, applied bytes and volatile provider (4428.8005ms)
✔ the running 0.1.0 process upgrades in place and retains its project, session, files and volatile model connection (9895.0835ms)
✔ HTTP worker identities require a current sealed plan, reject overrides and close unused plans without touching source (21248.1469ms)
✔ HTTP unknown recovery only accepts the original evidence SHA and preserves changed source for explicit retention (29608.7575ms)
✔ verified isolated outputs form a new parent batch, rerun real checks and commit only reviewed source (80311.2369ms)
✔ overlapping worker paths are explicit conflicts even if each independent task checks successfully (50957.4644ms)
✔ source edits after isolated work preserve the owner file and invalidate collection without proposing a merge (37715.0531ms)
✔ changed child bytes after parent preview block the entire batch before any parent write (51368.6365ms)
✔ corrupt original check logs cannot become collected outputs or a parent review (41047.3368ms)
✔ stopping a parent accounts late responses and executes none of their proposed tools (23553.32ms)
✔ unreported interrupted requests reserve estimated usage and the total deadline cannot reset on resume (20809.5409ms)
✔ a deadline before worker request dispatch records no invented usage and cannot reset on resume (24568.3948ms)
✔ a response exceeding the cumulative token allocation stops before registering or executing its tools (19303.1857ms)
✔ cumulative tool allocation rejects the entire returned batch before any tool side effect (15903.1255ms)
✔ single response 16 tool ceiling rejects the entire returned batch before any tool side effect (16848.1117ms)
✔ consumed request steps remain consumed across explicit worker resumes (19912.7209ms)
✔ a saved completed response with missing usage is conservatively accounted exactly once during recovery (14944.3871ms)
✔ an unexecuted second workspace plan resumes its original identity without recreating the first checkout (17411.6006ms)
✔ a verified interrupted parent proposal resumes sealing without duplicating or replacing its original change (61228.0032ms)
✔ recovery attaches a sealed parent batch saved before its group reference without creating another batch (52979.8415ms)
✔ explicitly retaining an unexecuted parent merge cancels its batch and frees the next parent goal (51099.7661ms)
✔ unknown collaboration blocks ordinary proposals but permits only original batch recovery and rollback (58109.3955ms)
✔ after a restart the batch application lazily reinstates the worker result guard before any parent write (47282.4957ms)
✔ unknown original tool inputs must match before acknowledgement, and reviewed checkpoint IDs preserve the unknown fact (17369.4979ms)
✔ a parent stop observes and waits for a manually approved child check and its real descendant process (17270.5913ms)
✔ a real host crash preserves unknown checks, refuses a still living owned PID and never replays a model or tool (31457.2019ms)
✔ a persisted child initialization resumes the original control run after a failure before its worker record (17843.9962ms)
✔ failed request intent persistence never dispatches a worker request or changes source (16344.4068ms)
✔ failed usage persistence rolls back the uncommitted usage and conservatively reserves one unknown request without tools (17647.6904ms)
✔ a child file change after observation prevents acknowledgement even when the tool inputs did not change (16384.3273ms)
✔ worker scopes and allocated budgets reject aliases, extra permissions and aggregate overflow (574.2788ms)
✔ two sequential Git workspaces freeze a new parent goal, separate owned tasks and preserve all source bytes (14782.5038ms)
✔ real isolated changes, approvals and npm checks produce original owned outputs, not model completion claims (41917.8906ms)
✔ forged worker tools cannot execute terminal or Git even when absent from the advertised tool list (15412.7344ms)
✔ a completion-only model pauses each worker without fabricated changes or checks (18687.6897ms)
✔ restart retains frozen scope and marks interrupted parent and workers unknown without sending requests (15818.2146ms)
✔ all actually returned same-version fragments reach the bounded request, including overlap and Unicode, with independent original provenance (4653.5615ms)
✔ unread gaps and different range origins remain separate; partial sending names the precise missing already-observed range (1103.1285ms)
✔ same-SHA conflicting overlaps and invalid coordinates fail closed, while a new SHA never inherits old fragments (1259.0006ms)
✔ versioned old views keep the original single-page semantics and verification (1033.2936ms)
✔ small code and actual failed command stay visible when their original multi-tool group cannot fit (2113.0184ms)
✔ mixed group retains six distinct small file observations with exact multilingual text and bounded request bytes (2631.3088ms)
✔ working observations are bound to the current Task and stop treating a pre-application file revision as current evidence (2061.517ms)
✔ a sealed observation rejects changed or missing original result artifacts instead of recreating evidence (1449.4945ms)
✔ a real passed-check observation is excluded after a source edit invalidates the check, while its historical sent evidence remains valid (6541.0296ms)
✔ budgeted file pages reconstruct UTF-8 text including long single lines and surrogate boundaries (6.3831ms)
✔ actual HTTP multi-tool reads survive an oversized runtime snapshot and the sealed evidence survives reopening (14478.3767ms)
✔ actual HTTP file pagination advances over a long file with each returned page visible in the next request (9557.1498ms)
✔ later mixed history pages cannot displace the actual module sources and complete requirements (6447.9309ms)
✔ a later narrow page of the same file revision cannot evict the wider actual observation (2841.66ms)
✔ six small source files stay complete across metadata turns even when mandatory evidence exceeds the soft target (3715.8846ms)
✔ historical read-artifact navigation points to the original artifact and next page, never its result wrapper (1357.5555ms)
✔ a 1.6.1 full-provenance sent view remains verifiable after introducing the compact view (1360.3986ms)
✔ actual HTTP requests keep four observed modules after a later large history read and preserve all source bytes (14202.2242ms)
✔ workspace HTTP ownership, reviewed creation and reopen survive restart without changing source files (11332.4665ms)
✔ isolated model Task applies a batch, executes real checks and commits at the Git review boundary with a source-bound report (37270.8929ms)
warning: in the working copy of 'src/空 格.mjs', LF will be replaced by CRLF the next time Git touches it
✔ worktree checkout disables hooks, fsmonitor and all filter command variants while honoring built-in EOL conversion (7038.9384ms)
✔ an ignored external file at checkout is observed and never accepted as pristine creation (6387.1036ms)
✔ outside content hidden by assume-unchanged cannot pass workspace verification and blocks further source execution (5268.1133ms)
✔ workspace preparation refuses known shared-repository activity and preserves archived child sessions (5602.1469ms)
✔ compact checkout identities create in deep Unicode source directories while retaining the human branch name (6864.2476ms)
✔ Windows Git metadata path limits reject creation before publishing a branch or changing source bytes (2878.1791ms)
✔ a pending source snapshot participates in shutdown and completes before storage can close (2328.9519ms)
✔ source configuration values stay private and a configuration change invalidates a saved review (3041.7886ms)
✔ workspace review rejects excluded dependency before any checkout (2853.427ms)
✔ workspace review rejects large committed blob before any checkout (2788.2964ms)
✔ workspace review rejects case collision before any checkout (2415.1146ms)
✔ workspace review rejects committed symbolic link before any checkout (2323.5716ms)
✔ workspace review rejects external config include before any checkout (1861.0759ms)
✔ actual process exit at workspace intent preserves source and reconciles without creation replay (5153.6282ms)
✔ actual process exit at workspace git_started preserves source and reconciles without creation replay (7972.5421ms)
✔ actual process exit at workspace git_finished preserves source and reconciles without creation replay (8188.6757ms)
✔ actual process exit at workspace verifying preserves source and reconciles without creation replay (7805.2006ms)
✔ actual process exit at workspace registering preserves source and reconciles without creation replay (8480.9786ms)
✔ actual process exit at workspace project_registered preserves source and reconciles without creation replay (10012.194ms)
✔ actual process exit at workspace session_registered preserves source and reconciles without creation replay (9257.6465ms)
✔ actual process exit at workspace ready preserves source and reconciles without creation replay (7325.4972ms)
✔ outside changes at project registration remain unknown and preserve the already registered child identity (8158.9613ms)
✔ managed worktree preserves source index and dirty files, opens a distinct project and binds Run identity (5794.2318ms)
✔ stale review, branch/path collisions and forged ownership do not run worktree creation (3657.4985ms)
✔ workspace project verification is read-only and rejects physical Git identity changes (4987.8826ms)
ℹ tests 926
ℹ suites 0
ℹ pass 926
ℹ fail 0
ℹ cancelled 0
ℹ skipped 0
ℹ todo 0
ℹ duration_ms 2143077.8077
@@ -0,0 +1,994 @@
> local-agent-workbench@1.25.3 test
> node --test --test-concurrency=2 tests/*.test.mjs
✔ Messages builds native system/tool history, bounds images, and seals a distinct mixed-pool variant (7.1432ms)
✔ Messages URL and native parameter contracts preserve other protocol paths and hard output budget (0.5816ms)
✔ native usage includes both cache buckets once, retains raw and rejects invalid counters (0.259ms)
✔ real HTTP Messages JSON executes one tool, retains opaque history, actual URL and cache accounting (3682.6337ms)
✔ real HTTP Messages SSE executes one tool, retains opaque history, actual URL and cache accounting (2824.3553ms)
✔ Messages api-key authentication is explicit; mismatched model remains blocked with original usage (5.9668ms)
✔ Messages bearer authentication is explicit; mismatched model remains blocked with original usage (5.9472ms)
✔ Messages redirect is not followed, either credential stays at the original server (4.2771ms)
✔ strict native SSE reads the nested start model and retains mismatch source without retries (3.016ms)
✔ a closed native tool block cannot execute before message_stop, then executes exactly once (2842.227ms)
✔ mixed Chat/Messages pool seals both variants and preserves one completed tool on failover (4069.0662ms)
✔ native JSON rejects incomplete stops, duplicate tools and hosted content (0.2749ms)
✔ Messages rejects empty, whitespace and thinking-only answers without rejecting valid text or tool calls (0.2124ms)
✔ empty Messages JSON fails once with original response and 6692/0 usage, before verification or tools (2488.8824ms)
✔ empty Messages SSE fails once with original response and 6692/0 usage, before verification or tools (1979.4285ms)
▶ strict SSE validates all blocks, terminal, event names, UTF8 and tool JSON before dispatch
✔ reject native stream 0 (1.1755ms)
✔ reject native stream 1 (0.2717ms)
✔ reject native stream 2 (0.1914ms)
✔ reject native stream 3 (0.1834ms)
✔ reject native stream 4 (0.1618ms)
✔ reject native stream 5 (0.1681ms)
✔ reject native stream 6 (0.1423ms)
✔ reject native stream 7 (0.1248ms)
✔ reject native stream 8 (0.2087ms)
✔ strict SSE validates all blocks, terminal, event names, UTF8 and tool JSON before dispatch (3.3056ms)
✔ upstream SSE error retains original type/request ID and initial usage, never retries (4.4715ms)
✔ stop cancels the owned stream and retains partial consumption without running tool or replaying (61.662ms)
✔ saved Messages import preserves auth choice, rejects endpoint duplication, and keeps source file untouched (704.5321ms)
✔ large source receipts use cursor-preserving invalidation without filling the event stream or changing the session (5.1678ms)
✔ small event updates retain the current session and actual cursor (0.1121ms)
✔ the bundled local service enforces host, origin, token and JSON object contracts (2305.3673ms)
✔ an in-flight connection probe prevents configuration changes and another probe until its actual result returns (2976.5136ms)
✔ API review applies verified content, detects conflicts, exports sessions and keeps keys in memory (4778.8466ms)
✔ a real bundle restart preserves proposals, interrupts old approvals and requests fresh consent on resume (12248.1301ms)
✔ runtime diagnostics and model-response validation use actual records and preserve the saved connection (2154.6808ms)
✔ persistent event replay is read-only and keeps run/artifact identity across reconnects (9740.0732ms)
✔ real browser HTTP plans, original field pages and PNG bytes preserve ownership and survive host restart (8276.9923ms)
✔ self harness browser tools require approval and feed verified actual page fields back to the original Task (15480.4049ms)
✔ browser host recovery retains unknown root outcome, blocks replay, and requires an explicit owned acknowledgement (4197.1913ms)
✔ late HTTP stop of a finished browser job cannot interrupt a later job in the same Task (7927.1675ms)
✔ actual failed page blocks source-passed delivery; a new passing page permits a bound commit and preserves both original records (20306.1265ms)
✔ a build-only change after Git review invalidates the actual browser evidence and leaves HEAD and the protected index intact (10195.3384ms)
✔ actual DOM controls include validated unique CSS selectors across repeated parent buttons and escaped labels (7915.3601ms)
✔ browser field coverage distinguishes sections and Run windows while retaining original source identity (0.4716ms)
✔ repeated original browser pages pause a live Task without re-opening the browser or breaking tool pairing (5951.2192ms)
✔ browser plan rejects outside builds, invalid interaction scope and unbounded steps before execution (0.8996ms)
✔ real owned Edge interactions, assertions, raw PNG and stopped process survive exact result reuse and host reopen (3748.0438ms)
✔ failed real page assertion preserves actual value and forbids silent retry of the original plan (8470.2653ms)
✔ build/source mutation and sealed step tampering reject before browser dispatch; earlier task cannot execute plan (2095.9601ms)
✔ outside network never reaches another local service, and original JavaScript/console failures remain failures (3084.0855ms)
✔ stopping a real in-flight browser preserves finished steps, closes its owned root and never replays (2958.4177ms)
✔ 原PNG损坏使原结果校验失败,不能以已有通过状态复用 (3991.7381ms)
✔ fixed browsing contracts reject local host APIs, credential URLs, protocols, cross-origin and arbitrary execution (0.9779ms)
✔ attested desktop queue preserves same-tab observations, original PNG bytes and approval-bound one-shot actions (1501.6585ms)
✔ page data cannot grant domains, interaction or other Task scope; nonce and ownership remain fixed (983.4316ms)
✔ unknown dispatched submission blocks even after a new observation, rejects late success and retains explicit acknowledgement (1582.3054ms)
✔ downloads require exact approved address, remain original binary artifacts, and cannot reuse consumed observation (800.0162ms)
✔ revoking shared persistent profile closes all its tabs and clears only that profile after original desktop acknowledgement (2029.9044ms)
✔ service recovery never reopens old tabs or replays dispatched actions and keeps unknown original receipts (1595.92ms)
✔ agent browser flow waits for original Task/Run approval, dispatches once, and returns paired original results (5823.7559ms)
✔ changed approved input and stopped dispatched agent action cannot be accepted or replayed (1204.6185ms)
✔ browser inventory gives bounded metadata and original references instead of repeating full website histories (1639.2534ms)
✔ receipt storage failure settles the original dispatched action as unknown and never hangs or accepts late success (1949.2577ms)
✔ input that fits the engine window but exceeds remaining credit diagnoses the credit limit (0.5822ms)
✔ window overflow and historical pauses retain distinct and honest diagnoses (0.0828ms)
✔ R observed reservation shortage keeps the window-fit diagnosis separate from Run/Task credit (0.0919ms)
✔ a budget pause persists concrete text/tool/history/image breakdown and never dispatches an oversized request (3560.6331ms)
✔ explicit frozen preview resumes the same Task in a new Run, preserves completed tools and does not change global config or selected model (4989.1003ms)
✔ insufficient budgets, silently clamped capacity and arbitrary config fields reject without a physical request or mutating task budget (3563.0133ms)
✔ changed session, changed registry, expiry and altered frozen preview all require a new explicit preview (4007.1529ms)
✔ preview preserves unknown invocation status and increasing Task allowance is committed only with confirmed new Run (6005.9552ms)
✔ batch APIs enforce ownership and review fingerprints, apply once, and persist the completed batch across restart (7059.9125ms)
✔ delivery reports bind actual batch identity and reject corrupted snapshot evidence without changing project files (4968.7893ms)
✔ a compatible model can seal proposals through the real HTTP loop while application stays behind user review (9465.9206ms)
✔ a sealed batch at the step limit hands off and continues the same Task into actual checks without replaying saved tools (22281.7815ms)
✔ applied batches cannot turn cancellation, other budgets, restart or unknown commands into a review continuation (2035.5094ms)
▶ complete sealed batch hands off after all tools settle with no next model request
✔ complete (6228.6206ms)
✔ later-error (6949.2145ms)
✔ complete sealed batch hands off after all tools settle with no next model request (13178.6089ms)
▶ partial, changed, failed or cancelled batches cannot claim complete review handoff
✔ partial (5563.852ms)
✔ new-proposal (5399.4306ms)
✔ failed (5241.4717ms)
✔ conflict (5469.3714ms)
✔ cancelled (4717.1909ms)
✔ unknown (5180.2197ms)
✔ tampered (5872.8531ms)
✔ partial, changed, failed or cancelled batches cannot claim complete review handoff (37445.7064ms)
✔ actual process exit at file-batch prepared preserves durable identity and never replays writes (1988.0563ms)
✔ actual process exit at file-batch application_started preserves durable identity and never replays writes (2307.566ms)
✔ actual process exit at file-batch write_intent preserves durable identity and never replays writes (2236.2999ms)
✔ actual process exit at file-batch staged preserves durable identity and never replays writes (2099.1863ms)
✔ actual process exit at file-batch replaced preserves durable identity and never replays writes (2164.0649ms)
✔ actual process exit at file-batch file_recorded preserves durable identity and never replays writes (2662.1836ms)
✔ actual process exit at file-batch finished preserves durable identity and never replays writes (2309.852ms)
✔ actual process exit at file-batch rollback_started preserves durable identity and never replays writes (3487.5399ms)
✔ actual process exit at file-batch restore_intent preserves durable identity and never replays writes (3708.904ms)
✔ actual process exit at file-batch restore_staged preserves durable identity and never replays writes (3687.0315ms)
✔ actual process exit at file-batch restored preserves durable identity and never replays writes (3041.8648ms)
✔ actual process exit at file-batch restore_recorded preserves durable identity and never replays writes (2608.1152ms)
✔ actual process exit at file-batch rollback_finished preserves durable identity and never replays writes (3120.6689ms)
warning: in the working copy of 'README.md', LF will be replaced by CRLF the next time Git touches it
warning: in the working copy of 'src/two.mjs', LF will be replaced by CRLF the next time Git touches it
✔ sealed batches preserve bytes until review, then apply and explicitly restore BOM/CRLF/new files (2662.6303ms)
Switched to a new branch 'owner-branch'
✔ a conflict in the last file or Git branch prevents writing every file in the batch (3272.071ms)
✔ duplicate canonical paths, ancestor collisions, links, dependencies and Windows aliases never form a batch (1340.5554ms)
✔ fingerprints, changed proposals and corrupt snapshot artifacts cannot write or restore any target (2718.8813ms)
✔ a failure after replacement restores only batch bytes and never retries an old application (1672.7031ms)
✔ rollback preserves owner edits, blocks other project writers, and explicit acknowledgement releases recovery (2490.3514ms)
✔ an unattempted owner-created file identical to a suggestion remains external and is never removed (1648.1335ms)
✔ active batch requests are serialized and cancellation leaves a verifiable ended batch (1667.1361ms)
✔ batch application and rollback preserve unrelated Git index and worktree bytes (3317.6108ms)
✔ recovery reuses exact staged bytes but retains a conflicting temporary file for owner inspection (2046.339ms)
✔ one Task can seal pending proposals from earlier Runs while snapshots retain each source Run (2091.4041ms)
✔ engineering inspection preserves a dirty Git baseline and treats project rules as data (2654.6254ms)
✔ real npm tests and build produce counted results, log artifacts and persistent verification (3084.5114ms)
✔ an actual failing test retains the nonzero exit code and failed count (2601.4536ms)
✔ unchanged automatic checks return original failure evidence, while explicit user reruns and changed source still execute (5285.1545ms)
✔ agent loops cannot hide identical failing checks behind new plan IDs or unrelated reads (10578.3598ms)
✔ zero tests, skipped tests and unrecognized stdout cannot become verification passes (8150.4561ms)
✔ a successful build with no selected tests does not verify the task (2229.9831ms)
✔ stale check plans and forged script names execute no project commands (1905.9631ms)
✔ applying a later proposal invalidates the prior passed engineering snapshot (3042.1188ms)
✔ stopping an actual check terminates its child process and blocks concurrent jobs (2025.3604ms)
✔ check timeout records a failure with its actual termination result (2236.2448ms)
✔ source edits during a successful check make the result stale (3410.9669ms)
✔ bounded large logs retain the test footer and readable artifact instead of inventing counts (1932.3749ms)
✔ model-selected checks require a run-bound decision and denial creates no check job (4322.3314ms)
✔ check APIs expose owned plans and jobs and preserve the authoritative results across restart (7427.8821ms)
✔ reviewed changes continue the original task without reopening a generic completed run (2900.1763ms)
✔ concurrent review-continuation API requests accept one run and retain the goal after restart (9888.6924ms)
✔ the model reads owned failure logs without running the check again (3464.4588ms)
✔ process exit at check queued preserves results and never replays commands (1490.7335ms)
✔ process exit at check result_committed preserves results and never replays commands (1966.9634ms)
✔ model code tools keep Task/Run ownership, index evidence and exact proposals across storage recovery (6432.9541ms)
✔ bundled code APIs isolate project results, validate queries and create review-only patches with conflicts (3668.6897ms)
✔ code tool arguments reject malformed ownership-independent inputs before side effects (0.8917ms)
✔ observe-only device authority cannot read, configure or clear project parsing caches (1132.5413ms)
✔ pairing is short lived, single-use, generation-bound and validates exact permissions (442.7775ms)
✔ device transaction failure neither consumes the ticket nor leaves a credential (691.9586ms)
✔ devices have project and action boundaries; rotation, expiration, revocation and host changes reject old authority (906.9628ms)
✔ persistent registry stores hashes and a bounded audit, never raw credentials (605.0953ms)
✔ real TLS verifies IP SAN, CA chain, encrypted restart and public fingerprint (880.3108ms)
✔ TLS sealing failure leaves listener closed and no plaintext or successful TLS registration (619.1946ms)
✔ HTTP pairing is concurrent single-use; secure cookie and strict origin/host checks (765.5979ms)
✔ gateway never exposes management routes, arbitrary assets or unpaired project data (1233.975ms)
✔ authorized project events paginate from original seq without leaking other project events or payloads (1343.7283ms)
✔ remote task projection excludes raw model history, host configuration, tool input and credentials (1460.9938ms)
✔ permission checks apply to direct remote mutation, not only UI controls (1085.6212ms)
✔ approval fingerprint and exact task/revision are bound; duplicate requests never execute twice (1011.7209ms)
✔ unknown original operation is recorded and never replayed after response failure (817.8507ms)
✔ revocation and gateway closure reject old access without invoking runtime stop (913.9779ms)
✔ rate limits reject pairing brute force and do not persist candidate secrets (829.772ms)
✔ real Windows CurrentUser DPAPI persists only a sealed CA and reopens it (1629.4825ms)
✔ mobile approval executes the original Runtime command exactly once and rejects the later desktop decision (2712.2019ms)
✔ desktop rejection wins a concurrent mobile request and original command remains unexecuted (1647.9466ms)
✔ phone stop waits for the original process exit and prevents later heartbeats and command replay (2761.4496ms)
✔ Run replacement during parent-stop wait is rejected rather than stopping the new Run (723.6222ms)
✔ diff pages preserve source SHA; report reads verify blobs and exclude unrelated artifact kinds (1135.5981ms)
✔ hot upgrade refuses an active gateway and restores closed gateway ownership on failure (701.9473ms)
✔ damaged device permissions or expiry fail closed and never overwrite the original registry (849.5353ms)
✔ device revocation destroys its actual authenticated TLS connection (1279.1989ms)
✔ companion observes only current Task screenshots, verifies original PNG bytes and loses access immediately on revoke (1562.7267ms)
✔ mobile browser observation is current Task/project scoped, preserves original PNG and exposes no browser control or profile data (1113.7203ms)
✔ actual config HTTP failure preserves the committed selection, config and volatile credential before a successful retry (873.5993ms)
✔ controlled installation pins the native kernel and bridge independently of the existing connection-probe identity (854.5115ms)
✔ controlled profiles disable every native project/terminal provider and native model/retry route using the actual include patch contract (0.7413ms)
✔ owned tool catalogues seal full schemas canonically and reject duplicate or malformed declarations (5.0058ms)
✔ bridge authentication, current Run, process, SDK session and browser-origin checks reject before dispatch (724.2057ms)
✔ bridge duplicates, unknown tools, malformed arguments and source changes are refused before a second side effect (762.2785ms)
✔ bridge model requests bind the exact catalogue, route and output bound and preserve raw request/result identity (638.9372ms)
✔ stopped and oversized bridge requests never reach a local model or tool callback (847.5664ms)
✔ a Windows virtual data-directory name binds new storage and credentials to its actual physical root (1809.3584ms)
✔ physical data-root adoption still rejects explicit junctions at the directory or any ancestor (845.7533ms)
✔ actual Windows CurrentUser DPAPI roundtrips owned UTF-8 and rejects damaged bytes and wrong entropy (1389.7178ms)
✔ only explicitly supplied keys persist; new provider restores original revisions and public views contain no key (2599.9777ms)
✔ registry and cipher commit failure restores original routing, keys, block and every database row (1858.9819ms)
✔ a two-key import with a protection failure rolls back all registry and cipher entries and retains its preview (2158.7069ms)
✔ connection revision changes retain the original saved block and report a source mismatch on restart (2443.9877ms)
✔ disable skips real restart loading, explicit enable restores, delete preserves memory and forget removes both (2740.7937ms)
✔ corrupt saved block and another SID do not load; failures are visible without returning ciphertext or plaintext (2645.758ms)
✔ legacy persistence requires a new explicit key and binds config and credential revisions; stale requests cannot save (2938.3809ms)
✔ owned plaintext keys never reach SQLite pages, WAL, config or public execution exports (1444.7079ms)
✔ reviewed application invalidates old checks, names current SHA and advances through a new approved check to Git preview (7483.4072ms)
✔ skip-tests, foreign task, unknown operations, diagnostic and parent review retain their boundaries (3517.626ms)
✔ before the first change, real baseline failures persist across later reads; passed baselines and changed sources cannot certify delivery (8802.4617ms)
✔ actual process exit at Git queued never replays a commit and can reconcile the durable result (13088.6136ms)
✔ actual process exit at Git preparing never replays a commit and can reconcile the durable result (14254.9591ms)
✔ actual process exit at Git index_locked never replays a commit and can reconcile the durable result (13129.382ms)
✔ actual process exit at Git prepared never replays a commit and can reconcile the durable result (11863.0829ms)
✔ actual process exit at Git ref_prepared never replays a commit and can reconcile the durable result (13461.5876ms)
✔ actual process exit at Git publishing never replays a commit and can reconcile the durable result (12870.2179ms)
✔ actual process exit at Git committed never replays a commit and can reconcile the durable result (9685.3023ms)
✔ actual process exit at Git index_published never replays a commit and can reconcile the durable result (10510.3092ms)
✔ actual process exit at Git finished never replays a commit and can reconcile the durable result (9577.8647ms)
✔ an unknown Git result cannot overwrite an externally changed index or delete an unrelated lock (8669.0714ms)
✔ known publication with a later source edit retains the commit and marks the delivery for inspection (8791.2068ms)
✔ CRLF normalization, multiple applied revisions and unrelated special index flags preserve exact bytes and ownership (10112.5524ms)
✔ adding new durable record kinds to an existing hot Store preserves all existing Maps and records (10471.4612ms)
✔ an unrelated owner deletion is fingerprinted and preserved while linked task paths are rejected (13488.5446ms)
✔ actual failed tests become a reviewed local commit and evidence report while preserving unrelated index and files (15667.5228ms)
✔ selected preexisting unstaged, staged and untracked owner edits cannot enter a task commit (15815.2772ms)
Switched to a new branch 'owner-branch'
✔ source, unrelated stage and branch changes after review invalidate a commit plan without changing HEAD (18173.4077ms)
Switched to a new branch 'delivery-branch'
✔ a later Git metadata change keeps matching source evidence valid but a real source edit requires checks (9397.832ms)
✔ a corrupt passed-check log and a pending proposal block Git delivery instead of inventing evidence (4680.0676ms)
HEAD is now at 95738c8 Sealed delivery fixture baseline
✔ Git filters, merge state, detached HEAD and old task ownership are explicitly rejected (20218.548ms)
✔ local snapshot commits disable repository hooks and signing and isolate the Git child environment (8088.0999ms)
✔ stopping before publication cancels safely; stopping after publication preserves the actual completed commit (15752.4611ms)
✔ delivery API confirms exactly one owned plan, serves verified reports and preserves results across restart (11869.6216ms)
✔ the model can prepare an owned Git review and evidence report while publishing remains a separate user action (11905.2653ms)
✔ HTTP Git review pairs deferred calls, survives reload and never requests more work until the saved preview is resolved (9698.2742ms)
✔ a reviewed Git preview can be revised or withdrawn without publishing or losing the current Task (8852.5483ms)
✔ actual Electron debug argument ordering retains exactly the compiled entry and rejects extra or repeated actions (1.0728ms)
✔ native bridge refuses other origins, subframes, other windows and unsupported inputs before connecting (549.8781ms)
✔ desktop preview opening requires a currently registered project preview and rejects iframe callers (635.5413ms)
✔ file reveal binds the current project, task, path, UTF-8 bytes and original SHA without executing the file (653.7522ms)
✔ file changed during asynchronous task recheck is not revealed as the old bytes (581.0342ms)
✔ desktop browser profile and aliases stay outside project inventory and file access (616.1325ms)
✔ native picker cancel and origin change after dialog publish no project registration (627.6214ms)
✔ pending native action is not dispatched twice, and current task review returns its original seq (492.1524ms)
✔ native stop cancellation and changed original host never stop a replacement process (591.5164ms)
✔ window register rejects host PID, stale installation, extra fields and missing creation identity (0.407ms)
✔ a rebuilt source archive binds actual generated client assets and retains its original running identity after disk changes (1898.4418ms)
✔ host challenge proof rejects a stale challenge, changed identity and changed signature (0.9943ms)
✔ stop confirmation binds the original instance, creation and loaded runtime (0.148ms)
✔ state paths reject escaping, linked ancestors, ambiguous hardlinks and damaged JSON (1043.9754ms)
✔ failed registration publishes no initialized memory and can complete after its actual inputs are supplied (2727.7382ms)
✔ host shutdown cancels an un-signalled HTTP model request and refuses later requests until recovered (1.4176ms)
✔ host shutdown cancels model-pool acquisition and releases queue ownership without inventing attempts (548.3186ms)
✔ full candidate validation returns the exact loaded-input host identity and rejects undeclared or altered files (844.943ms)
✔ fast per-file scan still rejects a junctioned subdirectory, a hard-linked file and a stream path with the original bytes intact (1012.478ms)
✔ installation requires separate absolute program/data directories and a candidate in its own versions namespace (1292.0528ms)
✔ inventory rejects a new hardlink on an optional file after its full-manifest scan and accepts the untouched candidate after removal (705.4721ms)
✔ staging retains the original data and pointer without querying or restarting a host (2776.8287ms)
✔ a post-pointer failure restores every original pointer byte and persistent row; a first-install failure removes only its published pointer (10390.9568ms)
✔ changed active manifest and concurrent pointer changes are rejected without overwriting the unknown record (4232.4444ms)
✔ an unregistered non-HTTP listener blocks preflight even when service-info is absent (2724.3631ms)
✔ uninstall archives its original pointer and preserves every data row/artifact instead of recursively deleting data (7157.6855ms)
✔ preflight rejects an unrelated program root or changed native bootstrap while preserving its original bytes (4250.3372ms)
✔ a changed owned bootstrap follows its candidate and a post-pointer fault restores both original bytes (17060.1579ms)
✔ an external bootstrap change after preflight is preserved and never overwritten as an update or rollback (8585.6484ms)
✔ owned startup retries only read-only observation after a slow health response and attests the same original child (3872.0569ms)
✔ unverified startup expires with its original live child and never treats timeout or an unrelated health PID as proof (5740.6364ms)
✔ diagnostic failures and repeated questions retain the original execution error, task, checks and read-only authority (2731.6658ms)
✔ diagnostic binding survives reopen, rejects foreign Task references and detects original error or artifact tampering (869.4472ms)
✔ a new Task never adopts a previous Task error from session.lastFailure (771.2781ms)
✔ verified research and explicit no-test scopes do not demand an unrelated test job or claim engineering success (2859.1054ms)
✔ unknown scope and unknown operations remain actionable even after a diagnostic answer or read proof (1372.6785ms)
✔ delivery status HTTP is read-only, selection-bound and equal to the saved report after host restart (5142.9662ms)
✔ routine engineering facts remove duplicate success excerpts while complete diagnostics and errors remain available (789.9072ms)
✔ a question about the previous workbench error receives local error context and current change states (2302.6089ms)
✔ an unknown model request stops, and explicit resume never replays the approved command (3355.572ms)
✔ exhausted refused connections preserve tool results, and explicit resume adds no new user turn or replay (3364.7716ms)
✔ stopping during model backoff cancels further requests and clears the running request state (1643.4769ms)
✔ HTTP authentication errors are classified, redacted and never automatically retried (17.3421ms)
✔ temporary HTTP service errors retry and recover with recorded attempt counts (13.1624ms)
✔ invalid and oversized protocol responses fail once with a useful response-format diagnosis (5.8087ms)
✔ HTTP rate limits expose their source and obey the service retry interval before recovery (1008.5299ms)
✔ Windows commands preserve Chinese through a PowerShell pipeline into Node (961.2004ms)
✔ Windows shell syntax errors stay readable and native command exit codes are retained (1108.6449ms)
✔ review and completed states do not turn a generic continue click into a new model request (775.8141ms)
✔ old session data receives a truthful historical connection diagnosis without rewriting history (765.6053ms)
✔ engine HTTP checks bind the detected installation and current task, page original bytes and preserve them across host restart (5383.9091ms)
✔ HTTP stop seals a pending SDK call, rejects automatic replay and keeps failures attached to the original session (4110.7924ms)
✔ sealed r1-off failure fits its original 16K config with authentic latest status and readable source (12.4433ms)
✔ sealed r1-on failure fits its original 16K config with authentic latest status and readable source (9.7237ms)
✔ sealed r2-on failure fits its original 16K config with authentic latest status and readable source (7.4435ms)
✔ sealed r2-off failure fits its original 16K config with authentic latest status and readable source (7.2101ms)
✔ archived native call pairs require exact saved arguments/results rather than model-written summaries (7.7787ms)
✔ full system/user requirements and mandatory current failure can refuse rather than be trimmed (12.5944ms)
✔ small native input is byte-preserving; developer instructions remain mandatory after compaction (0.1642ms)
✔ older failing checks and pending proposals remain visible after a later large file read (14.4686ms)
✔ bounded result keeps every check status and cuts Unicode only on whole characters (0.449ms)
✔ controlled engine HTTP freezes one child, pages original owned input/report, rejects foreign reads and prevents changed-source reuse (15145.7298ms)
✔ a sealed whole-batch review pauses the owned native session without certifying engine completion or losing the saved host result (14330.9558ms)
✔ a changed review seal still closes the owned native process and cannot claim a verified review handoff (14653.7397ms)
✔ native text and call IDs preserve original result pairing while unsupported content fails closed (0.4414ms)
✔ one engine child freezes real installation identity before creating tasks and rejects a changed source (8688.9197ms)
✔ owned native tools return saved results, use real approvals/checks and account each actual request once across new engine Runs (28180.1356ms)
✔ forged native scope cannot change a parent file and a final completion claim cannot replace worker proof (13309.7501ms)
✔ duplicate native IDs are rejected without replaying a saved tool (11997.9325ms)
✔ native input exceeding the context budget is retained and sends no model request (11206.7839ms)
✔ bounded native sends verify original tool provenance and reject forged views or missing versions (13450.5496ms)
✔ stop retains a late reported model usage once and executes no tool from that output (11962.3268ms)
✔ a source change after the original model intent stops before native tools execute (11057.3597ms)
✔ an engine source change while approval is pending invalidates the old permission before any check starts (14832.7533ms)
✔ missing or noncompleted turn end cannot be adopted even with an owned receipt and assistant text (0.5735ms)
✔ closed root with an original live descendant remains unknown and retains its process tree (0.2874ms)
✔ host recovery reserves original pending attempts exactly once and explicit review preserves unknown outcomes (15757.2819ms)
✔ corrupted model reservation cannot reduce recovered budget or permit a new engine request (12844.3918ms)
✔ engine detection seals the actual CLI, SDK, profile and Node and rejects mixed versions without launch (861.5292ms)
✔ owned engine probe waits for real ready identity, verifies six native responses and saves immutable task evidence (1498.2991ms)
✔ an invalid startup nonce is rejected before initialize and preserves the process close and partial report (1544.7595ms)
✔ SDK identity and session directory mismatches retain raw responses and never become verified (4487.3297ms)
✔ cancel during a pending initialize closes only the owned process without sending an unsupported MCP cancellation (1637.0627ms)
✔ cancel before readiness sends no protocol request and a user retry creates a different process and record (2009.2106ms)
✔ source or scratch changes and unexpected model traffic fail after preserving the actual observed result (4426.5657ms)
✔ unowned idle notifications stay quarantined and server requests receive a rejection without granting operations (2864.8945ms)
✔ engine probe environment excludes parent secrets and NODE_OPTIONS while preserving operating system paths (1862.5582ms)
✔ foreign sessions and altered reports or task bindings cannot reuse verified engine evidence (1455.8956ms)
✔ restart never replays an active engine check or stops a historical PID, and keeps prior task state (1299.547ms)
✔ model argument validation and runtime activity do not grant external engine prompt or project execution (578.4506ms)
✔ a follow-up question receives the actual failed engine check from its original Task and keeps its report unchanged (2436.103ms)
✔ unfixed dependency URLs and invalid integrity are reported without exposing URL credentials or allowing a plan (1090.7739ms)
✔ hot upgrade refuses an active dependency operation or evaluation before changing runtime objects (0.5804ms)
✔ read-only environment diagnosis identifies actual runtime, missing dependencies and sealed inputs without installing or running scripts (6264.1643ms)
✔ diagnosis rejects unsupported managers, lock integrity and credential configuration without reading config content (5401.3112ms)
✔ approved real npm ci runs only in owned worktree, skips scripts, retains logs and binds later actual tests (10866.7851ms)
✔ explicitly allowing lifecycle scripts executes sealed root script and preserves the separate preparation result (8854.1571ms)
✔ stale lockfile, tampered plan/config, cross-session identities and mismatched download sources never dispatch npm (7066.5551ms)
✔ registry failure keeps original npm exit/log and performs no automatic fallback or replay (8816.2117ms)
✔ stopping a hanging dependency download records only its owned process and preserves stopped logs (8441.0162ms)
✔ restart retains unknown preparation, blocks writes and checks, and requires explicit read-only reconciliation (6965.8196ms)
✔ HTTP environment flow survives refresh, requires explicit approval and preserves task and artifact ownership (13799.228ms)
✔ Python environment diagnosis reads declarations and runtime identity without executing project code or installing packages (1011.5866ms)
✔ empty build cache directories keep dependency identity while added executable files invalidate it (1157.2217ms)
✔ dependency snapshots reject external links and incomplete budgets instead of claiming environment equivalence (776.0192ms)
✔ model-requested environment preparation uses the original approval boundary and never treats preparation as passed engineering checks (13071.4885ms)
✔ timeout and offline cache miss retain actual failures and new attempts require newly sealed plans (12779.3681ms)
✔ both evaluation worktrees can approve fixed dependency preparation and different actual dependencies reject candidate execution (30377.171ms)
✔ evaluation denominator distinguishes attempted unsealed candidates from unexecuted samples (0.3145ms)
✔ evaluation summary retains sealed outcomes and stopped or failed original attempts (0.4346ms)
✔ evaluation contract fixes bounded samples, source paths, checks and aggregate budget (495.6177ms)
✔ same-source workspaces seal real commands before execution and preserve original staged and unstaged bytes (15175.8828ms)
✔ actual original failures and candidate success are both preserved, then adoption requires parent review, checks and Git (61641.1839ms)
✔ candidate cannot shrink tests and changing original test bytes rejects execution before any model request (22968.3598ms)
✔ missing upstream usage remains unknown and does not become a free run or invented cost (39621.2438ms)
✔ missing raw check logs preserve unknown result and original sample denominator (38674.1546ms)
✔ source and model changes never silently replace an existing batch input (3692.1131ms)
✔ model revision change before preparing a fixed sample cannot silently use the replacement connection (4079.6613ms)
✔ HTTP evaluations require host token, original fingerprint and session ownership; worker API cannot bypass evaluation flow (17100.5952ms)
✔ sample queue cannot skip or reset attempted samples, and reports freeze further writes (29490.5238ms)
✔ stop cancels only current candidate and preserves partial results instead of replaying a request (25949.8334ms)
✔ restart recovers interrupted evaluations as unknown and keeps completed checks and plan identity (21803.5427ms)
✔ JSON migration survives process exit before commit and retains exact legacy backups (2313.3777ms)
✔ process exit after_state rolls back state, event and outbox together (1754.3393ms)
✔ process exit after_event rolls back state, event and outbox together (2096.7451ms)
✔ crash recovery at accepted retains execution identity and never replays side effects (1948.2587ms)
✔ crash recovery at queued retains execution identity and never replays side effects (3240.3766ms)
✔ crash recovery at approval_wait retains execution identity and never replays side effects (3415.2897ms)
✔ crash recovery at approved retains execution identity and never replays side effects (3115.3911ms)
✔ crash recovery at effect_unrecorded retains execution identity and never replays side effects (3540.3093ms)
✔ crash recovery at result_committed retains execution identity and never replays side effects (3790.6646ms)
✔ crash recovery at before_next_model retains execution identity and never replays side effects (3799.2686ms)
✔ state and event failures publish no success and preserve the last committed memory snapshot (1016.9312ms)
✔ tool result artifacts and event cursors retain project ownership across restart (2435.712ms)
✔ token budgets pause before tools execute and continuation keeps original task and accumulated usage (2169.7654ms)
✔ a time budget interrupts an unresponsive model and persists a resumable run (1343.6043ms)
✔ a corrupted large result artifact cannot become a recovered success (1142.8234ms)
✔ hard-link aliases cannot expose protected bytes through reads, proposals, inventory or automatic project rules (790.5336ms)
✔ ambiguous project hard links are explicit and ordinary independent file copies remain readable (809.3703ms)
✔ current-file HTTP receipts bind Task/Run and bytes, persist on restart and end without altering original runs or user Git data (6440.3135ms)
✔ file targets exclude referenced inputs and prohibited subsequent modifications (1.6794ms)
✔ full actual read and existing receipt complete a no-test file task without fabricating changes, checks or commits (3366.6377ms)
✔ partial pages, partial lines, wrong hashes and failed reads cannot be adopted; contiguous full paging can (1356.2645ms)
✔ user edits and new requirements invalidate receipts, and a second task must explicitly adopt a source in its own project/session (1241.1135ms)
✔ adopting an unchanged file never satisfies an explicit modification or a new modification instruction (862.3203ms)
✔ required checks and stale passing flags cannot be bypassed by current file receipts (1202.8287ms)
✔ results plan proofs bind current file receipts while changes proofs still demand real applied change sets (908.0737ms)
✔ duplicate receipts cannot create progress indefinitely or reset no-progress reads (3416.6638ms)
✔ manual completion preserves the failed original run and refuses unknown operations (1223.4514ms)
✔ completion refuses active work, files changed during inspection and a newly selected task (928.9619ms)
✔ unrelated files cannot stand in for named output paths, nor can a late apply satisfy newer instructions with an old proposal (1288.3918ms)
✔ delivery and mobile projections distinguish adopted bytes, new changes, no tests and no commit; mutations cannot rewrite receipts (1556.9747ms)
✔ proposals preserve project bytes until review; apply backs up the exact original bytes (952.7156ms)
✔ new nested files are created only on apply and unchanged content creates no proposal (584.2236ms)
✔ stale reads and changed baselines are rejected without overwriting external edits (823.9262ms)
✔ file paths reject traversal, absolute paths, credentials and outside junctions (912.3869ms)
✔ read and search expose bounded text scopes and reject non UTF-8 / oversized files (642.9473ms)
✔ project-local junctions cannot expose or create protected Git and credential files (581.2252ms)
✔ Windows trailing-dot and trailing-space path aliases cannot bypass protected paths (645.6425ms)
✔ a starter is copied into user data once; restarting from another program version preserves existing projects and editable bytes (1301.8121ms)
✔ startup does not insert a demo or rewrite/reorder an existing user project registry (1288.9956ms)
✔ PNG bytes are fully decoded, metadata-free and bounded; JPEG EXIF orientation is applied to the sent PNG (105.7313ms)
✔ forged type, short headers, CRC damage, incomplete chunks, trailing payloads, animation and dimension bombs reject (23.7624ms)
✔ draft image ownership, withdrawal, reload and atomic Task binding retain original SHA and original bytes (1732.8876ms)
✔ a failed session transaction restores draft status and cannot leave a Run consuming the attachment (982.9136ms)
✔ draft limits, duplicate selection, source changes and cross-Task association reject before dispatch (1623.5042ms)
✔ demo and legacy paths refuse image dispatch without creating a Task or consuming the draft (994.1068ms)
✔ chat-completions sends one owned image per actual request, pairs original tools and verifies exact sealed bytes (2727.6958ms)
✔ responses sends one owned image per actual request, pairs original tools and verifies exact sealed bytes (3439.8926ms)
✔ mixed pool excludes unknown vision before lease and sends zero HTTP to the incompatible member (1888.5918ms)
✔ vision probe requires correct randomized pixel order and evidence expires when the connection revision changes (1141.38ms)
✔ HTTP upload/read/withdraw uses session authorization, returns byte SHA and refuses unknown ownership (833.6016ms)
✔ image accounting excludes Base64 as text but seals every original request byte and reserves image input separately (3.1189ms)
✔ actual model comes from owned consumption, excludes other Tasks and deduplicates the current Run (1.6161ms)
✔ mixed routes retain each original model/protocol/outcome and unknown use is never reported as zero (0.1484ms)
✔ zero dispatch, missing ledger, missing model and truncated Run history never borrow configured model identity (0.2206ms)
✔ repeated attempt identities across Runs and contradictory usage stay explicit instead of inflating successes (0.1156ms)
✔ observed route never certifies returned identity and retains mismatches and original source (0.1526ms)
✔ directory picker reads one level, preserves empty folders, skips credential and linked folders (701.6152ms)
✔ HTML preview serves encoded names and relative modules on an isolated loopback origin; closing revokes it (631.3283ms)
✔ HTML preview refuses credentials, hidden inventories, traversal, external junctions, hardlinks and unsupported file types (560.3547ms)
✔ runtime shutdown closes local preview ports and pending starts cannot escape shutdown (678.324ms)
✔ idle file previews expire and release their loopback servers (695.6717ms)
✔ directory and preview HTTP controls require workbench authentication; project pages cannot read bootstrap (2324.1347ms)
✔ MCP registration stays disabled until an explicit real stdio probe and schema-based tool selection, with actual process cleanup (2255.2941ms)
✔ MCP probe holds the project operation lock until its owned process and report are settled (1567.2059ms)
✔ MCP validates the declared JSON schema subset and rejects unsupported schemas before external dispatch (0.6406ms)
✔ actual model MCP call approves long schema-valid arguments, validates the sent catalog and stores one original result without leaking environment secrets (5199.3439ms)
✔ denying an MCP call never starts another process or executes the external tool (3998.9279ms)
✔ actual handshake drift rejects tools/call even if program entry hash is unchanged (5287.2123ms)
✔ cancelled dispatched external side effect stays unknown, closes its owned process and cannot replay on reopen (4449.8105ms)
✔ changed executable input, foreign project IDs and corrupted original contracts block enable and dispatch (1956.2002ms)
✔ registered MCP HTTP API uses fixed execution contracts and project-owned probe records with no model requests (3953.3643ms)
✔ unknown MCP pauses later tool calls and all project writers; fingerprinted acknowledgement clears only the review gates and explicit resume does not replay (6494.792ms)
✔ MCP acknowledgement persistence failure restores both job and tool review gates; corrupt evidence cannot be acknowledged (5354.5523ms)
✔ actual host exit after an external effect restores unknown jobs without a fabricated report, allows source-bound review and never replays the call (4016.0582ms)
✔ changed enabled MCP programs can be explicitly disabled without launching or trusting their changed source (1849.1171ms)
✔ stdio preserves split UTF-8 frames and final replies drained after process exit (2114.9829ms)
✔ server-initiated model requests are refused instead of inheriting provider privileges (788.6378ms)
✔ invalid or over-budget protocol invalid closes only its owned process and is not retried (893.3673ms)
✔ invalid or over-budget protocol oversized closes only its owned process and is not retried (822.1523ms)
✔ cancel and timeout do not resend a dispatched request and pre-cancel has no request trace (884.692ms)
✔ startup failure remains undispatched and has no process PID to terminate (621.3192ms)
✔ REA launch filters a fixture credential canary and rejects changed entry identity before a new connection (1137.0712ms)
✔ schema1 connections remain readable without inventing observed capabilities or migrating original state (771.2645ms)
✔ protocol, unsupported capabilities and stream choice persist; client claims cannot forge verified evidence (895.5633ms)
✔ ability evidence belongs to exact connection and credential revision; unknown and unsupported reject required routing (0.3455ms)
✔ invalid declarations and inconsistent stream settings reject before registry, revision and credentials mutate (579.0634ms)
✔ upstream declarations include brand metadata and remain distinct from adapter support or verification (1.3171ms)
✔ new upstream retrieves more than 100 models without a manual model; choosing several stores one key and frozen metadata (1010.0639ms)
✔ Responses and DeepSeek list use bearer; Messages root and v1 endpoints paginate with selected authentication (623.1469ms)
✔ refresh preserves existing key objects, encrypted block, parameters and user capability overrides without re-encryption (2024.8244ms)
✔ shared credentials restore from the original saved block; source disable, endpoint changes and key replacement cannot revive frozen runs (3597.7247ms)
✔ automatic selection reuses capability and health filtering, manual selection sends the exact chosen model and shared auth (1031.6342ms)
✔ unknown catalog capabilities require explicit declarations; upstream negative and retired models cannot be overridden (974.4568ms)
✔ model reasoning controls use declared effort and mode ranges without silently enabling unsupported options (781.0448ms)
✔ cancelled, expired, changed-source and failed catalogs cannot save or overwrite a later selection (969.88ms)
✔ pagination loops, conflicting models and upstream errors fail once without creating members or model attempts (541.522ms)
✔ limits and transaction failures preserve original registry, config, cipher and preview for explicit correction (1534.8902ms)
✔ a model that switches from a shared to an explicit credential receives a new revision for opaque-history isolation (622.4869ms)
✔ compiled HTTP directory flow persists selection and references, gates changes during fetch and permits explicit cancel (2308.6634ms)
✔ large directories declare partial coverage and a batch crossing the local connection limit leaves no partial members (4247.7903ms)
✔ a distinct upstream account at the same endpoint and model never silently reuses another account (882.2949ms)
✔ source changes invalidate an in-flight child probe and host shutdown cancels directory work and drops transient previews (1108.1457ms)
✔ model registry keeps original credentials/endpoint intact, rejects stale edits, and forgets only volatile keys on restart (1567.345ms)
✔ config import previews exact revisions, keeps secrets volatile, syncs only when enabled and preserves manual edits on conflicts (1168.4683ms)
✔ real HTTP pool fails over 429 to another model and never repeats completed tools; attempts seal actual request bytes and Run identity (4421.8691ms)
✔ pool rotation and weights select members on new Runs; disabled and missing credentials are excluded without requests (6088.8958ms)
✔ 401 blocks the credential revision; updating the key restores availability without leaking error echoes (1991.0143ms)
✔ queued request cancellation and in-flight disable release only owned leases and do not dispatch another HTTP request (1184.6924ms)
✔ compiled model-center HTTP API persists public profiles, blocks forged origins, and local operations never invoke a model (2533.5023ms)
✔ model registry disk failures restore routing, pool profiles and credential references without partial updates (838.8313ms)
✔ model selection and saved configuration commit together and failures preserve the original mode and endpoint (785.1763ms)
✔ configuration imports roll back the entire batch and retain one usable preview after validation or commit failure (925.3949ms)
✔ HTTP quota retains native code/type/param/request id, bounded original and allowlisted headers, never repeats that connection (2713.9062ms)
✔ Retry-After above 90 seconds remains exact and a short local budget cannot trigger an early retry (2.0292ms)
✔ Retry-After date, expired date, huge value and invalid header retain explicit semantics (1.1818ms)
✔ unclassified 429 remains unspecified and every bounded actual retry retains its own original error (1774.8377ms)
✔ oversized HTTP error saves exactly the bounded prefix and labels its incomplete range (2389.882ms)
✔ Chat SSE error preserves upstream fields, original failed event, partial tool and reported usage without executing or retrying (2494.3184ms)
✔ Responses failed SSE retains original event/response identity and usage (2333.0126ms)
✔ Responses incomplete SSE retains original event/response identity and usage (1981.9849ms)
✔ unsupported successful Responses output retains original JSON and usage with a local protocol origin (2039.3045ms)
✔ HTTP 200 native error is still an upstream failure rather than a fabricated successful answer (0.8158ms)
✔ JSON-escaped credential echoes are redacted in the original preview and capture as well as parsed fields (0.5872ms)
✔ quota failover uses another independent pool member and never replays a failed stream (3291.507ms)
✔ native parameters are isolated by pool member, preserve Responses state and bind the original output reservation (0.3328ms)
✔ unsupported native fields and invalid values reject before dispatch instead of silently dropping or overriding constraints (0.3598ms)
✔ native settings for legacy and registered profiles persist without changing credentials or old records and reject stale edits (1186.2597ms)
✔ compiled native settings API saves/restores exact parameters and rejects reserved fields without mutating the profile (2246.3566ms)
✔ legacy connection test sends its saved native fields and selected output limit field without changing the configuration (2543.5329ms)
✔ identity distinguishes missing, invalid, mixed, exact and explicitly declared aliases without guessing (2.3832ms)
✔ chat-completions JSON rejects a different model before tool dispatch, retaining source and reported usage (1852.1058ms)
✔ chat-completions SSE rejects a different model before tool dispatch, retaining source and reported usage (2532.8426ms)
✔ responses JSON rejects a different model before tool dispatch, retaining source and reported usage (2967.1414ms)
✔ responses SSE rejects a different model before tool dispatch, retaining source and reported usage (1762.0642ms)
✔ record policy accepts missing identity but keeps it explicit, and strict declared mapping is frozen in the consumption (1.2381ms)
✔ mixed streamed identities cannot pass strict mode, and interrupted streams retain observed identifiers without authenticating them (1.4234ms)
✔ identity refusal never fails over to another pool member and raw response model cannot leak known credentials (2073.963ms)
✔ strict refusal remains visible even when the same original response exhausts the budget (1782.2239ms)
✔ strict identity crosses the controlled engine bridge with its original source and zero tool dispatch (13200.7313ms)
✔ frozen direct connection policy and delivery reports preserve response identity without filling historical gaps (2849.2017ms)
✔ saved engine budgets verify all Chat/Responses route variants and reject a mismatched output allowance (801.8858ms)
✔ owned CC Switch SQLite imports the selected TOML Responses wire_api and leaves unknown clients untouched (1522.8908ms)
✔ Responses executes one original tool through Runtime, carries encrypted Items, reports raw usage and seals exact wire bytes (3309.2314ms)
✔ mixed-protocol pool keeps both actual variants and changes route on 429 without re-executing completed tools (3385.0748ms)
▶ Responses incomplete, failed, missing call identity and unsupported hosted outputs never dispatch a tool
✔ Reject {"body":{"id":"resp_owned","status":"incomplete","output":[{"type":"function_call","id":"fc_partial" (1597.2483ms)
✔ Reject {"body":{"id":"resp_owned","status":"failed","output":[{"type":"function_call","id":"fc_failed","cal (2337.0336ms)
✔ Reject {"body":{"id":"resp_owned","status":"completed","output":[{"type":"function_call","id":"fc_bad","cal (1672.8251ms)
✔ Reject {"body":{"id":"resp_owned","status":"completed","output":[{"type":"computer_call","id":"not-supporte (2010.0088ms)
✔ Responses incomplete, failed, missing call identity and unsupported hosted outputs never dispatch a tool (7617.8736ms)
▶ invalid JSON in a nonstream Chat or Responses tool group prevents every tool and pool replay
✔ chat-completions (2257.8578ms)
✔ responses (1449.4892ms)
✔ invalid JSON in a nonstream Chat or Responses tool group prevents every tool and pool replay (3707.7084ms)
✔ wire adapter preserves tool IDs but never passes Responses-private Items to Chat or another endpoint (0.1972ms)
✔ connection JSON import accepts explicit Responses protocol and continues to reject unknown protocols without reading client sessions (890.3178ms)
✔ reservation uses actual known usage from the same route, never lowers estimates or rewrites original receipts (1.8471ms)
✔ reservation verification binds original Task/Run consumption and rejects changed usage or foreign provenance (0.8092ms)
✔ known high input calibrates the next request and stops before dispatch while preserving the completed read (2731.8463ms)
✔ one reported overrun stops legacy requests and keeps actual use and excess (2729.098ms)
✔ one reported overrun stops pool requests and keeps actual use and excess (3133.7581ms)
✔ one reported overrun stops connection requests and keeps actual use and excess (2912.6472ms)
✔ reported usage below the limit prevents an unaffordable legacy retry before another dispatch (2566.7979ms)
✔ reported usage below the limit prevents an unaffordable pool retry before another dispatch (3484.5692ms)
✔ native bridge seals calibrated refusal with original consumption and preserves versioned request verification (14211.3147ms)
✔ original R v1/v2 stay verifiable while v3 excludes the isolated input outlier and keeps the ledger intact (4.0209ms)
✔ v3 never excludes a repeated high ratio, the latest receipt, or a short history (3.0736ms)
✔ reservation proofs reject reordered latest high receipts, altered history and duplicates while old snapshots survive later consumption (3.1901ms)
✔ input-ratio exclusion preserves independently reported output maxima without rewriting historical v2 (2.0423ms)
✔ capability routing skips text-only members before acquiring a lease and records exact refusal in the original Run (3077.1571ms)
✔ all incompatible members reject with no network request, no lease, preserved Task and explicit non-dispatched ledger (2264.458ms)
✔ explicit synthetic tool probe verifies only actual matching call, never executes it and loses verification on revision changes (1487.4726ms)
✔ plain text probe cannot verify tools; wrong tool nonce and client supplied evidence cannot claim verified support (1360.6443ms)
✔ a disabled connection invalidates an in-flight probe even when toggle mutates its original registry object (1289.9408ms)
✔ selected connection owns the actual endpoint, credential, wire model and mismatch even when legacy is configured for GPT (5480.0036ms)
✔ budget-only API and demo/real selection preserve routes, reject injected endpoint fields and publish authoritative selection without keys (3221.3771ms)
✔ another session cannot change default model, mode or budgets while a request is active (4415.7861ms)
✔ HTTP download preserves original malformed SSE bytes and artifact SHA instead of decoding invalid UTF8 (2323.1673ms)
✔ terminated SSE adapts empty optional tool arguments, retains the exact wire bytes and executes the validated batch once (4627.2122ms)
✔ reported SSE usage survives rejected required arguments "" without dispatching any tool (2336.8455ms)
✔ reported SSE usage survives rejected required arguments "{\"path\":" without dispatching any tool (2459.9344ms)
✔ reported SSE usage survives rejected required arguments "null" without dispatching any tool (1858.8257ms)
✔ a stream without DONE retains reported usage but cannot adapt or execute even an optional tool (2812.9347ms)
✔ a timeout after reasoning-only fragments creates one unknown receipt and never invents reported tokens (2655.1755ms)
✔ real Chat SSE carries UTF8 and split tool JSON; Runtime executes only after DONE, stores partial progress and original usage (4281.186ms)
✔ Responses typed SSE validates original item identity, terminal data, raw usage and tool result pair (4219.2483ms)
✔ incomplete network stream retains unknown attempt and never automatically switches pool or dispatches fragment tools (2668.9743ms)
✔ stopping an active HTTP stream closes only original request and keeps partial text without executing tools (2836.7206ms)
✔ explicit streaming probe uses actual SSE despite unknown declaration and binds its own original attempt (1142.2169ms)
✔ coalesced initial SSE events publish the latest partial state even when the upstream stays idle (3302.4441ms)
✔ stream timeout closes the original HTTP response, preserves redacted capture and never retries (1750.6196ms)
✔ stream idle timeout preserves incomplete tool arguments, records unknown use and never fails over (1806.6599ms)
✔ SSE accepts CR-only final event boundaries and retains exact invalid UTF8 bytes without dispatch (0.7877ms)
▶ stream parser rejects truncation, finish length, malformed JSON, duplicate sequence, changed final item and oversized data
✔ reject 0 (0.2103ms)
✔ reject 1 (0.1716ms)
✔ reject 2 (0.1345ms)
✔ reject 3 (4.9131ms)
✔ reject 4 (0.1627ms)
✔ reject 5 (0.4196ms)
✔ stream parser rejects truncation, finish length, malformed JSON, duplicate sequence, changed final item and oversized data (6.4868ms)
✔ compiled host retains interrupted HTTP 200 and explicitly resumes after the recorded cooldown (10151.114ms)
✔ compiled host retains interrupted HTTP 503 and explicitly resumes after the recorded cooldown (11048.8561ms)
✔ owned IPv4 service works after a failed IPv6 candidate without a second model POST or global DNS change (26.0676ms)
✔ environment HTTP proxy tunnels the owned request and NO_PROXY sends a separate direct request (11.2377ms)
✔ original connect timeout has a bounded nested cause and does not masquerade as response generation timeout (0.7068ms)
✔ POST accepted then reset is unknown, is charged once and never automatically retried (3.585ms)
✔ partial HTTP bodies retain original request id and bytes, and an unknown pool attempt never switches model (3271.2403ms)
✔ response wait timeout after an accepted POST is unknown; cancellation closes the owned socket and permits explicit reuse (137.3501ms)
✔ older configurations receive bounded wait defaults without changing the original object (1.0414ms)
✔ active stream survives sixty seconds; byte activity resets idle but cannot extend total deadline (0.7902ms)
✔ silent stream has its own accurate deadline; caller cancellation and cleanup preserve their cause (1.4219ms)
✔ waiting for headers is bounded independently of stream-idle time (0.2262ms)
✔ wait settings persist, pin new runs and leave model pool, credentials and old run snapshots intact (1788.8645ms)
✔ plain endpoint checks only TCP, records family coverage and never reads model credentials or writes routing/health/attempts (1772.3694ms)
✔ local DNS failure is separate from TCP/TLS and arbitrary input, stale revision and TLS overrides are rejected (1171.0842ms)
✔ refused TCP socket preserves its stage and cannot masquerade as a TLS or API response (1892.2189ms)
✔ TLS validation succeeds for a trusted owned certificate and rejects the same untrusted certificate (2025.0288ms)
✔ a TCP peer that never answers TLS is bounded, and cancellation waits for owned sockets to close (2390.9122ms)
✔ a stalled lookup is cancellable and the overall deadline bounds DNS without late socket work (1605.6904ms)
✔ dual-stack fallback and explicit family results remain separate and never change global DNS defaults (1109.9428ms)
✔ proxy settings are reported as scope only; diagnostics do not contact the proxy or leak its credentials (1573.2193ms)
✔ history is bounded, survives reopening, marks interrupted work and reports changed revisions without rewriting frozen targets (1631.9412ms)
✔ HTTP API, host activity, configuration edits, model start and upgrade honor the same active diagnostic and stop operation (1592.3579ms)
✔ provider shutdown cancels diagnostics and permits explicit reuse after a failed host stop (1668.5923ms)
✔ unchanged proposals cannot hide repeated original-log reads or identical session inspections (10437.8114ms)
✔ reusing a verified proof never resets identical-state detection or creates a new proof (7463.3668ms)
✔ coverage merges adjacent pages, separates changed sources and never claims old Run coverage in a continuation (0.7394ms)
✔ paged inspection finds a corrupted old artifact outside the recent fifty and retains original evidence without model/tool execution (4877.8371ms)
✔ inspection rejects ambiguous/out-of-range pagination and reports unknown tool/task identity without resolving it (644.4409ms)
✔ original source locks all 60 byte-identical files including modules outside the nine-file prompt experiment (1058.6639ms)
✔ a new original executable or missing license cannot silently join the fixed trusted-source profile (957.5441ms)
✔ original mode has a separate source identity and does not accept retry or privilege fields (673.1169ms)
✔ actual original model input requires the complete postPrompt once in system context and cannot borrow a user or tool copy (0.297ms)
✔ Windows directory permissions normalize overlapping grants without allowing an unrelated sibling (861.6801ms)
✔ durability broker rejects foreign authentication, outside paths and wrong file identity before real fsync (1230.1035ms)
✔ exact patches preserve BOM/CRLF, unicode and all unselected bytes and back up the original on apply (906.3343ms)
✔ several patches are matched against one original snapshot in any input order and keep LF (943.5005ms)
✔ mixed line endings require exact anchors and preserve untouched formatting (915.4535ms)
✔ ambiguous, overlapping, missing and malformed patch anchors never create a proposal or write (906.741ms)
✔ stale original hashes and changes before apply preserve owner edits and mark conflicts (1271.8908ms)
✔ no-op BOM patches/full content produce no change and replacements cannot split Unicode bytes (920.4442ms)
✔ real owned Vite starts only after approval, records OS ownership and live changes, and stops its tree (9971.1959ms)
✔ occupied ports and approval-time conflicts never take over an unrelated listener (6490.8212ms)
✔ background preview final-log failure retains unknown evidence without an unhandled startup promise (9002.8956ms)
✔ stale source, tampered plans, wrong ownership and arbitrary scripts cannot dispatch a preview (6771.4419ms)
✔ restart retains original unknown preview, blocks mutation, reconciles absent identity without replay (7382.4105ms)
✔ an actual shell command keeps Node and UTF-8 output while omitting host-only credentials (897.665ms)
✔ real npm tests and build do not inherit arbitrary host credentials or Node injection settings (2626.3606ms)
✔ real Python tests and compilation keep their sealed inputs without host credentials (2665.2012ms)
✔ a process exiting between the Windows snapshot and owner query is absent only after a fresh query (1892.0245ms)
✔ owner observation errors remain errors while a process is present, replaced, or cannot be rechecked (1638.4072ms)
✔ cache controls require desktop authority, exact inputs and owned project identity while preserving task evidence (2759.144ms)
✔ cache enablement and derived reuse survive a real host restart without becoming task recovery state (4240.2963ms)
✔ persistent parsing is project/SHA/config bound and produces identical semantic indexes after restart (1250.3785ms)
warning: in the working copy of 'README.md', LF will be replaced by CRLF the next time Git touches it
warning: in the working copy of 'main.ts', LF will be replaced by CRLF the next time Git touches it
warning: in the working copy of 'value.ts', LF will be replaced by CRLF the next time Git touches it
✔ same-size restored-mtime changes, additions, renames and deletions cannot reuse stale entries or import targets (1132.4292ms)
Switched to a new branch 'other'
✔ branch and commit changes invalidate even byte-identical sources (2286.4231ms)
✔ corrupt, malformed and schema-invalid records fall back to fresh parsing (1159.0674ms)
✔ disable and clear persist without touching sources, sessions, configuration or original artifacts (803.1168ms)
✔ cache capacity bounds disk and memory by bytes and LRU project count (1108.7167ms)
✔ incomplete scans do not load or save cached entries; unavailable files leave the current result scope (1007.4926ms)
✔ project and cache junctions cannot expose or delete an outside directory (699.4272ms)
✔ concurrent file changes refuse a snapshot and clear during pending work cannot repopulate it (776.7454ms)
✔ file access mode changes invalidate parsed material even when bytes are unchanged (805.5765ms)
✔ tracked directory replaced by a junction invalidates access scope and never uses its old cache (1718.9754ms)
✔ JS/TS/React AST indexes actual declarations, exports, imports and verified source lines (574.4854ms)
✔ default expressions, named default exports and declared functions retain their AST evidence (852.5179ms)
✔ incremental refresh invalidates changed/deleted symbols and recalculates reused import targets (1137.1483ms)
✔ Python candidates exclude comments and multiline strings and declare limited evidence (567.4408ms)
✔ indexes isolate project identity and root and evict bounded caches without sharing results (838.1146ms)
✔ Git ignore, credential paths and links stay outside the index and no project files are executed (1286.3987ms)
✔ malformed, binary, non-UTF8 and bounded files expose partial coverage while good files remain searchable (742.0061ms)
✔ index query rejects invalid limits/types and does not imply reference or alias resolution (892.2226ms)
✔ read-only engineering inspection disables configured Git clean filters and preserves index bytes (1780.7366ms)
✔ Python multiline declarations retain candidate scope without interpreting expressions as declarations (932.2723ms)
✔ TypeScript module candidates support import-equals and MTS/CTS while opaque imports stay explicit (821.4106ms)
✔ actual HTTP requests reuse only this project source-checked memory as redacted data and preserve sent evidence after archive/restart (2365.4897ms)
✔ actual HTTP request excludes archived, timed-out and foreign-project memories (4107.9777ms)
✔ actual next HTTP request removes expired memory after a tool observes a source change without replaying that tool (3381.8366ms)
✔ incomplete inventory and excluded-file SHA changes cannot enter actual HTTP memory context or inspect_session snapshots (2891.4578ms)
✔ corrupt original artifact expires its memory and cannot reach an HTTP model request (2545.2057ms)
✔ memory JSON is bounded in escaped UTF-8 bytes and entry count, skips oversized entries and retains complete multilingual values (3850.7869ms)
✔ optional memory yields to the complete goal and original code evidence under the whole-request budget (1321.9466ms)
✔ changed sealed memory artifacts or binding fields fail verification while the original request remains immutable (2713.3749ms)
✔ a malformed project declaration omits optional memory while the repair Task still reaches the owned HTTP service (2504.3667ms)
✔ an expiration persistence failure prevents dispatch and is not hidden as an optional scan failure (2081.7095ms)
✔ project memories persist across restart, keep source provenance, and are visible across tasks (1581.3089ms)
✔ source changes expire current project memories and keep the original evidence (988.1028ms)
✔ incomplete source scans mark current memories unverified without expiring or reusing them (1081.9321ms)
✔ explicit file provenance is rechecked even when the inventory excludes that path (1090.0023ms)
✔ execution snapshots do not expose project memories before their source scope is validated (893.2578ms)
✔ project memory records stay isolated and archive requires the owning session (983.2993ms)
✔ project memory HTTP API exposes read-only source status and explicit writes (2590.8453ms)
✔ two Purge source versions lock nine original inputs without executing JavaScript or postinstall and remain disabled by default (799.786ms)
✔ incomplete, unsupported, relative and oversized sources cannot enable the experiment (932.5911ms)
✔ invalid source fingerprints and invented privilege or retry fields fail before any parent Task is created (2372.1485ms)
✔ source edits between detection and plan are rejected without creating a parent Task (1825.5434ms)
✔ source edits after sealing cannot launch a child and do not reset its cumulative budget (10170.8763ms)
✔ literal experiment text is sealed into the parent, worker and each new Run while actual host approvals and saved results remain required (30184.3953ms)
✔ source changes after a model response stop before any native tool is accepted, retaining the actual usage once (11464.9125ms)
✔ source changes while approval is pending invalidate the old permission before an engineering check starts (15554.2094ms)
✔ altered text, interpolation, section order and policy flags cannot reuse a sealed experiment (2056.7091ms)
✔ the observed SDK system-message layout and the explicit system field both require exactly one effective section, never a user or tool copy (2460.2151ms)
✔ a missing or duplicated experimental section in the actual input is rejected before any model dispatch (21685.1241ms)
✔ experimental prompt overhead still participates in the original complete-input budget guard (11188.3772ms)
✔ source detection HTTP remains local-authenticated and read-only with no configured model requests (1756.1074ms)
✔ Python HTTP plans keep npm defaults, require owned approval and preserve actual results across restart (10722.5431ms)
✔ an isolated three-module Python Task repairs, checks and delivers a source-bound Git report without extra model calls (37971.4573ms)
✔ Python static inspection does not import project code or probe a project interpreter before approval (1068.6452ms)
✔ a real three-module Python failure, reviewed repairs, tests and compilation retain exact structured evidence (4459.5359ms)
✔ unittest counts repeated execution ordinals and subtest skips rather than deduplicating display names (2773.6898ms)
✔ unittest empty, skipped, expected failure and class initialization errors never verify a task (8534.4187ms)
✔ stdout claims cannot replace actual unittest results and an early exit cannot forge a successful collector (3003.4091ms)
✔ compilation reads actual Python source bytes without executing top-level project code (2998.329ms)
✔ Node and Python mixed projects retain the existing npm default and require an explicit Python selection (2964.5817ms)
✔ changed project interpreters and forged Python plan bindings execute no check (1773.9666ms)
✔ malformed TOML and invalid Python test paths are reported without expanding npm check scope (1443.8878ms)
✔ stopping a real Python check terminates its child and keeps the completed result as cancelled (2053.4187ms)
✔ real pytest pass, failure, skip, collection errors and teardown errors use collected case outcomes (14172.7228ms)
✔ a real project virtual environment is detected statically and its actual version is recorded only during the approved check (9631.9855ms)
✔ a missing interpreter and repeated pytest item objects cannot create a false verification pass (2871.486ms)
✔ a mixed turn with new bytes and three repeated modules advances, while repeat-only and independent no-op guards still pause (0.9812ms)
✔ runtime sends the next real HTTP request after a W-shaped mixed-progress turn instead of stopping on its trailing repeats (6978.5763ms)
✔ continuation reads advance through historical pages without a false pause while repeated current-Run pages still pause (1.2555ms)
✔ a new tool page replaces old latest-feedback space and still obeys its escaped protocol budget (0.3815ms)
✔ budgeted artifact pages count escaped protocol bytes and reconstruct original multilingual JSON without losing characters (8.5389ms)
✔ known artifact coverage detects semantic repeats despite changing limits, and advancing pages or direct evidence restore progress (0.3445ms)
✔ actual large artifact reads are present in the next bounded HTTP request and next_offset progresses without redoing tools (5037.9859ms)
✔ a real repeated artifact loop pauses the same Task after recorded results, survives reload and only resumes explicitly (7763.4675ms)
✔ actual HTTP continuation after reload sends distinct historically covered pages without false no_progress or replay (8459.2398ms)
✔ continuations distinguish unread source, later original line ranges and already returned result wrappers (772.3921ms)
✔ pure-file batch returns unused small-page space while mixed and historical allocation stays conservative (0.3155ms)
✔ v3 pools mixed file/log reads, serves small sources first and leaves historical/mutating batches unchanged (0.2755ms)
✔ v3 W-shaped mixed reading returns complete small modules and substantially larger first test pages in the next bounded request (4609.7057ms)
✔ observed upstream high input pauses before a new request, preserves reported usage and distinguishes the limiting ledger (694.2727ms)
✔ chat-completions: complete actual Unicode/CRLF pages and small source survive the next bounded wire request (4262.561ms)
✔ responses: complete actual Unicode/CRLF pages and small source survive the next bounded wire request (4505.0366ms)
✔ anthropic-messages: complete actual Unicode/CRLF pages and small source survive the next bounded wire request (4450.3231ms)
✔ brand and custom-compatible profiles send native reasoning controls per protocol without changing output budgets (2.1295ms)
✔ opaque Chat reasoning returns only to its original endpoint, model, connection and credential revision; bytes count against the actual request (0.7881ms)
✔ saved profile changes preserve credential references and old frozen Run settings, including the legacy connection after restart (1289.4425ms)
✔ DeepSeek JSON reasoning survives real HTTP tool rounds and restart without entering visible assistant text (6512.1045ms)
✔ DeepSeek SSE reasoning survives real HTTP tool rounds and restart without entering visible assistant text (5445.6155ms)
✔ malformed or post-terminal Chat reasoning is rejected with the raw stream retained (1.7655ms)
✔ connection import synchronizes profile, native fields and declared abilities without editing the source or manufacturing verification (735.8437ms)
✔ compiled HTTP persists profile and reasoning choices for every protocol and rejects incompatible/stale edits atomically (2312.9042ms)
✔ an installed app outside the source drive resolves registered SDK and Purge through actual input verification (1033.6513ms)
✔ standalone registration resolves the upstream metadata directory and missing upstream falls back to the other registered source (865.3966ms)
✔ registered hints keep explicit roots, environment and existing app-relative candidates in their original priority (940.1884ms)
✔ unknown IDs, duplicate registered IDs, wrong versions and relative or inconsistent metadata are not lookup candidates (833.7125ms)
✔ oversized, malformed, invalid UTF-8 and non-file registry inputs cannot silently enable a source (811.6462ms)
✔ a linked registry directory is rejected without reading or deleting its external source (1028.271ms)
✔ matching research declarations do not replace actual installed version and critical-source checks (792.0323ms)
✔ runtime guides map archived evidence while all original reports and successful/failed source bytes remain independently verifiable (1096.8248ms)
✔ missing required research reports fail packaging rather than creating a successful placeholder (731.2657ms)
✔ full verification reports only verified bytes, complete declared scope and actual directory inventory; changed/deleted files still fail (772.2312ms)
✔ preflight and activation continue one monotonic operation bound to installer hash; terminal replay and changed origin are rejected (8150.2345ms)
✔ preflight failures and fixed-candidate mismatches retain the actual reason without publishing a pointer (4633.03ms)
✔ cancel finishes only the original ready preparation, preserves data and pointer and refuses replay (6792.3867ms)
✔ the installation lock rejects concurrent same-id operations before they can overwrite original progress (4854.5545ms)
✔ original post-publication failure is linked to its rollback receipt and every original byte survives (4836.0162ms)
✔ a candidate mutation at the publication boundary still fails even after an earlier successful full validation (8329.2865ms)
✔ public progress marks a lost process unknown without editing its original record or claiming completion (6557.3328ms)
✔ throwing console/UI progress observers cannot decide the installation result (4461.1598ms)
✔ a progress-only write failure after committed publication returns the authoritative receipt and a warning (4457.4661ms)
✔ browser and MCP approvals use actual tool categories; mobile exposes only category while keeping original fingerprint (655.9965ms)
✔ the compiled runtime version constant tracks package.json so a release bump cannot ship a stale served version (8.0338ms)
✔ context APIs retain exact request identity, paginate archived groups and reject foreign sessions/runs (2936.7272ms)
✔ delivery reports verify request body, sealed plan and original archived bytes, and show corruption as needing verification (2189.336ms)
✔ native seed framing reserves capacity before optional history without changing the selected window (914.3936ms)
✔ direct code evidence remains selected while recent historical-index browsing is archived within the complete request budget (1930.4436ms)
✔ one long user turn becomes bounded complete groups with exact UTF-8 history and recent JSON arguments (3313.1016ms)
✔ tool definitions, exact system bytes, output reserve and remaining Run/Task budget are accounted before requests (520.7559ms)
✔ oversized required user instructions pause intact and can be explicitly resumed after increasing the window (1574.8652ms)
✔ all task instructions, constraints and acceptance remain in the runtime system instead of an eight-item or byte prefix (1483.0234ms)
✔ a newest manageable multi-tool group stays exact even beyond the working target; oversized earlier groups are archived (1233.6982ms)
✔ invalid orphan result is rejected before archive or network preparation (673.1078ms)
✔ invalid missing result is rejected before archive or network preparation (600.0548ms)
✔ invalid wrong result is rejected before archive or network preparation (532.8327ms)
✔ invalid duplicate call is rejected before archive or network preparation (1057.8139ms)
✔ a group larger than the artifact limit uses verified Unicode-safe chunks without replacing tool JSON (1833.0448ms)
✔ changed or missing archived evidence stops reuse and is never silently recreated (1102.622ms)
✔ transaction failure prevents publishing a plan and any corresponding model request (1195.1979ms)
✔ the Provider rejects body mutation after a sealed plan before calling HTTP (1017.4837ms)
✔ HTTP failure and explicit same-Task resume reuse exact completed results and request hashes without repeating a command (9787.1651ms)
✔ a real process exit after plan persistence marks interruption and never issues an automatic model request (1562.7772ms)
✔ local draft preserves exact original ranges and duplicate source instances without inferring outputs or accepting facts (857.2345ms)
✔ review keeps unknown blocking, references distinct from output and boundaries distinct from compliance evidence (808.3288ms)
✔ draft adoption must account for every original range and does not fill paths from a reference (1297.762ms)
✔ source and requirement changes, other Tasks and tampered draft bytes are rejected without losing originals (1553.0997ms)
✔ partial explicit source review leaves the other source unreviewed while old file evidence remains valid (1461.0198ms)
✔ bounded draft does not claim full source coverage and cannot acknowledge an omitted remainder (1108.3715ms)
✔ draft HTTP preserves its original artifact across restart and rejects stale scope and source versions (5027.823ms)
✔ requirements HTTP binds selection, enforces optimistic revisions, restores original proofs and leaves provider history untouched (5300.1372ms)
✔ research evidence binds actual original bytes; model completion is not user acceptance (1588.3508ms)
✔ same text with distinct user source instances stays distinct; an appended requirement only invalidates its own revision (1777.7912ms)
✔ foreign project, old Task and fabricated evidence strings are rejected with preserved failed proofs (2379.6887ms)
✔ source changes invalidate prior proofs and cannot certify a read that was already stale at binding (2411.108ms)
✔ missing or tampered artifacts never become evidence and original failures remain after recovery (1821.6784ms)
✔ checks omitted by explicit agreement are reported as not executed and never passed (950.4131ms)
✔ user confirmation stays separate from engineering evidence, is bounded to current inputs and cannot be silently converted (1663.4172ms)
✔ existing multi-file results keep provenance and invalidate only evidence whose files changed (1909.8761ms)
✔ unrelated additions preserve the checked file facts but remain visible as unreviewed user sources (1719.9621ms)
✔ actual Git preview and publication recheck changed requirement bindings without touching HEAD or the owner index (7952.8631ms)
✔ no-test file application and manual completion keep pending user requirements for explicit review (2191.6839ms)
✔ revision races, forged source ranges, evidence fields and deletion cannot overwrite sealed requirements (1109.5206ms)
✔ requirement binding prevents completion and Git acceptance after a review changes; legacy tasks stay explicit (3296.6613ms)
✔ deep Windows data directories keep REA output within the external path allowance and retain owned evidence (2634.6122ms)
✔ user research registration creates an owned Task without model requests and seals the exact artifact identity (1143.0797ms)
✔ complete research binds original Evidence and Bundle, pages exact fields and reuses verified results after reload (2227.6075ms)
✔ artifact changes before analysis refuse dispatch and changes during analysis preserve unverified original responses (3325.1072ms)
✔ owned cancellation closes the connection, concurrent and automatic retries are refused, explicit new analysis preserves history (2349.8766ms)
✔ a dispatched transport timeout stays unknown and reload never launches or kills a saved PID (1655.8657ms)
✔ corrupt Evidence parts and changed binding records are refused instead of silently re-analysed (1663.3205ms)
✔ research documents over the single-artifact limit preserve exact Unicode bytes with owned verifiable parts (992.5582ms)
✔ research tools are only sent for registered Task artifacts and actual HTTP uses the sealed tool contract without replay (8584.1709ms)
✔ file identity rejects directory and invalid formats without running a target (744.2352ms)
✔ a stop from the event loop interrupts immediate model replies before later tools (2344.9672ms)
✔ a stop after the first tool preserves a complete interrupted multi-call protocol group (2982.9717ms)
✔ the compatible HTTP loop preserves call IDs, validates errors and waits before executing commands (7997.5799ms)
✔ denial prevents command execution and produces a paired denied result (3054.044ms)
✔ stopping during approval interrupts pending calls without executing or replaying them (2395.1701ms)
✔ stopping a running fixture command terminates its own process tree and retains uncertain state (2761.7679ms)
✔ command output decodes split UTF-8 and excludes the workbench key from child environment (1201.841ms)
✔ model-list testing uses a tentative key without changing the saved connection and redacts service errors (6.264ms)
✔ truncated or repeated model calls fail before execution; tool input validation rejects unknown capabilities (4385.1706ms)
✔ request projection keeps exact tool pairs and complete recent user instructions while preserving disk history (1483.3135ms)
✔ local skill revisions require explicit enable, retain original text, survive restart and stay within the selected project (1384.3182ms)
✔ registration rejects non-SKILL paths, oversized/non-UTF8 files, duplicate sources and wrong field shapes without running project code (989.4471ms)
✔ actual HTTP models receive only enabled current-project metadata, then a redacted owned skill page with complete tool pairing (4443.365ms)
✔ source edits after a catalog request prevent the read and appear as changed on the next request without replay (3702.3702ms)
✔ disabled and foreign skills cannot be read by guessing their IDs from another project (3942.4423ms)
✔ skill catalog takes remaining escaped JSON budget and cannot evict the user goal or current code (1424.9501ms)
✔ multilingual skill body pages fit actual tool JSON bytes and every page uses the same immutable artifact (5648.4984ms)
✔ registry failures roll back enable and refresh state; corrupted snapshots block requests before network (2197.9955ms)
✔ saved skill catalogs retain SHA-bound historical evidence and reject corrupt catalog blobs (2449.2245ms)
✔ HTTP skill administration enforces project scope, revision checks and idle mutation without model calls (3331.1797ms)
✔ JSON state keeps an atomic backup and preserves corrupt primary files (809.2887ms)
✔ tool-pair recovery fills missing results and drops orphan or duplicate results (0.431ms)
✔ restart pauses unfinished work, retains proposals and never replays a command (957.6528ms)
✔ public snapshots carry increasing revisions without the private provider transcript or config key (901.6242ms)
✔ a checkpoint seals all task requirements and original results without a model request and explicit continuation keeps the same Task without replay (5511.4042ms)
✔ checkpoint sources, requirements, plan and original tool changes block continuation and leave original checkpoint bytes intact (4018.8214ms)
✔ unknown tool effects and pending proposals stay blocked in recovery packages rather than being changed into completed work (4760.6143ms)
✔ checkpoint and referenced result corruption are rejected before dispatch and historical blobs are never rebuilt (4044.3839ms)
✔ checkpoint ownership, exact confirmation and concurrent operations reject foreign or stale resumes (3941.3642ms)
✔ Git identity changes and altered plan proofs cannot be accepted as the sealed recovery state (4388.0183ms)
✔ checkpoint records survive SQLite reopen and do not trigger automatic model or tool replay (3907.6294ms)
✔ incomplete scans and in-progress project activity refuse checkpoint creation while private credentials are redacted from saved requirements (4195.4467ms)
✔ oversized recovery evidence is refused before reading or publishing the checkpoint (3570.3359ms)
✔ compiled checkpoint HTTP APIs bind current Task/Run, isolate sessions and only resume on an explicit matching confirmation (8917.9589ms)
✔ actual process exit before checkpoint commit retains only durable records without replay (3922.8998ms)
✔ actual process exit after checkpoint commit retains only durable records without replay (4202.3818ms)
✔ a real two-reply failure replay reaches request three at 16K with complete goal, original sources and sealed tool evidence (9967.5517ms)
✔ request deduplication retains long original intent, extra requirements, constraints and acceptance without changing the stored contract (2682.1344ms)
✔ only the routine context avoids plan echoes; diagnostic API and requests without an included plan keep original observations (891.2611ms)
✔ operational questions are diagnostic while explicit repair requests remain execution instructions (1.3395ms)
✔ an explicit prohibition on running engineering checks is not interpreted as a requirement to run them (1.6106ms)
✔ file delivery requires a positive instruction, not saved evidence or a prohibited write (0.2846ms)
✔ an unsubstantiated text claim cannot complete a requested file delivery (1836.5775ms)
✔ HTML without tests ends at reviewed file delivery and never offers automatic rechecking (3551.3439ms)
✔ diagnostic answers retain the unfinished task, limits and check result, and enforce read-only tools (5250.2505ms)
✔ review continuation is a system action and adds no testing demand or fake user instruction (3592.9272ms)
✔ a provider cannot execute tests against an explicit no-test task even if it ignores advertised tools (2568.7412ms)
✔ a budget pause becomes visible only after its input snapshot can be sealed with the original Run (2236.0594ms)
✔ alternating duplicate file reads and listing files still pauses without treating new pages as duplicates (20495.5581ms)
✔ task-plan HTTP mutations bind current Task/Run and retain original evidence through real host restart (5477.6094ms)
✔ concurrent stale HTTP plan revisions cannot replace conditions or attach proofs to a different plan (3393.3297ms)
✔ a real escaped long-file tool message keeps its proof identity within a tight request budget (1162.0607ms)
✔ task plans reject fabricated completion, invalid scopes and cyclic or missing dependencies (0.5295ms)
✔ sealed plans retain goal, contract hash and original steps through SQLite reopen without executing tools (1044.9042ms)
✔ proofs require original owned reads, honor dependencies and reject altered invocation arguments and bytes (1451.4756ms)
✔ new user requirements invalidate current proofs and reseal the same original operations in a new revision (1390.9966ms)
✔ applied change-set evidence binds reviewed current files and invalidates dependent steps after outside edits (2037.344ms)
✔ actual npm checks are verified by original counted logs and current source, with failure history retained (7554.3918ms)
✔ real desktop and mobile browser proofs bind required actions and PNGs, rejecting source changes, a later failed plan and changed builds (12649.6005ms)
✔ Git delivery blocks incomplete plans and changed preview bindings, preserving HEAD and owner index before a fresh verified commit (14120.1496ms)
✔ the runtime supplies owned read IDs and completes a plan only after original step evidence verification (3978.2922ms)
✔ premature model completion pauses the original Task and resume preserves the existing plan without replay (4434.9304ms)
✔ plan operations participate in project activity and orderly shutdown (689.6387ms)
✔ actual process exit before plan save restores only committed plan evidence without replay (1277.6037ms)
✔ actual process exit after plan save restores only committed plan evidence without replay (1694.2527ms)
✔ actual process exit before proof save restores only committed plan evidence without replay (1734.6071ms)
✔ actual process exit after proof save restores only committed plan evidence without replay (1221.621ms)
✔ the latest oversized session result reaches the next request together with already read source modules (2078.8419ms)
✔ every call in a large latest group keeps its actual outcome, including denied and unknown results (1509.301ms)
✔ a small complete latest tool group is sent once and needs no duplicate feedback view (1237.2728ms)
✔ real HTTP repeated state inspection pauses after three identical engineering outcomes and preserves tool pairing (3987.0436ms)
✔ fixed normal exploration, pagination, SHA changes and repeated ranges are external observations, not a score (1.4278ms)
✔ read-only research is valid work; model summaries and unchanged timestamped listings do not become engineering evidence (0.3948ms)
✔ new permissions, network and broken tools are failure evidence; repeated invalid inputs remain candidates (0.4676ms)
✔ waiting for original long checks, review and approvals stays waiting across repeated observations (0.5147ms)
✔ baseline facts, real repairs, current-source check results and requirement evidence have distinct growth (0.2107ms)
✔ bounded memory keeps unknown instead of claiming an unretained range is repeated; Task/Run instances cannot mix (12.2181ms)
✔ actual runtime records growing pages then a repeated page without changing model rounds, stop rules or original tool pairs (4414.5593ms)
✔ actual invalid tool calls repeat under the original step budget; observation never adds an early stop (3974.0812ms)
✔ actual command approval is observed as waiting and denial stays a permission failure (2891.84ms)
✔ network errors remain model failures with shadow failure evidence, never a fabricated no-progress cause (4053.2774ms)
✔ observer receipt failure cannot change tool results, model budget or completion (1901.4391ms)
✔ a later receipt recovers after an observation write failure without misaligning retained source receipts (2949.0179ms)
✔ real applied files, counted checks and current requirement proofs grow after an existing-facts baseline (4482.7316ms)
✔ receipt retention stays aligned at the cap and a foreign step cannot enter the Run (3273.7797ms)
✔ an actual approved long check records its owned running job as waiting before the original counted result (7340.6226ms)
✔ a live SQLite host with the actual 1.9 artifact object upgrades, keeps its original blobs and verifies real browser PNG bytes (4727.4445ms)
✔ failed upgrade restores the original artifact object, provider and request handler (2458.1617ms)
✔ claiming the same runtime owner twice preserves its original metadata and every database row (890.0951ms)
✔ sessions without controlled-engine history retain their prior export shape and never expose another session record (1232.3735ms)
✔ live workers, controlled drivers, or MCP operations refuse an upgrade before changing handlers, managers or database rows (641.098ms)
✔ an idle upgrade loads the newly introduced engine record map, resets obsolete managers and preserves original rows, sessions and provider objects (677.3722ms)
✔ an installation failure restores old managers, original record maps, handlers and all existing database rows (1637.9706ms)
✔ the owned controlled driver remains an active session and project even after outer task activity has ended (799.1373ms)
✔ release and upgrade help or invalid arguments have no file or network setup side effects in an independent directory (1171.0513ms)
✔ hot upgrade adopts a retained virtual data root without reopening SQLite or changing rows, Provider or routing (1628.3253ms)
✔ failed adoption restores the original virtual data-root name and original open database (857.4654ms)
✔ hot upgrade refreshes retained model and importer methods while preserving credential and routing objects (917.1708ms)
✔ hot upgrade refuses active probes and restores retained model methods if installation fails (1021.8501ms)
✖ the upgrade compatibility gate includes the served current version and the previous formal patch (2.5575ms)
✖ a directly started compiled host upgrades twice with the same PID, sessions, applied bytes and volatile provider (2970.6116ms)
✖ the running 0.1.0 process upgrades in place and retains its project, session, files and volatile model connection (5814.7735ms)
✔ HTTP worker identities require a current sealed plan, reject overrides and close unused plans without touching source (17706.3101ms)
✔ HTTP unknown recovery only accepts the original evidence SHA and preserves changed source for explicit retention (24125.8188ms)
✔ verified isolated outputs form a new parent batch, rerun real checks and commit only reviewed source (65058.7726ms)
✔ overlapping worker paths are explicit conflicts even if each independent task checks successfully (40919.6233ms)
✔ source edits after isolated work preserve the owner file and invalidate collection without proposing a merge (37654.7434ms)
✔ changed child bytes after parent preview block the entire batch before any parent write (53570.0964ms)
✔ corrupt original check logs cannot become collected outputs or a parent review (43109.2794ms)
✔ stopping a parent accounts late responses and executes none of their proposed tools (18641.929ms)
✔ unreported interrupted requests reserve estimated usage and the total deadline cannot reset on resume (18348.2826ms)
✔ a deadline before worker request dispatch records no invented usage and cannot reset on resume (16917.04ms)
✔ a response exceeding the cumulative token allocation stops before registering or executing its tools (19186.9652ms)
✔ cumulative tool allocation rejects the entire returned batch before any tool side effect (18346.6963ms)
✔ single response 16 tool ceiling rejects the entire returned batch before any tool side effect (17884.3638ms)
✔ consumed request steps remain consumed across explicit worker resumes (20658.5195ms)
✔ a saved completed response with missing usage is conservatively accounted exactly once during recovery (15899.3434ms)
✔ an unexecuted second workspace plan resumes its original identity without recreating the first checkout (18153.7599ms)
✔ a verified interrupted parent proposal resumes sealing without duplicating or replacing its original change (63627.3686ms)
✔ recovery attaches a sealed parent batch saved before its group reference without creating another batch (54543.8981ms)
✔ explicitly retaining an unexecuted parent merge cancels its batch and frees the next parent goal (50609.7543ms)
✔ unknown collaboration blocks ordinary proposals but permits only original batch recovery and rollback (55320.2331ms)
✔ after a restart the batch application lazily reinstates the worker result guard before any parent write (49963.0208ms)
✔ unknown original tool inputs must match before acknowledgement, and reviewed checkpoint IDs preserve the unknown fact (19385.0839ms)
✔ a parent stop observes and waits for a manually approved child check and its real descendant process (17525.5781ms)
✔ a real host crash preserves unknown checks, refuses a still living owned PID and never replays a model or tool (31004.8753ms)
✔ a persisted child initialization resumes the original control run after a failure before its worker record (18353.0462ms)
✔ failed request intent persistence never dispatches a worker request or changes source (17379.579ms)
✔ failed usage persistence rolls back the uncommitted usage and conservatively reserves one unknown request without tools (18225.067ms)
✔ a child file change after observation prevents acknowledgement even when the tool inputs did not change (17065.4101ms)
✔ worker scopes and allocated budgets reject aliases, extra permissions and aggregate overflow (520.2716ms)
✔ two sequential Git workspaces freeze a new parent goal, separate owned tasks and preserve all source bytes (16868.8025ms)
✔ real isolated changes, approvals and npm checks produce original owned outputs, not model completion claims (40735.0601ms)
✔ forged worker tools cannot execute terminal or Git even when absent from the advertised tool list (16180.8984ms)
✔ a completion-only model pauses each worker without fabricated changes or checks (19265.0491ms)
✔ restart retains frozen scope and marks interrupted parent and workers unknown without sending requests (16123.6708ms)
✔ all actually returned same-version fragments reach the bounded request, including overlap and Unicode, with independent original provenance (4642.4665ms)
✔ unread gaps and different range origins remain separate; partial sending names the precise missing already-observed range (939.4852ms)
✔ same-SHA conflicting overlaps and invalid coordinates fail closed, while a new SHA never inherits old fragments (1222.3438ms)
✔ versioned old views keep the original single-page semantics and verification (786.2239ms)
✔ small code and actual failed command stay visible when their original multi-tool group cannot fit (1186.1329ms)
✔ mixed group retains six distinct small file observations with exact multilingual text and bounded request bytes (1780.5298ms)
✔ working observations are bound to the current Task and stop treating a pre-application file revision as current evidence (1570.7003ms)
✔ a sealed observation rejects changed or missing original result artifacts instead of recreating evidence (1006.0475ms)
✔ a real passed-check observation is excluded after a source edit invalidates the check, while its historical sent evidence remains valid (5465.2028ms)
✔ budgeted file pages reconstruct UTF-8 text including long single lines and surrogate boundaries (4.5044ms)
✔ actual HTTP multi-tool reads survive an oversized runtime snapshot and the sealed evidence survives reopening (9009.643ms)
✔ actual HTTP file pagination advances over a long file with each returned page visible in the next request (8386.6515ms)
✔ later mixed history pages cannot displace the actual module sources and complete requirements (7013.3628ms)
✔ a later narrow page of the same file revision cannot evict the wider actual observation (3574.8663ms)
✔ six small source files stay complete across metadata turns even when mandatory evidence exceeds the soft target (3707.3347ms)
✔ historical read-artifact navigation points to the original artifact and next page, never its result wrapper (1155.5957ms)
✔ a 1.6.1 full-provenance sent view remains verifiable after introducing the compact view (1179.9006ms)
✔ actual HTTP requests keep four observed modules after a later large history read and preserve all source bytes (9840.8566ms)
✔ workspace HTTP ownership, reviewed creation and reopen survive restart without changing source files (11129.1854ms)
✔ isolated model Task applies a batch, executes real checks and commits at the Git review boundary with a source-bound report (38802.664ms)
warning: in the working copy of 'src/空 格.mjs', LF will be replaced by CRLF the next time Git touches it
✔ worktree checkout disables hooks, fsmonitor and all filter command variants while honoring built-in EOL conversion (7745.9429ms)
✔ an ignored external file at checkout is observed and never accepted as pristine creation (6284.3279ms)
✔ outside content hidden by assume-unchanged cannot pass workspace verification and blocks further source execution (5470.3047ms)
✔ workspace preparation refuses known shared-repository activity and preserves archived child sessions (6465.2534ms)
✔ compact checkout identities create in deep Unicode source directories while retaining the human branch name (6445.6113ms)
✔ Windows Git metadata path limits reject creation before publishing a branch or changing source bytes (2773.0664ms)
✔ a pending source snapshot participates in shutdown and completes before storage can close (2363.997ms)
✔ source configuration values stay private and a configuration change invalidates a saved review (2812.225ms)
✔ workspace review rejects excluded dependency before any checkout (2795.4688ms)
✔ workspace review rejects large committed blob before any checkout (2352.5125ms)
✔ workspace review rejects case collision before any checkout (2768.6148ms)
✔ workspace review rejects committed symbolic link before any checkout (2619.0046ms)
✔ workspace review rejects external config include before any checkout (1862.636ms)
✔ actual process exit at workspace intent preserves source and reconciles without creation replay (5552.0108ms)
✔ actual process exit at workspace git_started preserves source and reconciles without creation replay (7455.6784ms)
✔ actual process exit at workspace git_finished preserves source and reconciles without creation replay (7481.9311ms)
✔ actual process exit at workspace verifying preserves source and reconciles without creation replay (7020.686ms)
✔ actual process exit at workspace registering preserves source and reconciles without creation replay (8913.6211ms)
✔ actual process exit at workspace project_registered preserves source and reconciles without creation replay (10333.7949ms)
✔ actual process exit at workspace session_registered preserves source and reconciles without creation replay (10599.7311ms)
✔ actual process exit at workspace ready preserves source and reconciles without creation replay (7584.7669ms)
✔ outside changes at project registration remain unknown and preserve the already registered child identity (8823.0339ms)
✔ managed worktree preserves source index and dirty files, opens a distinct project and binds Run identity (7119.6261ms)
✔ stale review, branch/path collisions and forged ownership do not run worktree creation (4396.1365ms)
✔ workspace project verification is read-only and rejects physical Git identity changes (6832.8905ms)
ℹ tests 926
ℹ suites 0
ℹ pass 923
ℹ fail 3
ℹ cancelled 0
ℹ skipped 0
ℹ todo 0
ℹ duration_ms 2083427.6906
✖ failing tests:
test at tests\upgrade.test.mjs:10:1
✖ the upgrade compatibility gate includes the served current version and the previous formal patch (2.5575ms)
AssertionError [ERR_ASSERTION]: The expression evaluated to a falsy value:
assert.ok(SUPPORTED_UPGRADE_VERSIONS.includes(APP_VERSION))
at TestContext.<anonymous> (file:///D:/Ai-project/REA%E9%80%86%E5%90%91/Agent-Workbench/tests/upgrade.test.mjs:11:10)
at Test.runInAsyncScope (node:async_hooks:227:14)
at Test.run (node:internal/test_runner/test:1402:25)
at Test.start (node:internal/test_runner/test:1262:17)
at startSubtestAfterBootstrap (node:internal/test_runner/harness:387:17) {
generatedMessage: true,
code: 'ERR_ASSERTION',
actual: false,
expected: true,
operator: '==',
diff: 'simple'
}
test at tests\upgrade.test.mjs:17:1
✖ a directly started compiled host upgrades twice with the same PID, sessions, applied bytes and volatile provider (2970.6116ms)
Error: This migration accepts a supported existing Workbench.
at upgradeRunning (file:///D:/Ai-project/REA%E9%80%86%E5%90%91/Agent-Workbench/scripts/upgrade-running.mjs:76:84)
at process.processTicksAndRejections (node:internal/process/task_queues:104:5)
at async TestContext.<anonymous> (file:///D:/Ai-project/REA%E9%80%86%E5%90%91/Agent-Workbench/tests/upgrade.test.mjs:31:20)
at async Test.run (node:internal/test_runner/test:1409:7)
at async Test.processPendingSubtests (node:internal/test_runner/test:974:7)
test at tests\upgrade.test.mjs:45:1
✖ the running 0.1.0 process upgrades in place and retains its project, session, files and volatile model connection (5814.7735ms)
Error: This migration accepts a supported existing Workbench.
at upgradeRunning (file:///D:/Ai-project/REA%E9%80%86%E5%90%91/Agent-Workbench/scripts/upgrade-running.mjs:76:84)
at process.processTicksAndRejections (node:internal/process/task_queues:104:5)
at async TestContext.<anonymous> (file:///D:/Ai-project/REA%E9%80%86%E5%90%91/Agent-Workbench/tests/upgrade.test.mjs:97:18)
at async Test.run (node:internal/test_runner/test:1409:7)
at async Test.processPendingSubtests (node:internal/test_runner/test:974:7)
@@ -0,0 +1,10 @@
{
"sourceCommit": "109e51247a47ea4160ac54fe652c745ee810ad58",
"status": "cancelled_scope_expansion",
"reason": "User added upstream catalog, multiple models and manual/automatic routing before packaging; preserve incomplete old-range test results and refreeze final source.",
"completedLines": 264,
"observedFailures": [],
"terminatedOwnedPid": 56056,
"realModelRequests": 0,
"recordedAt": "2026-11-04T16:27:37.0733172Z"
}
@@ -0,0 +1,291 @@
✔ Messages builds native system/tool history, bounds images, and seals a distinct mixed-pool variant (7.5352ms)
✔ Messages URL and native parameter contracts preserve other protocol paths and hard output budget (0.7113ms)
✔ native usage includes both cache buckets once, retains raw and rejects invalid counters (0.1684ms)
✔ real HTTP Messages JSON executes one tool, retains opaque history, actual URL and cache accounting (3768.657ms)
✔ real HTTP Messages SSE executes one tool, retains opaque history, actual URL and cache accounting (3165.2554ms)
✔ Messages api-key authentication is explicit; mismatched model remains blocked with original usage (5.6114ms)
✔ Messages bearer authentication is explicit; mismatched model remains blocked with original usage (4.5747ms)
✔ Messages redirect is not followed, either credential stays at the original server (3.948ms)
✔ strict native SSE reads the nested start model and retains mismatch source without retries (2.8382ms)
✔ a closed native tool block cannot execute before message_stop, then executes exactly once (3031.2422ms)
✔ mixed Chat/Messages pool seals both variants and preserves one completed tool on failover (4108.912ms)
✔ native JSON rejects incomplete stops, duplicate tools and hosted content (0.5529ms)
✔ Messages rejects empty, whitespace and thinking-only answers without rejecting valid text or tool calls (0.393ms)
✔ empty Messages JSON fails once with original response and 6692/0 usage, before verification or tools (2388.731ms)
✔ empty Messages SSE fails once with original response and 6692/0 usage, before verification or tools (2236.7692ms)
▶ strict SSE validates all blocks, terminal, event names, UTF8 and tool JSON before dispatch
✔ reject native stream 0 (1.1658ms)
✔ reject native stream 1 (0.2455ms)
✔ reject native stream 2 (0.1854ms)
✔ reject native stream 3 (0.1828ms)
✔ reject native stream 4 (0.1607ms)
✔ reject native stream 5 (0.1679ms)
✔ reject native stream 6 (0.1496ms)
✔ reject native stream 7 (0.1262ms)
✔ reject native stream 8 (0.1797ms)
✔ strict SSE validates all blocks, terminal, event names, UTF8 and tool JSON before dispatch (3.2387ms)
✔ upstream SSE error retains original type/request ID and initial usage, never retries (4.5943ms)
✔ stop cancels the owned stream and retains partial consumption without running tool or replaying (53.916ms)
✔ saved Messages import preserves auth choice, rejects endpoint duplication, and keeps source file untouched (598.865ms)
✔ large source receipts use cursor-preserving invalidation without filling the event stream or changing the session (5.4902ms)
✔ small event updates retain the current session and actual cursor (0.1333ms)
✔ the bundled local service enforces host, origin, token and JSON object contracts (2596.4617ms)
✔ an in-flight connection probe prevents configuration changes and another probe until its actual result returns (2577.4142ms)
✔ API review applies verified content, detects conflicts, exports sessions and keeps keys in memory (5211.4745ms)
✔ a real bundle restart preserves proposals, interrupts old approvals and requests fresh consent on resume (12037.7562ms)
✔ runtime diagnostics and model-response validation use actual records and preserve the saved connection (2656.2461ms)
✔ persistent event replay is read-only and keeps run/artifact identity across reconnects (8398.5137ms)
✔ real browser HTTP plans, original field pages and PNG bytes preserve ownership and survive host restart (8386.4094ms)
✔ self harness browser tools require approval and feed verified actual page fields back to the original Task (13468.9904ms)
✔ browser host recovery retains unknown root outcome, blocks replay, and requires an explicit owned acknowledgement (3551.089ms)
✔ late HTTP stop of a finished browser job cannot interrupt a later job in the same Task (7395.801ms)
✔ actual failed page blocks source-passed delivery; a new passing page permits a bound commit and preserves both original records (19309.9019ms)
✔ a build-only change after Git review invalidates the actual browser evidence and leaves HEAD and the protected index intact (9628.7245ms)
✔ actual DOM controls include validated unique CSS selectors across repeated parent buttons and escaped labels (7575.0662ms)
✔ browser field coverage distinguishes sections and Run windows while retaining original source identity (0.451ms)
✔ repeated original browser pages pause a live Task without re-opening the browser or breaking tool pairing (5543.4574ms)
✔ browser plan rejects outside builds, invalid interaction scope and unbounded steps before execution (0.9011ms)
✔ real owned Edge interactions, assertions, raw PNG and stopped process survive exact result reuse and host reopen (4002.9149ms)
✔ failed real page assertion preserves actual value and forbids silent retry of the original plan (8539.0896ms)
✔ build/source mutation and sealed step tampering reject before browser dispatch; earlier task cannot execute plan (1782.0619ms)
✔ outside network never reaches another local service, and original JavaScript/console failures remain failures (3274.4592ms)
✔ stopping a real in-flight browser preserves finished steps, closes its owned root and never replays (2782.7547ms)
✔ 原PNG损坏使原结果校验失败,不能以已有通过状态复用 (3122.4693ms)
✔ fixed browsing contracts reject local host APIs, credential URLs, protocols, cross-origin and arbitrary execution (1.008ms)
✔ attested desktop queue preserves same-tab observations, original PNG bytes and approval-bound one-shot actions (1688.6062ms)
✔ page data cannot grant domains, interaction or other Task scope; nonce and ownership remain fixed (1237.5537ms)
✔ unknown dispatched submission blocks even after a new observation, rejects late success and retains explicit acknowledgement (1582.8256ms)
✔ downloads require exact approved address, remain original binary artifacts, and cannot reuse consumed observation (927.1039ms)
✔ revoking shared persistent profile closes all its tabs and clears only that profile after original desktop acknowledgement (2150.2181ms)
✔ service recovery never reopens old tabs or replays dispatched actions and keeps unknown original receipts (1220.5736ms)
✔ agent browser flow waits for original Task/Run approval, dispatches once, and returns paired original results (5066.8215ms)
✔ changed approved input and stopped dispatched agent action cannot be accepted or replayed (1412.2295ms)
✔ browser inventory gives bounded metadata and original references instead of repeating full website histories (1098.0591ms)
✔ receipt storage failure settles the original dispatched action as unknown and never hangs or accepts late success (1530.3448ms)
✔ input that fits the engine window but exceeds remaining credit diagnoses the credit limit (0.4889ms)
✔ window overflow and historical pauses retain distinct and honest diagnoses (0.0804ms)
✔ R observed reservation shortage keeps the window-fit diagnosis separate from Run/Task credit (0.094ms)
✔ a budget pause persists concrete text/tool/history/image breakdown and never dispatches an oversized request (2459.1563ms)
✔ explicit frozen preview resumes the same Task in a new Run, preserves completed tools and does not change global config or selected model (4590.4763ms)
✔ insufficient budgets, silently clamped capacity and arbitrary config fields reject without a physical request or mutating task budget (2733.5784ms)
✔ changed session, changed registry, expiry and altered frozen preview all require a new explicit preview (3522.9859ms)
✔ preview preserves unknown invocation status and increasing Task allowance is committed only with confirmed new Run (4695.3813ms)
✔ batch APIs enforce ownership and review fingerprints, apply once, and persist the completed batch across restart (6600.9151ms)
✔ delivery reports bind actual batch identity and reject corrupted snapshot evidence without changing project files (4543.0338ms)
✔ a compatible model can seal proposals through the real HTTP loop while application stays behind user review (8688.0691ms)
✔ a sealed batch at the step limit hands off and continues the same Task into actual checks without replaying saved tools (19438.3173ms)
✔ applied batches cannot turn cancellation, other budgets, restart or unknown commands into a review continuation (2350.1324ms)
▶ complete sealed batch hands off after all tools settle with no next model request
✔ complete (5539.0166ms)
✔ later-error (5832.3898ms)
✔ complete sealed batch hands off after all tools settle with no next model request (11372.1548ms)
▶ partial, changed, failed or cancelled batches cannot claim complete review handoff
✔ partial (5081.4276ms)
✔ new-proposal (5175.7006ms)
✔ failed (4963.8489ms)
✔ conflict (5303.5908ms)
✔ cancelled (5120.9529ms)
✔ unknown (5754.5617ms)
✔ tampered (5718.0502ms)
✔ partial, changed, failed or cancelled batches cannot claim complete review handoff (37119.2538ms)
✔ actual process exit at file-batch prepared preserves durable identity and never replays writes (2130.9515ms)
✔ actual process exit at file-batch application_started preserves durable identity and never replays writes (2084.4481ms)
✔ actual process exit at file-batch write_intent preserves durable identity and never replays writes (2351.3987ms)
✔ actual process exit at file-batch staged preserves durable identity and never replays writes (2455.7ms)
✔ actual process exit at file-batch replaced preserves durable identity and never replays writes (2524.961ms)
✔ actual process exit at file-batch file_recorded preserves durable identity and never replays writes (3216.226ms)
✔ actual process exit at file-batch finished preserves durable identity and never replays writes (2740.304ms)
✔ actual process exit at file-batch rollback_started preserves durable identity and never replays writes (3539.5528ms)
✔ actual process exit at file-batch restore_intent preserves durable identity and never replays writes (3274.1664ms)
✔ actual process exit at file-batch restore_staged preserves durable identity and never replays writes (3832.3605ms)
✔ actual process exit at file-batch restored preserves durable identity and never replays writes (3264.8252ms)
✔ actual process exit at file-batch restore_recorded preserves durable identity and never replays writes (3177.7021ms)
✔ actual process exit at file-batch rollback_finished preserves durable identity and never replays writes (3227.4966ms)
warning: in the working copy of 'README.md', LF will be replaced by CRLF the next time Git touches it
warning: in the working copy of 'src/two.mjs', LF will be replaced by CRLF the next time Git touches it
✔ sealed batches preserve bytes until review, then apply and explicitly restore BOM/CRLF/new files (3287.3111ms)
Switched to a new branch 'owner-branch'
✔ a conflict in the last file or Git branch prevents writing every file in the batch (3217.57ms)
✔ duplicate canonical paths, ancestor collisions, links, dependencies and Windows aliases never form a batch (1190.6273ms)
✔ fingerprints, changed proposals and corrupt snapshot artifacts cannot write or restore any target (2699.0279ms)
✔ a failure after replacement restores only batch bytes and never retries an old application (2279.5923ms)
✔ rollback preserves owner edits, blocks other project writers, and explicit acknowledgement releases recovery (2276.9629ms)
✔ an unattempted owner-created file identical to a suggestion remains external and is never removed (1940.903ms)
✔ active batch requests are serialized and cancellation leaves a verifiable ended batch (1550.5976ms)
✔ batch application and rollback preserve unrelated Git index and worktree bytes (3737.6643ms)
✔ recovery reuses exact staged bytes but retains a conflicting temporary file for owner inspection (2343.0169ms)
✔ one Task can seal pending proposals from earlier Runs while snapshots retain each source Run (1532.9081ms)
✔ engineering inspection preserves a dirty Git baseline and treats project rules as data (2961.4516ms)
✔ real npm tests and build produce counted results, log artifacts and persistent verification (3163.1878ms)
✔ an actual failing test retains the nonzero exit code and failed count (2497.3883ms)
✔ unchanged automatic checks return original failure evidence, while explicit user reruns and changed source still execute (4878.2722ms)
✔ agent loops cannot hide identical failing checks behind new plan IDs or unrelated reads (11096.6936ms)
✔ zero tests, skipped tests and unrecognized stdout cannot become verification passes (9429.7766ms)
✔ a successful build with no selected tests does not verify the task (2556.0986ms)
✔ stale check plans and forged script names execute no project commands (1834.5862ms)
✔ applying a later proposal invalidates the prior passed engineering snapshot (3165.9654ms)
✔ stopping an actual check terminates its child process and blocks concurrent jobs (1991.5313ms)
✔ check timeout records a failure with its actual termination result (2776.7217ms)
✔ source edits during a successful check make the result stale (3304.3391ms)
✔ bounded large logs retain the test footer and readable artifact instead of inventing counts (2534.164ms)
✔ model-selected checks require a run-bound decision and denial creates no check job (4533.7187ms)
✔ check APIs expose owned plans and jobs and preserve the authoritative results across restart (6924.6756ms)
✔ reviewed changes continue the original task without reopening a generic completed run (2954.4643ms)
✔ concurrent review-continuation API requests accept one run and retain the goal after restart (10402.8488ms)
✔ the model reads owned failure logs without running the check again (3456.2405ms)
✔ process exit at check queued preserves results and never replays commands (1691.0745ms)
✔ process exit at check result_committed preserves results and never replays commands (1956.5706ms)
✔ model code tools keep Task/Run ownership, index evidence and exact proposals across storage recovery (6061.1508ms)
✔ bundled code APIs isolate project results, validate queries and create review-only patches with conflicts (3190.0548ms)
✔ code tool arguments reject malformed ownership-independent inputs before side effects (0.4198ms)
✔ observe-only device authority cannot read, configure or clear project parsing caches (889.7012ms)
✔ pairing is short lived, single-use, generation-bound and validates exact permissions (866.2099ms)
✔ device transaction failure neither consumes the ticket nor leaves a credential (902.673ms)
✔ devices have project and action boundaries; rotation, expiration, revocation and host changes reject old authority (905.7276ms)
✔ persistent registry stores hashes and a bounded audit, never raw credentials (997.6635ms)
✔ real TLS verifies IP SAN, CA chain, encrypted restart and public fingerprint (950.5456ms)
✔ TLS sealing failure leaves listener closed and no plaintext or successful TLS registration (799.7289ms)
✔ HTTP pairing is concurrent single-use; secure cookie and strict origin/host checks (1399.609ms)
✔ gateway never exposes management routes, arbitrary assets or unpaired project data (916.6665ms)
✔ authorized project events paginate from original seq without leaking other project events or payloads (1732.4956ms)
✔ remote task projection excludes raw model history, host configuration, tool input and credentials (896.3314ms)
✔ permission checks apply to direct remote mutation, not only UI controls (1088.2767ms)
✔ approval fingerprint and exact task/revision are bound; duplicate requests never execute twice (1030.8732ms)
✔ unknown original operation is recorded and never replayed after response failure (1291.8353ms)
✔ revocation and gateway closure reject old access without invoking runtime stop (915.0914ms)
✔ rate limits reject pairing brute force and do not persist candidate secrets (1455.1545ms)
✔ real Windows CurrentUser DPAPI persists only a sealed CA and reopens it (2386.0571ms)
✔ mobile approval executes the original Runtime command exactly once and rejects the later desktop decision (2770.9563ms)
✔ desktop rejection wins a concurrent mobile request and original command remains unexecuted (2742.615ms)
✔ phone stop waits for the original process exit and prevents later heartbeats and command replay (2895.2393ms)
✔ Run replacement during parent-stop wait is rejected rather than stopping the new Run (687.7078ms)
✔ diff pages preserve source SHA; report reads verify blobs and exclude unrelated artifact kinds (1112.6832ms)
✔ hot upgrade refuses an active gateway and restores closed gateway ownership on failure (944.9518ms)
✔ damaged device permissions or expiry fail closed and never overwrite the original registry (712.0815ms)
✔ device revocation destroys its actual authenticated TLS connection (1148.3148ms)
✔ companion observes only current Task screenshots, verifies original PNG bytes and loses access immediately on revoke (1346.7096ms)
✔ mobile browser observation is current Task/project scoped, preserves original PNG and exposes no browser control or profile data (1393.1087ms)
✔ actual config HTTP failure preserves the committed selection, config and volatile credential before a successful retry (1310.3125ms)
✔ controlled installation pins the native kernel and bridge independently of the existing connection-probe identity (1013.7264ms)
✔ controlled profiles disable every native project/terminal provider and native model/retry route using the actual include patch contract (0.5478ms)
✔ owned tool catalogues seal full schemas canonically and reject duplicate or malformed declarations (5.0612ms)
✔ bridge authentication, current Run, process, SDK session and browser-origin checks reject before dispatch (652.9015ms)
✔ bridge duplicates, unknown tools, malformed arguments and source changes are refused before a second side effect (733.763ms)
✔ bridge model requests bind the exact catalogue, route and output bound and preserve raw request/result identity (682.4089ms)
✔ stopped and oversized bridge requests never reach a local model or tool callback (761.9142ms)
✔ a Windows virtual data-directory name binds new storage and credentials to its actual physical root (1879.8795ms)
✔ physical data-root adoption still rejects explicit junctions at the directory or any ancestor (741.4235ms)
✔ actual Windows CurrentUser DPAPI roundtrips owned UTF-8 and rejects damaged bytes and wrong entropy (1297.4389ms)
✔ only explicitly supplied keys persist; new provider restores original revisions and public views contain no key (3022.9618ms)
✔ registry and cipher commit failure restores original routing, keys, block and every database row (1773.4233ms)
✔ a two-key import with a protection failure rolls back all registry and cipher entries and retains its preview (2181.7461ms)
✔ connection revision changes retain the original saved block and report a source mismatch on restart (2495.0308ms)
✔ disable skips real restart loading, explicit enable restores, delete preserves memory and forget removes both (2669.7261ms)
✔ corrupt saved block and another SID do not load; failures are visible without returning ciphertext or plaintext (2790.2703ms)
✔ legacy persistence requires a new explicit key and binds config and credential revisions; stale requests cannot save (2899.1186ms)
✔ owned plaintext keys never reach SQLite pages, WAL, config or public execution exports (1573.9556ms)
✔ reviewed application invalidates old checks, names current SHA and advances through a new approved check to Git preview (8790.0795ms)
✔ skip-tests, foreign task, unknown operations, diagnostic and parent review retain their boundaries (3773.0723ms)
✔ before the first change, real baseline failures persist across later reads; passed baselines and changed sources cannot certify delivery (8779.0426ms)
✔ actual process exit at Git queued never replays a commit and can reconcile the durable result (13583.9221ms)
✔ actual process exit at Git preparing never replays a commit and can reconcile the durable result (14162.7925ms)
✔ actual process exit at Git index_locked never replays a commit and can reconcile the durable result (15975.5443ms)
✔ actual process exit at Git prepared never replays a commit and can reconcile the durable result (16063.5577ms)
✔ actual process exit at Git ref_prepared never replays a commit and can reconcile the durable result (16923.2613ms)
✔ actual process exit at Git publishing never replays a commit and can reconcile the durable result (17122.1499ms)
✔ actual process exit at Git committed never replays a commit and can reconcile the durable result (12859.1287ms)
✔ actual process exit at Git index_published never replays a commit and can reconcile the durable result (13486.3737ms)
✔ actual process exit at Git finished never replays a commit and can reconcile the durable result (8627.3641ms)
✔ an unknown Git result cannot overwrite an externally changed index or delete an unrelated lock (8436.9194ms)
✔ known publication with a later source edit retains the commit and marks the delivery for inspection (9722.1788ms)
✔ CRLF normalization, multiple applied revisions and unrelated special index flags preserve exact bytes and ownership (11242.7322ms)
✔ adding new durable record kinds to an existing hot Store preserves all existing Maps and records (7931.8402ms)
✔ an unrelated owner deletion is fingerprinted and preserved while linked task paths are rejected (13372.1161ms)
✔ actual failed tests become a reviewed local commit and evidence report while preserving unrelated index and files (16151.2154ms)
✔ selected preexisting unstaged, staged and untracked owner edits cannot enter a task commit (19683.3082ms)
Switched to a new branch 'owner-branch'
✔ source, unrelated stage and branch changes after review invalidate a commit plan without changing HEAD (23028.3159ms)
Switched to a new branch 'delivery-branch'
✔ a later Git metadata change keeps matching source evidence valid but a real source edit requires checks (12231.5204ms)
✔ a corrupt passed-check log and a pending proposal block Git delivery instead of inventing evidence (6104.0081ms)
HEAD is now at c0e9adc Sealed delivery fixture baseline
✔ Git filters, merge state, detached HEAD and old task ownership are explicitly rejected (24570.6056ms)
✔ local snapshot commits disable repository hooks and signing and isolate the Git child environment (8499.0624ms)
✔ stopping before publication cancels safely; stopping after publication preserves the actual completed commit (15892.5302ms)
✔ delivery API confirms exactly one owned plan, serves verified reports and preserves results across restart (13600.0445ms)
✔ the model can prepare an owned Git review and evidence report while publishing remains a separate user action (9688.4723ms)
✔ HTTP Git review pairs deferred calls, survives reload and never requests more work until the saved preview is resolved (10399.7665ms)
✔ a reviewed Git preview can be revised or withdrawn without publishing or losing the current Task (9285.7517ms)
✔ actual Electron debug argument ordering retains exactly the compiled entry and rejects extra or repeated actions (1.1848ms)
✔ native bridge refuses other origins, subframes, other windows and unsupported inputs before connecting (512.3666ms)
✔ desktop preview opening requires a currently registered project preview and rejects iframe callers (628.521ms)
✔ file reveal binds the current project, task, path, UTF-8 bytes and original SHA without executing the file (1056.0635ms)
✔ file changed during asynchronous task recheck is not revealed as the old bytes (641.9925ms)
✔ desktop browser profile and aliases stay outside project inventory and file access (974.3896ms)
✔ native picker cancel and origin change after dialog publish no project registration (582.7162ms)
✔ pending native action is not dispatched twice, and current task review returns its original seq (624.6805ms)
✔ native stop cancellation and changed original host never stop a replacement process (632.9859ms)
✔ window register rejects host PID, stale installation, extra fields and missing creation identity (0.3833ms)
✔ a rebuilt source archive binds actual generated client assets and retains its original running identity after disk changes (1766.145ms)
✔ host challenge proof rejects a stale challenge, changed identity and changed signature (0.8294ms)
✔ stop confirmation binds the original instance, creation and loaded runtime (0.136ms)
✔ state paths reject escaping, linked ancestors, ambiguous hardlinks and damaged JSON (952.6159ms)
✔ failed registration publishes no initialized memory and can complete after its actual inputs are supplied (2595.2259ms)
✔ host shutdown cancels an un-signalled HTTP model request and refuses later requests until recovered (1.5739ms)
✔ host shutdown cancels model-pool acquisition and releases queue ownership without inventing attempts (1004.7195ms)
✔ full candidate validation returns the exact loaded-input host identity and rejects undeclared or altered files (885.7308ms)
✔ fast per-file scan still rejects a junctioned subdirectory, a hard-linked file and a stream path with the original bytes intact (1017ms)
✔ installation requires separate absolute program/data directories and a candidate in its own versions namespace (1217.1969ms)
✔ inventory rejects a new hardlink on an optional file after its full-manifest scan and accepts the untouched candidate after removal (652.7417ms)
✔ staging retains the original data and pointer without querying or restarting a host (3180.4634ms)
✔ a post-pointer failure restores every original pointer byte and persistent row; a first-install failure removes only its published pointer (11178.8377ms)
✔ changed active manifest and concurrent pointer changes are rejected without overwriting the unknown record (4307.3952ms)
✔ an unregistered non-HTTP listener blocks preflight even when service-info is absent (2550.6911ms)
✔ uninstall archives its original pointer and preserves every data row/artifact instead of recursively deleting data (7273.2218ms)
✔ preflight rejects an unrelated program root or changed native bootstrap while preserving its original bytes (4438.6831ms)
✔ a changed owned bootstrap follows its candidate and a post-pointer fault restores both original bytes (16531.6276ms)
✔ an external bootstrap change after preflight is preserved and never overwritten as an update or rollback (8049.29ms)
✔ owned startup retries only read-only observation after a slow health response and attests the same original child (4337.727ms)
✔ unverified startup expires with its original live child and never treats timeout or an unrelated health PID as proof (5908.0335ms)
✔ diagnostic failures and repeated questions retain the original execution error, task, checks and read-only authority (2143.4679ms)
✔ diagnostic binding survives reopen, rejects foreign Task references and detects original error or artifact tampering (1091.446ms)
✔ a new Task never adopts a previous Task error from session.lastFailure (1012.2443ms)
✔ verified research and explicit no-test scopes do not demand an unrelated test job or claim engineering success (3012.5116ms)
✔ unknown scope and unknown operations remain actionable even after a diagnostic answer or read proof (1047.3595ms)
✔ delivery status HTTP is read-only, selection-bound and equal to the saved report after host restart (5101.1775ms)
✔ routine engineering facts remove duplicate success excerpts while complete diagnostics and errors remain available (655.5671ms)
✔ a question about the previous workbench error receives local error context and current change states (1953.8633ms)
✔ an unknown model request stops, and explicit resume never replays the approved command (3940.5575ms)
✔ exhausted refused connections preserve tool results, and explicit resume adds no new user turn or replay (3115.4404ms)
✔ stopping during model backoff cancels further requests and clears the running request state (1649.6299ms)
✔ HTTP authentication errors are classified, redacted and never automatically retried (17.1097ms)
✔ temporary HTTP service errors retry and recover with recorded attempt counts (17.7165ms)
✔ invalid and oversized protocol responses fail once with a useful response-format diagnosis (5.3243ms)
✔ HTTP rate limits expose their source and obey the service retry interval before recovery (1009.2634ms)
✔ Windows commands preserve Chinese through a PowerShell pipeline into Node (782.5448ms)
✔ Windows shell syntax errors stay readable and native command exit codes are retained (1003.112ms)
✔ review and completed states do not turn a generic continue click into a new model request (745.5677ms)
✔ old session data receives a truthful historical connection diagnosis without rewriting history (683.48ms)
✔ engine HTTP checks bind the detected installation and current task, page original bytes and preserve them across host restart (5621.3275ms)
✔ HTTP stop seals a pending SDK call, rejects automatic replay and keeps failures attached to the original session (4399.3421ms)
✔ sealed r1-off failure fits its original 16K config with authentic latest status and readable source (132.3009ms)
✔ sealed r1-on failure fits its original 16K config with authentic latest status and readable source (77.4701ms)
✔ sealed r2-on failure fits its original 16K config with authentic latest status and readable source (36.6452ms)
✔ sealed r2-off failure fits its original 16K config with authentic latest status and readable source (43.6342ms)
✔ archived native call pairs require exact saved arguments/results rather than model-written summaries (8.5148ms)
✔ full system/user requirements and mandatory current failure can refuse rather than be trimmed (12.85ms)
✔ small native input is byte-preserving; developer instructions remain mandatory after compaction (0.191ms)
✔ older failing checks and pending proposals remain visible after a later large file read (15.1039ms)
✔ bounded result keeps every check status and cuts Unicode only on whole characters (0.4966ms)
✔ controlled engine HTTP freezes one child, pages original owned input/report, rejects foreign reads and prevents changed-source reuse (15251.4503ms)
✔ a sealed whole-batch review pauses the owned native session without certifying engine completion or losing the saved host result (14782.4442ms)
✔ a changed review seal still closes the owned native process and cannot claim a verified review handoff (15007.5049ms)
✔ native text and call IDs preserve original result pairing while unsupported content fails closed (0.4767ms)
✔ one engine child freezes real installation identity before creating tasks and rejects a changed source (8927.2184ms)
✔ owned native tools return saved results, use real approvals/checks and account each actual request once across new engine Runs (26740.4614ms)
✔ forged native scope cannot change a parent file and a final completion claim cannot replace worker proof (13747.1114ms)
✔ duplicate native IDs are rejected without replaying a saved tool (13010.9053ms)
✔ native input exceeding the context budget is retained and sends no model request (11322.004ms)
✔ bounded native sends verify original tool provenance and reject forged views or missing versions (13820.2123ms)
✔ stop retains a late reported model usage once and executes no tool from that output (10855.3183ms)
✔ a source change after the original model intent stops before native tools execute (10640.6312ms)
@@ -0,0 +1,88 @@
> local-agent-workbench@1.25.3 package
> node scripts/package.mjs --immutable
{
"portableRoot": "D:\\Ai-project\\REA逆向\\Agent-Workbench\\releases\\Agent-Workbench-1.25.3-win-x64-20261104175949830",
"sourceRoot": "D:\\Ai-project\\REA逆向\\Agent-Workbench\\releases\\Agent-Workbench-1.25.3-source-20261104175949830",
"evidenceRoot": "D:\\Ai-project\\REA逆向\\Agent-Workbench\\releases\\Agent-Workbench-1.25.3-evidence-20261104175949830",
"portable": {
"path": "D:\\Ai-project\\REA逆向\\Agent-Workbench\\releases\\Agent-Workbench-1.25.3-win-x64-20261104175949830.zip",
"bytes": 243407090,
"sha256": "379b164cff06b2e6c2c09fff9f74ed9e8cb795914d0c80ff1e4131c0aa209e7b"
},
"source": {
"path": "D:\\Ai-project\\REA逆向\\Agent-Workbench\\releases\\Agent-Workbench-1.25.3-source-20261104175949830.zip",
"bytes": 1247166991,
"sha256": "485d8c26543caa7f055e0d5463492c28057aabb012bb1090cc2755a1454d0a94"
},
"evidence": {
"path": "D:\\Ai-project\\REA逆向\\Agent-Workbench\\releases\\Agent-Workbench-1.25.3-evidence-20261104175949830.zip",
"bytes": 1005656605,
"sha256": "768174cba6832ba457ba0b0701e88906c75120f201c3da9c0fdebfdd102f5025"
},
"documentation": {
"sourceCommit": "6741510b2dc9dd1191ae7f459474a981ed1d07c9",
"originalFiles": 26646,
"originalBytes": 3173916990,
"shippedFiles": 194,
"shippedBytes": 1403919,
"evidenceManifestSha256": "ae8b6deeafe711055cc9574966a1a1d2f0699d3b4b4aaa7730b47726397104a6"
},
"installer": {
"path": "D:\\Ai-project\\REA逆向\\Agent-Workbench\\releases\\Agent-Workbench-1.25.3-win-x64-20261104175949830-Setup.exe",
"bytes": 213768104,
"sha256": "80d3f9269fbdcc057412a4dbde1f5894cab953b103bfca4ea2478639a57a2132",
"compiler": {
"version": "6.3.1",
"sha256": "3b22d36fc7e7c421d8caf780008ecdb42b14b3f0371b9bafe3cad64e6648d9cf"
},
"source": "D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\installer-build-f7799718-ded5-4fc5-9e60-bb3d3ac7941d\\installer.iss",
"version": "1.25.3",
"candidateDirectory": "Agent-Workbench-1.25.3-win-x64-20261104175949830",
"qaFixture": null
},
"desktopRuntime": {
"version": "45.2.0",
"sha256": "010ea51bfb528c8143513439aa93fb40754403daa2023f9decd4517da71706ab",
"bytes": 249810432,
"executable": "desktop/AgentWorkbench.exe",
"scope": "Fixed original Windows Electron runtime, full license texts, isolated application loader and native bootstrap; original executable hash unchanged."
},
"nativeLauncher": {
"bytes": 12288,
"sha256": "e3d31cb7ce5009d3ca5a96d04dc16b1add0bd1f5cde7e1304461de6713d37463",
"inputs": {
"compiler": "46809206887326d2d24db1eff1f3064de972c3451abe766b49111450a5e08e00",
"source": "1c8e7d7ee5c383897bec48e9aadb8051b47c09686f74bbf34e0b6b695cd01ae5",
"icon": "f1879074e45f01c480ad9bf50b818637cc60ad56526056c886f808a4812d7b0c"
},
"compiler": "Windows .NET Framework C# compiler; source-bound cached original output, no deterministic-build claim."
},
"nodeVersion": "v24.22.1",
"reaRuntime": {
"version": "6.8.0",
"protocolVersion": "2025-03-26",
"packages": 149,
"files": 8413,
"bytes": 142395402,
"lockSha256": "687fb3fa3089a12f239dd5e87d9fa07df55a87df4935931d495a92600a1b0f6e"
},
"portableFiles": 10989,
"licenseTexts": 135,
"browserRuntime": {
"playwrightVersion": "1.64.1",
"packages": [
{
"name": "playwright",
"version": "1.64.1",
"files": 60
},
{
"name": "playwright-core",
"version": "1.64.1",
"files": 115
}
]
}
}
@@ -0,0 +1,88 @@
> local-agent-workbench@1.25.3 package
> node scripts/package.mjs --immutable
{
"portableRoot": "D:\\Ai-project\\REA逆向\\Agent-Workbench\\releases\\Agent-Workbench-1.25.3-win-x64-20261104181218275",
"sourceRoot": "D:\\Ai-project\\REA逆向\\Agent-Workbench\\releases\\Agent-Workbench-1.25.3-source-20261104181218275",
"evidenceRoot": "D:\\Ai-project\\REA逆向\\Agent-Workbench\\releases\\Agent-Workbench-1.25.3-evidence-20261104181218275",
"portable": {
"path": "D:\\Ai-project\\REA逆向\\Agent-Workbench\\releases\\Agent-Workbench-1.25.3-win-x64-20261104181218275.zip",
"bytes": 243407817,
"sha256": "43f33f855715c9b1ce4ce5f1e4fcdfdd2bdbe7d6e6a0c61b647a6b75775cd170"
},
"source": {
"path": "D:\\Ai-project\\REA逆向\\Agent-Workbench\\releases\\Agent-Workbench-1.25.3-source-20261104181218275.zip",
"bytes": 1247167737,
"sha256": "990dbb481ce83a350f6f7ec4ab468849ca50f999fa3e94579765aa9f8111d039"
},
"evidence": {
"path": "D:\\Ai-project\\REA逆向\\Agent-Workbench\\releases\\Agent-Workbench-1.25.3-evidence-20261104181218275.zip",
"bytes": 1005656997,
"sha256": "6bc64a506eca1fb1bdce8b2a1d0ae681ec237aba6970d29bdde4ad17e4afd3aa"
},
"documentation": {
"sourceCommit": "1bd94856d31a4d122eb03a9d7d199b42dbabacc6",
"originalFiles": 26646,
"originalBytes": 3173917785,
"shippedFiles": 194,
"shippedBytes": 1404714,
"evidenceManifestSha256": "3b56e6be40266d5e47089bd96d3db0f53aa09fabf3952089648240778c602990"
},
"installer": {
"path": "D:\\Ai-project\\REA逆向\\Agent-Workbench\\releases\\Agent-Workbench-1.25.3-win-x64-20261104181218275-Setup.exe",
"bytes": 213769026,
"sha256": "55f144ea2698ed3ac3da8d74ca9287fc02f7844dc23511113ba098154d8d4be7",
"compiler": {
"version": "6.3.1",
"sha256": "3b22d36fc7e7c421d8caf780008ecdb42b14b3f0371b9bafe3cad64e6648d9cf"
},
"source": "D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\installer-build-16d31904-3f5a-4c0b-9590-d153904d4787\\installer.iss",
"version": "1.25.3",
"candidateDirectory": "Agent-Workbench-1.25.3-win-x64-20261104181218275",
"qaFixture": null
},
"desktopRuntime": {
"version": "45.2.0",
"sha256": "010ea51bfb528c8143513439aa93fb40754403daa2023f9decd4517da71706ab",
"bytes": 249810432,
"executable": "desktop/AgentWorkbench.exe",
"scope": "Fixed original Windows Electron runtime, full license texts, isolated application loader and native bootstrap; original executable hash unchanged."
},
"nativeLauncher": {
"bytes": 12288,
"sha256": "e3d31cb7ce5009d3ca5a96d04dc16b1add0bd1f5cde7e1304461de6713d37463",
"inputs": {
"compiler": "46809206887326d2d24db1eff1f3064de972c3451abe766b49111450a5e08e00",
"source": "1c8e7d7ee5c383897bec48e9aadb8051b47c09686f74bbf34e0b6b695cd01ae5",
"icon": "f1879074e45f01c480ad9bf50b818637cc60ad56526056c886f808a4812d7b0c"
},
"compiler": "Windows .NET Framework C# compiler; source-bound cached original output, no deterministic-build claim."
},
"nodeVersion": "v24.22.1",
"reaRuntime": {
"version": "6.8.0",
"protocolVersion": "2025-03-26",
"packages": 149,
"files": 8413,
"bytes": 142395402,
"lockSha256": "687fb3fa3089a12f239dd5e87d9fa07df55a87df4935931d495a92600a1b0f6e"
},
"portableFiles": 10989,
"licenseTexts": 135,
"browserRuntime": {
"playwrightVersion": "1.64.1",
"packages": [
{
"name": "playwright",
"version": "1.64.1",
"files": 60
},
{
"name": "playwright-core",
"version": "1.64.1",
"files": 115
}
]
}
}
File diff suppressed because one or more lines are too long
File diff suppressed because one or more lines are too long
@@ -0,0 +1,70 @@
{
"success": true,
"sourceCommit": "1bd94856d31a4d122eb03a9d7d199b42dbabacc6",
"candidate": "D:\\Ai-project\\REA逆向\\Agent-Workbench\\releases\\Agent-Workbench-1.25.3-win-x64-20261104181218275",
"stages": [
"rebuild",
"archives",
"portable",
"source-reasoning",
"source-reasoning-desktop",
"source-catalog",
"source-catalog-desktop",
"source-a",
"source-a-desktop",
"source-b",
"source-b-desktop",
"source-c",
"source-c-desktop",
"source-d",
"source-d-desktop",
"portable-reasoning",
"portable-reasoning-desktop",
"portable-catalog",
"portable-catalog-desktop",
"portable-fidelity",
"portable-messages",
"portable-a",
"portable-a-desktop",
"portable-b",
"portable-b-desktop",
"portable-c",
"portable-c-desktop",
"portable-d",
"portable-d-desktop",
"portable-files",
"portable-results",
"portable-reading",
"install"
],
"fullSuiteSourceCommit": "6741510b2dc9dd1191ae7f459474a981ed1d07c9",
"acceptanceOnlyDelta": [
{
"path": "scripts/reasoning-release.mjs",
"beforeSha256": "002440c9226c7824fe49725abd9d027527373adfa9c5dc144c35ffe3bd934fa4",
"afterSha256": "ce38a40fb9140f1be8a305b7e0226c4c8fa3d4162f192c03c9132a55b81e2523"
},
{
"path": "scripts/verify-controlled-release.mjs",
"beforeSha256": "dd24176e7173350b514df79f255d86daffee1a037dd5d9df4e71e1db414a8a55",
"afterSha256": "d80b2f410214662b040ec7dcaf508721aafe96a51f6667aa7ed609b545d909d1"
}
],
"inputs": 652,
"compiled": 22,
"wholeSuite": {
"tests": 926,
"pass": 926,
"fail": 0,
"cancelled": 0,
"skipped": 0,
"todo": 0,
"duration_ms": 2143077.8077
},
"fullLogSha256": "520b223452f59fd0b53d9418bdf7a5dba7c0d2713f4291bc805aec3ca037d245",
"sourceUIGroups": 78,
"portableUIGroups": 126,
"installedUIGroups": 103,
"realModelRequests": 0,
"formalUpgradeYet": false
}
@@ -0,0 +1,36 @@
import fs from 'node:fs';
import path from 'node:path';
import assert from 'node:assert/strict';
import { createHash } from 'node:crypto';
import { execFileSync } from 'node:child_process';
import { stages, groups, installedGroups, suiteStats, checkReceipt } from './acceptance.mjs';
const root = path.resolve('.qa/reasoning-1.25.3');
const read = file => JSON.parse(fs.readFileSync(path.join(root, file), 'utf8').replace(/^\uFEFF/, ''));
const hash = file => createHash('sha256').update(fs.readFileSync(file)).digest('hex');
const freeze = read('release-verified/freeze.json'), candidate = read('release-verified/candidate.json');
assert.equal(execFileSync('git', ['status', '--porcelain'], { encoding: 'utf8', windowsHide: true }).trim(), '');
assert.equal(execFileSync('git', ['rev-parse', 'HEAD'], { encoding: 'utf8', windowsHide: true }).trim(), freeze.sourceCommit);
assert.deepEqual(JSON.parse(fs.readFileSync('releases/latest.json')), candidate);
for (const row of freeze.inputs) assert.equal(hash(row.path), row.sha256, row.path);
for (const row of freeze.compiled) {
assert.equal(hash(row.path), row.sha256, row.path); assert.equal(hash(path.join(candidate.portableRoot, row.path)), row.sha256, row.path);
}
const wholeSuite = suiteStats(fs.readFileSync(path.join(root, 'full-after-upgrade-fix.txt'), 'utf8'));
const delta = read('candidate-input-delta.json'); assert.equal(delta.success, true); assert.equal(delta.after, freeze.sourceCommit);
assert.equal(delta.before, '6741510b2dc9dd1191ae7f459474a981ed1d07c9'); assert.equal(delta.compiledExact, 22);
assert.ok(wholeSuite.tests > 900); assert.equal(wholeSuite.pass, wholeSuite.tests);
for (const name of ['fail', 'cancelled', 'skipped', 'todo']) assert.equal(wholeSuite[name], 0);
for (const stage of stages) {
const suffix = ['portable-fidelity', 'portable-messages'].includes(stage) ? '-reviewed' : '';
const receipt = read('release-verified/receipts/' + stage + suffix + '.json');
if (suffix) assert.equal(hash(receipt.resultPath), receipt.result.metadataAdapter.originalResultSha256);
checkReceipt(stage, receipt, freeze, candidate); assert.equal(hash(receipt.log), receipt.logSha256);
}
const result = { success: true, sourceCommit: freeze.sourceCommit, candidate: candidate.portableRoot, stages,
fullSuiteSourceCommit: delta.before, acceptanceOnlyDelta: delta.changes,
inputs: freeze.inputs.length, compiled: freeze.compiled.length, wholeSuite, fullLogSha256: hash(path.join(root, 'full-after-upgrade-fix.txt')),
sourceUIGroups: Object.entries(groups).filter(([stage]) => stage.startsWith('source-')).reduce((sum, [, count]) => sum + count, 0),
portableUIGroups: Object.entries(groups).filter(([stage]) => stage.startsWith('portable-')).reduce((sum, [, count]) => sum + count, 0),
installedUIGroups: installedGroups.reduce((sum, count) => sum + count, 0), realModelRequests: 0, formalUpgradeYet: false };
fs.writeFileSync(path.join(root, 'preinstall-review.json'), JSON.stringify(result, null, 2) + '\n', { flag: 'wx' });
console.log(JSON.stringify(result));
@@ -0,0 +1,47 @@
import fs from 'node:fs';
import path from 'node:path';
import assert from 'node:assert/strict';
import { createHash } from 'node:crypto';
import { spawn } from 'node:child_process';
import { projectEnvironment } from '../../src/core/process-environment.mjs';
const root = path.resolve('.qa/reasoning-1.25.3/release-verified'), mode = process.argv[2];
assert.ok(['portable-fidelity', 'portable-messages'].includes(mode));
const read = file => JSON.parse(fs.readFileSync(file, 'utf8').replace(/^\uFEFF/, ''));
const hash = file => createHash('sha256').update(fs.readFileSync(file)).digest('hex');
const freeze = read(path.join(root, 'freeze.json')), candidate = read(path.join(root, 'candidate.json'));
assert.deepEqual(read('releases/latest.json'), candidate);
for (const row of freeze.inputs) assert.equal(hash(row.path), row.sha256);
for (const row of freeze.compiled) assert.equal(hash(path.join(candidate.portableRoot, row.path)), row.sha256);
const output = path.join(root, 'receipts', mode + '-reviewed.json'); assert.equal(fs.existsSync(output), false);
let log, startedAt = new Date().toISOString();
if (mode === 'portable-fidelity') {
const failed = read(path.join(root, 'receipts/portable-fidelity.json')); assert.equal(failed.success, false);
assert.match(failed.error, /undefined/); log = failed.log; assert.equal(hash(log), failed.logSha256);
} else {
log = path.join(root, 'logs', mode + '-reviewed.log'); const fd = fs.openSync(log, 'wx');
let code;
try { code = await new Promise((resolve, reject) => {
const child = spawn(process.execPath, ['scripts/verify-anthropic-messages-ui.mjs', '--app-root', candidate.portableRoot],
{ env: projectEnvironment({ AGENT_WORKBENCH_QA_APP_ROOT: candidate.portableRoot }), windowsHide: true, stdio: ['ignore', fd, fd] });
child.once('error', reject); child.once('close', resolve);
}); } finally { fs.closeSync(fd); }
assert.equal(code, 0, 'Preserve the original result and inspect ' + log);
}
const summaries = fs.readFileSync(log, 'utf8').trim().split(/\r?\n/).filter(line => line.startsWith('{')).flatMap(line => { try { return [JSON.parse(line)]; } catch { return []; } });
const summary = summaries.findLast(row => row?.success !== undefined); assert.equal(summary.success, true);
const resultPath = path.join(summary.directory ?? summary.root, 'result.json'), original = read(resultPath);
assert.equal(original.success, true); assert.equal(original.checks.length, mode === 'portable-fidelity' ? 6 : 7);
assert.deepEqual(original.errors, []); assert.equal(original.realModelRequests, 0);
assert.equal(fs.realpathSync.native(original.appRoot), fs.realpathSync.native(candidate.portableRoot));
const manifest = read(path.join(candidate.portableRoot, 'MANIFEST.json')); assert.equal(manifest.version, '1.25.3');
const compiled = fs.readFileSync(path.join(candidate.portableRoot, 'dist-server/server.mjs'), 'utf8');
assert.equal(/APP_VERSION\s*=\s*["']([0-9.]+)["']/.exec(compiled)?.[1], '1.25.3');
if (mode === 'portable-messages') assert.equal(original.testedVersion, '1.25.3');
const adapter = { originalResultSha256: hash(resultPath), originalVersionField: mode === 'portable-messages' ? 'testedVersion' : null,
versionSource: mode === 'portable-messages' ? 'original observed health testedVersion' : 'original exact appRoot plus candidate manifest and every attested compiled file; original legacy proof contains no observed health version',
compiledFilesVerified: freeze.compiled.length, assertionsAndTimeoutsUnchanged: true, originalResultUnmodified: true };
const result = { ...original, version: '1.25.3', metadataAdapter: adapter };
const receipt = { success: true, mode: mode + '-reviewed', resultPath, result, log, logSha256: hash(log), startedAt,
finishedAt: new Date().toISOString(), sourceCommit: freeze.sourceCommit, candidate: candidate.portableRoot, realModelRequests: 0 };
fs.writeFileSync(output, JSON.stringify(receipt, null, 2) + '\n', { flag: 'wx' });
console.log(JSON.stringify({ success: true, mode: receipt.mode, checks: original.checks.length, adapter, output }));
@@ -0,0 +1,13 @@
import { spawn } from 'node:child_process';
import assert from 'node:assert/strict';
const stages = process.argv.slice(2);
assert.ok(stages.length);
for (const stage of stages) {
const started = Date.now();
const code = await new Promise((resolve, reject) => {
const child = spawn(process.execPath, ['scripts/reasoning-release.mjs', stage, 'release-verified'], { windowsHide: true, stdio: 'inherit' });
child.once('error', reject); child.once('close', resolve);
});
assert.equal(code, 0, 'Stopped at ' + stage + '; inspect the preserved receipt before any further attempt.');
console.log('COMPLETED ' + stage + ' ' + (Date.now() - started) + 'ms');
}
@@ -0,0 +1,92 @@
import fs from 'node:fs';
import path from 'node:path';
import assert from 'node:assert/strict';
import { createHash } from 'node:crypto';
import { suiteStats } from './acceptance.mjs';
const source = path.resolve('.qa/reasoning-1.25.3'), development = path.resolve('.qa/reasoning-controls');
const root = path.resolve('docs/model-catalog-1.25.3-evidence');
const read = file => JSON.parse(fs.readFileSync(file, 'utf8').replace(/^\uFEFF/, ''));
const sha = bytes => createHash('sha256').update(bytes).digest('hex');
assert.equal(fs.existsSync(root), false, 'Never replace sealed evidence.');
for (const file of ['preinstall-review.json', 'formal/final.json', 'formal-web/result.json', 'formal-native/result.json']) assert.equal(read(path.join(source, file)).success, true, file);
const freeze = read(path.join(source, 'release-verified/freeze.json')), candidate = read(path.join(source, 'release-verified/candidate.json'));
assert.deepEqual(read('releases/latest.json'), candidate); assert.equal(candidate.documentation.sourceCommit, freeze.sourceCommit);
fs.mkdirSync(root); const entries = [], originals = new Set();
const supported = /\.(?:json|jsonl|log|txt|tap|png|jpe?g|md|mjs|ts|html)$/;
function copy(file, relative) {
file = path.resolve(file); if (originals.has(file)) return; originals.add(file);
const stat = fs.lstatSync(file); assert.equal(stat.isSymbolicLink(), false); assert.equal(stat.isFile(), true);
const bytes = fs.readFileSync(file); relative = relative.replaceAll('\\', '/');
if (relative.endsWith('.mjs')) relative += '.txt';
const target = path.resolve(root, relative); assert.ok(target.startsWith(root + path.sep));
fs.mkdirSync(path.dirname(target), { recursive: true }); fs.writeFileSync(target, bytes, { flag: 'wx' });
entries.push({ path: relative, bytes: bytes.length, sha256: sha(bytes), source: file });
}
function top(directory, destination) {
for (const row of fs.readdirSync(directory, { withFileTypes: true })) if (row.isFile() && supported.test(row.name)) {
if (directory === source && ['continuation.md', 'report-draft.md'].includes(row.name)) continue;
copy(path.join(directory, row.name), path.posix.join(destination, row.name));
}
}
function walk(directory, destination) {
for (const row of fs.readdirSync(directory, { withFileTypes: true })) {
const file = path.join(directory, row.name), next = path.posix.join(destination, row.name);
assert.equal(fs.lstatSync(file).isSymbolicLink(), false);
if (row.isDirectory()) walk(file, next); else copy(file, next);
}
}
function ui(directory, destination) {
assert.ok(path.resolve(directory).startsWith(path.resolve('.qa') + path.sep)); top(directory, destination);
for (const folder of ['artifacts', 'snapshots']) if (fs.existsSync(path.join(directory, folder))) walk(path.join(directory, folder), destination + '/' + folder);
}
function resultFromLog(file) {
const lines = fs.readFileSync(file, 'utf8').trim().split(/\r?\n/);
const rows = lines.flatMap((line, index) => {
if (!line.startsWith('{')) return [];
try { return [JSON.parse(line)]; } catch {}
try { return [JSON.parse(lines.slice(index).join('\n'))]; } catch { return []; }
});
const summary = rows.findLast(row => row && (row.root || row.directory || row.output));
return summary ? summary.output ?? path.join(summary.directory ?? summary.root, 'result.json') : null;
}
top(source, 'development'); top(development, 'development/controls');
for (const row of fs.readdirSync(development, { withFileTypes: true })) if (row.isDirectory()) {
if (row.name === 'public-docs') walk(path.join(development, row.name), 'development/controls/public-docs');
else ui(path.join(development, row.name), 'validation/development-' + row.name);
}
for (const row of fs.readdirSync(development, { withFileTypes: true }).filter(row => row.isFile() && /\.(txt|log)$/.test(row.name))) {
const resultFile = resultFromLog(path.join(development, row.name));
if (resultFile && fs.existsSync(resultFile)) ui(path.dirname(resultFile), 'validation/development-' + path.basename(row.name, path.extname(row.name)));
}
for (const folder of ['release', 'release-final', 'release-fixed', 'release-verified', 'formal', 'formal-web', 'formal-native']) if (fs.existsSync(path.join(source, folder))) walk(path.join(source, folder), folder);
const receiptsRoot = path.join(source, 'release-verified/receipts');
for (const file of fs.readdirSync(receiptsRoot).filter(name => /^(source|portable)-/.test(name) && name.endsWith('.json'))) {
const receipt = read(path.join(receiptsRoot, file)); if (!receipt.success) continue; ui(path.dirname(receipt.resultPath), 'validation/' + receipt.mode);
}
const build = read(path.join(receiptsRoot, 'rebuild.json')); top(build.result.root, 'validation/rebuild');
const installation = read(path.join(receiptsRoot, 'install.json')); top(installation.result.root, 'validation/install');
for (const [index, item] of installation.result.installedUI.proofs.entries()) {
const resultFile = resultFromLog(item.log); assert.ok(resultFile); ui(path.dirname(resultFile), 'validation/installed-' + index);
}
for (const row of freeze.inputs) { assert.equal(sha(fs.readFileSync(row.path)), row.sha256); copy(row.path, 'source-bytes/' + row.path); }
for (const row of freeze.compiled) { const file = path.join(candidate.portableRoot, row.path); assert.equal(sha(fs.readFileSync(file)), row.sha256); copy(file, 'compiled/' + row.path); }
for (const kind of ['portable', 'source']) copy(path.join(candidate[kind + 'Root'], 'MANIFEST.json'), kind + '-manifest.json');
const formal = read(path.join(source, 'formal/final.json')), review = read(path.join(source, 'preinstall-review.json'));
const identity = { version: '1.25.3', sourceCommit: freeze.sourceCommit, candidate: path.basename(candidate.portableRoot),
installer: candidate.installer, archives: Object.fromEntries(['portable', 'source', 'evidence'].map(kind => [kind, candidate[kind]])),
portableManifestSha256: sha(fs.readFileSync(path.join(candidate.portableRoot, 'MANIFEST.json'))), sourceManifestSha256: sha(fs.readFileSync(path.join(candidate.sourceRoot, 'MANIFEST.json'))),
formalPid: formal.pid, runtimeRoot: formal.runtimeRoot, retention: formal.retention,
inputs: freeze.inputs.length, independentInputs: build.result.inputCount, compiled: freeze.compiled.length,
fullSuiteResult: suiteStats(fs.readFileSync(path.join(source, 'full-after-upgrade-fix.txt'), 'utf8')), fullLogSha256: review.fullLogSha256,
fullSuiteSourceCommit: review.fullSuiteSourceCommit, acceptanceOnlyDelta: review.acceptanceOnlyDelta,
sourceUIGroups: review.sourceUIGroups, portableUIGroups: review.portableUIGroups, isolatedInstallGroups: 12,
installedUISuites: installation.result.installedUI.proofs.length, installedUIGroups: review.installedUIGroups,
realModelRequests: 0, realCatalogRequests: 0,
unresolved: ['real upstream catalog and metadata coverage', 'real autonomous engineering delivery', 'intermittent network root cause', 'real reservation outlier boundary benefit', 'original Purge AB/BA', 'historical concurrent-load startup and approval timeout root causes', 'unsupported audio/video/server tools and structured-output execution'] };
fs.writeFileSync('docs/model-catalog-1.25.3-release-identity.json', JSON.stringify(identity, null, 2) + '\n', { flag: 'wx' });
copy('docs/model-catalog-1.25.3-release-identity.json', 'release-identity.json');
entries.sort((a, b) => a.path.localeCompare(b.path));
const manifest = { schemaVersion: 1, version: '1.25.3', sourceCommit: freeze.sourceCommit, entries, count: entries.length,
bytes: entries.reduce((sum, row) => sum + row.bytes, 0), scope: 'Original selected development failures and passes, superseded freeze and cancelled test, exact final source/compile inputs, independent release/UI/install and retained formal state evidence. No real keys, environments, node_modules, installer binaries or unrelated project files copied.' };
fs.writeFileSync(path.join(root, 'manifest.json'), JSON.stringify(manifest, null, 2) + '\n', { flag: 'wx' });
console.log(JSON.stringify({ success: true, count: manifest.count, bytes: manifest.bytes, manifestSha256: sha(fs.readFileSync(path.join(root, 'manifest.json'))), sourceCommit: freeze.sourceCommit }));
@@ -0,0 +1,44 @@
✔ full candidate validation returns the exact loaded-input host identity and rejects undeclared or altered files (732.8244ms)
✔ fast per-file scan still rejects a junctioned subdirectory, a hard-linked file and a stream path with the original bytes intact (659.1989ms)
✔ installation requires separate absolute program/data directories and a candidate in its own versions namespace (858.8613ms)
✔ inventory rejects a new hardlink on an optional file after its full-manifest scan and accepts the untouched candidate after removal (654.7992ms)
✔ staging retains the original data and pointer without querying or restarting a host (2854.7439ms)
✔ a post-pointer failure restores every original pointer byte and persistent row; a first-install failure removes only its published pointer (9532.3433ms)
✔ changed active manifest and concurrent pointer changes are rejected without overwriting the unknown record (3543.2254ms)
✔ an unregistered non-HTTP listener blocks preflight even when service-info is absent (2376.3425ms)
✔ uninstall archives its original pointer and preserves every data row/artifact instead of recursively deleting data (5751.7785ms)
✔ preflight rejects an unrelated program root or changed native bootstrap while preserving its original bytes (3996.78ms)
✔ a changed owned bootstrap follows its candidate and a post-pointer fault restores both original bytes (13361.9209ms)
✔ an external bootstrap change after preflight is preserved and never overwritten as an update or rollback (6423.7351ms)
✔ runtime guides map archived evidence while all original reports and successful/failed source bytes remain independently verifiable (542.7522ms)
✔ missing required research reports fail packaging rather than creating a successful placeholder (532.0127ms)
✔ full verification reports only verified bytes, complete declared scope and actual directory inventory; changed/deleted files still fail (735.1457ms)
✔ preflight and activation continue one monotonic operation bound to installer hash; terminal replay and changed origin are rejected (7282.1999ms)
✔ preflight failures and fixed-candidate mismatches retain the actual reason without publishing a pointer (4272.4734ms)
✔ cancel finishes only the original ready preparation, preserves data and pointer and refuses replay (6235.4835ms)
✔ the installation lock rejects concurrent same-id operations before they can overwrite original progress (3702.8829ms)
✔ original post-publication failure is linked to its rollback receipt and every original byte survives (3257.0596ms)
✔ a candidate mutation at the publication boundary still fails even after an earlier successful full validation (6233.715ms)
✔ public progress marks a lost process unknown without editing its original record or claiming completion (4496.5685ms)
✔ throwing console/UI progress observers cannot decide the installation result (3421.7103ms)
✔ a progress-only write failure after committed publication returns the authoritative receipt and a warning (3525.3689ms)
✔ browser and MCP approvals use actual tool categories; mobile exposes only category while keeping original fingerprint (552.6892ms)
✔ the compiled runtime version constant tracks package.json so a release bump cannot ship a stale served version (8.0653ms)
✔ claiming the same runtime owner twice preserves its original metadata and every database row (616.1756ms)
✔ sessions without controlled-engine history retain their prior export shape and never expose another session record (564.5963ms)
✔ live workers, controlled drivers, or MCP operations refuse an upgrade before changing handlers, managers or database rows (560.6814ms)
✔ an idle upgrade loads the newly introduced engine record map, resets obsolete managers and preserves original rows, sessions and provider objects (652.6293ms)
✔ an installation failure restores old managers, original record maps, handlers and all existing database rows (1352.2131ms)
✔ the owned controlled driver remains an active session and project even after outer task activity has ended (616.8538ms)
✔ release and upgrade help or invalid arguments have no file or network setup side effects in an independent directory (807.4632ms)
✔ the upgrade compatibility gate includes the served current version and the previous formal patch (0.4124ms)
✔ a directly started compiled host upgrades twice with the same PID, sessions, applied bytes and volatile provider (3461.9663ms)
✔ the running 0.1.0 process upgrades in place and retains its project, session, files and volatile model connection (6653.1317ms)
ℹ tests 36
ℹ suites 0
ℹ pass 36
ℹ fail 0
ℹ cancelled 0
ℹ skipped 0
ℹ todo 0
ℹ duration_ms 111458.0338
@@ -0,0 +1,14 @@
✔ a live SQLite host with the actual 1.9 artifact object upgrades, keeps its original blobs and verifies real browser PNG bytes (3525.1557ms)
✔ failed upgrade restores the original artifact object, provider and request handler (1540.6865ms)
✔ hot upgrade adopts a retained virtual data root without reopening SQLite or changing rows, Provider or routing (1248.8616ms)
✔ failed adoption restores the original virtual data-root name and original open database (550.136ms)
✔ hot upgrade refreshes retained model and importer methods while preserving credential and routing objects (572.0636ms)
✔ hot upgrade refuses active probes and restores retained model methods if installation fails (510.1146ms)
ℹ tests 6
ℹ suites 0
ℹ pass 6
ℹ fail 0
ℹ cancelled 0
ℹ skipped 0
ℹ todo 0
ℹ duration_ms 8559.6168
@@ -0,0 +1,134 @@
import fs from 'node:fs';
import path from 'node:path';
import assert from 'node:assert/strict';
import { createHash } from 'node:crypto';
import { execFileSync } from 'node:child_process';
const root = path.resolve('docs/model-catalog-1.25.3-evidence'), index = process.argv.includes('--index');
assert.ok(process.argv.slice(2).every(value => value === '--index'));
const hash = bytes => createHash('sha256').update(bytes).digest('hex');
const blob = bytes => createHash('sha1').update(Buffer.from('blob ' + bytes.length + '\0')).update(bytes).digest('hex');
const bytes = relative => fs.readFileSync(path.join(root, relative));
const read = relative => JSON.parse(bytes(relative).toString('utf8').replace(/^\uFEFF/, ''));
const manifest = read('manifest.json'), names = new Set(), originalFiles = new Map(); let total = 0;
for (const row of manifest.entries) {
assert.ok(row.path && !path.isAbsolute(row.path) && !row.path.includes('\\') && !row.path.split('/').includes('..'));
assert.equal(names.has(row.path), false); names.add(row.path);
const value = bytes(row.path); assert.equal(value.length, row.bytes); assert.equal(hash(value), row.sha256, row.path); total += value.length;
originalFiles.set(path.resolve(row.source), row.path);
}
assert.equal(manifest.count, names.size); assert.equal(manifest.bytes, total);
const localOriginal = file => { const name = originalFiles.get(path.resolve(file)); assert.ok(name, file); return name; };
const freeze = read('release-verified/freeze.json'), candidate = read('release-verified/candidate.json'), identity = read('release-identity.json');
assert.equal(manifest.version, '1.25.3'); assert.equal(manifest.sourceCommit, freeze.sourceCommit); assert.equal(candidate.documentation.sourceCommit, freeze.sourceCommit);
assert.equal(identity.sourceCommit, freeze.sourceCommit); assert.equal(identity.candidate, path.basename(candidate.portableRoot));
assert.equal(hash(bytes('portable-manifest.json')), identity.portableManifestSha256); assert.equal(hash(bytes('source-manifest.json')), identity.sourceManifestSha256);
assert.equal(identity.formalPid, 36552); assert.deepEqual(identity.installer, candidate.installer);
const sourceTree = execFileSync('git', ['ls-tree', '-r', '-z', '--full-tree', freeze.sourceCommit], { encoding: 'utf8', maxBuffer: 16 * 1024 * 1024, windowsHide: true }).split('\0').filter(Boolean);
const sourceBlobs = new Map(sourceTree.map(row => { const [meta, file] = row.split('\t'); return [file, meta.split(' ')[2]]; }));
for (const row of freeze.inputs) {
const value = bytes('source-bytes/' + row.path + (row.path.endsWith('.mjs') ? '.txt' : '')); assert.equal(hash(value), row.sha256, row.path);
const oid = sourceBlobs.get(row.path); assert.ok(oid, row.path);
if (blob(value) !== oid) { assert.equal(value.includes(0), false); assert.equal(blob(Buffer.from(value.toString('utf8').replaceAll('\r\n', '\n'))), oid, row.path); }
}
for (const row of freeze.compiled) assert.equal(hash(bytes('compiled/' + row.path + (row.path.endsWith('.mjs') ? '.txt' : ''))), row.sha256, row.path);
const groups = {
'source-reasoning': 6, 'source-reasoning-desktop': 6, 'source-catalog': 8, 'source-catalog-desktop': 8,
'source-a': 7, 'source-a-desktop': 3, 'source-b': 9, 'source-b-desktop': 9, 'source-c': 6, 'source-c-desktop': 6, 'source-d': 5, 'source-d-desktop': 5,
'portable-reasoning': 6, 'portable-reasoning-desktop': 6, 'portable-catalog': 8, 'portable-catalog-desktop': 8, 'portable-fidelity': 6, 'portable-messages': 7,
'portable-a': 7, 'portable-a-desktop': 3, 'portable-b': 9, 'portable-b-desktop': 9, 'portable-c': 6, 'portable-c-desktop': 6, 'portable-d': 5, 'portable-d-desktop': 5,
'portable-files': 9, 'portable-results': 16, 'portable-reading': 10,
};
const stages = ['rebuild', 'archives', 'portable', ...Object.keys(groups), 'install'];
const receipts = Object.fromEntries(stages.map(stage => {
const suffix = ['portable-fidelity', 'portable-messages'].includes(stage) ? '-reviewed' : '';
const receipt = read('release-verified/receipts/' + stage + suffix + '.json'); assert.equal(receipt.success, true, stage);
if (suffix) {
assert.equal(hash(bytes(localOriginal(receipt.resultPath))), receipt.result.metadataAdapter.originalResultSha256);
assert.equal(receipt.result.metadataAdapter.compiledFilesVerified, 22); assert.equal(receipt.result.metadataAdapter.originalResultUnmodified, true);
}
assert.equal(receipt.sourceCommit, freeze.sourceCommit); assert.equal(receipt.candidate, candidate.portableRoot); assert.equal(receipt.realModelRequests, 0);
assert.equal(hash(bytes('release-verified/logs/' + path.basename(receipt.log))), receipt.logSha256); assert.equal(receipt.result.success, true);
if (groups[stage]) {
assert.equal((receipt.result.checks ?? receipt.result.passed).length, groups[stage], stage);
assert.deepEqual(receipt.result.errors, []); assert.equal(receipt.result.realModelRequests, 0);
if (stage !== 'portable-reading') assert.equal(receipt.result.version, '1.25.3');
}
return [stage, receipt];
}));
const build = receipts.rebuild.result; assert.equal(build.inputCount, freeze.inputs.length + 5); assert.equal(build.compiled.length, 22); assert.deepEqual(build.mismatches, []);
for (const row of freeze.inputs) assert.equal(build.inputs.find(item => item.path === row.path)?.sha256, row.sha256);
for (const row of build.compiled) assert.deepEqual(row.actual, row.expected);
for (const kind of ['portable', 'source', 'evidence']) {
const archive = receipts.archives.result.archives.find(row => row.kind === kind);
assert.equal(archive.crcPassed, true); assert.equal(archive.manifestAndDirectoryMatch, true); assert.equal(archive.sha256, candidate[kind].sha256);
}
const installation = receipts.install.result; assert.equal(installation.results.length, 12); assert.ok(installation.results.every(row => row.passed));
const installedGroups = [16, 9, 7, 3, 9, 9, 6, 6, 5, 5, 6, 6, 8, 8]; assert.equal(installation.installedUI.proofs.length, installedGroups.length);
for (const [i, item] of installation.installedUI.proofs.entries()) {
const proof = item.proof; assert.equal(proof.success, true); assert.equal(proof.version, '1.25.3'); assert.deepEqual(proof.errors, []); assert.equal(proof.realModelRequests, 0);
assert.equal((proof.checks ?? proof.passed).length, installedGroups[i]); assert.equal(hash(bytes(localOriginal(item.log))), item.sha256);
}
const stats = {};
for (const name of ['tests', 'pass', 'fail', 'cancelled', 'skipped', 'todo', 'duration_ms']) {
const match = bytes('development/full-after-upgrade-fix.txt').toString('utf8').match(new RegExp('^[ℹ#] ' + name + ' ([0-9.]+)\\s*$', 'm'));
assert.ok(match, 'Complete test footer: ' + name); stats[name] = Number(match[1]);
}
assert.deepEqual(stats, identity.fullSuiteResult); assert.ok(stats.tests > 900); assert.equal(stats.pass, stats.tests);
for (const name of ['fail', 'cancelled', 'skipped', 'todo']) assert.equal(stats[name], 0);
assert.equal(hash(bytes('development/full-after-upgrade-fix.txt')), identity.fullLogSha256);
assert.equal(read('development/preinstall-review.json').success, true);
const delta = read('development/candidate-input-delta.json');
assert.equal(delta.success, true); assert.equal(delta.before, identity.fullSuiteSourceCommit); assert.equal(delta.after, freeze.sourceCommit); assert.equal(delta.compiledExact, 22);
assert.deepEqual(delta.changes.map(row => row.path), ['scripts/reasoning-release.mjs', 'scripts/verify-controlled-release.mjs']);
const testedFreeze = read('release-fixed/freeze.json'); assert.equal(testedFreeze.sourceCommit, delta.before); assert.deepEqual(testedFreeze.compiled, freeze.compiled);
for (const row of freeze.inputs) {
const old = testedFreeze.inputs.find(item => item.path === row.path); assert.ok(old);
if (!delta.changes.some(item => item.path === row.path)) assert.equal(row.sha256, old.sha256, row.path);
}
const firstPortable = read('release-fixed/receipts/portable.json'); assert.equal(firstPortable.success, false);
assert.equal(hash(bytes('release-fixed/logs/portable.log')), firstPortable.logSha256);
assert.match(bytes('release-fixed/logs/portable.log').toString('utf8'), /includes\(expectedVersion\)/);
const oldMetadata = read('release-verified/receipts/portable-fidelity.json'); assert.equal(oldMetadata.success, false); assert.match(oldMetadata.error, /undefined/);
assert.equal(read('release-final/freeze.json').sourceCommit, 'bb4b497713894bdd9016e18a0009cbc333d2be93');
const originalFull = bytes('development/full-final.txt').toString('utf8');
for (const pattern of [/^ℹ tests 926\s*$/m, /^ℹ pass 923\s*$/m, /^ℹ fail 3\s*$/m, /^ℹ cancelled 0\s*$/m]) assert.match(originalFull, pattern);
assert.match(originalFull, /This migration accepts a supported existing Workbench/);
assert.equal(read('release/freeze.json').sourceCommit.startsWith('109e5124'), true);
assert.ok(bytes('development/full-source.txt').length > 0); assert.ok(names.has('development/full-source-cancelled-scope.json'));
const developmentCatalogFailures = [...names].filter(name => name.startsWith('validation/development-catalog-web-') && name.endsWith('/result.json')).filter(name => read(name).success === false);
assert.ok(developmentCatalogFailures.length > 0, 'Preserve the first UI assertion failure.');
const before = read('formal/state-before.json');
for (const stage of ['upgrade', 'final']) {
const proof = read('formal/' + stage + '.json'), state = read('formal/' + stage + '-state-after.json');
assert.equal(proof.success, true); assert.equal(proof.version, '1.25.3'); assert.equal(proof.pid, 36552); assert.equal(proof.inspectorClosed, true); assert.equal(proof.desktopWindows, 0);
for (const key of ['pid', 'hasKey', 'configSha256', 'projectsSha256', 'sessionsSha256', 'eventCursor', 'exports', 'persistent', 'modelCenterSha256']) assert.deepEqual(state[key], before[key], key);
assert.deepEqual(state.database.tables, before.database.tables); assert.deepEqual(state.database.sessionIds, before.database.sessionIds);
}
assert.equal(read('formal/installer-exit.json').code, 0);
for (const file of ['formal-web/result.json', 'formal-native/result.json']) {
const proof = read(file); assert.equal(proof.success, true); assert.equal(proof.version, '1.25.3'); assert.equal(proof.pid ?? proof.formalPid, 36552); assert.equal(proof.realModelRequests, 0);
}
assert.equal(read('formal-web/result.json').realCatalogRequests, 0); assert.deepEqual(read('formal-web/result.json').rejectedWrites, []);
const native = read('formal-native/result.json');
assert.equal(native.realCatalogRequests, 0); assert.equal(native.configurationSaved, false); assert.equal(native.keyReadOrResaved, false);
assert.equal(native.closedNativeWindows, true); assert.equal(read('formal-native/launch.json').code, 0);
for (const stage of ['01-task', '02-model-center', '03-upstream-form', '04-deepseek-reasoning']) {
const observation = read('formal-native/' + stage + '.json');
assert.ok(observation.screenshots.length > 0);
for (const screenshot of observation.screenshots) {
assert.ok(['image/jpeg', 'image/png'].includes(screenshot.mime));
assert.ok(names.has('formal-native/' + screenshot.file));
const value = bytes('formal-native/' + screenshot.file);
if (screenshot.mime === 'image/jpeg') assert.equal(value.subarray(0, 3).toString('hex'), 'ffd8ff');
else assert.equal(value.subarray(0, 8).toString('hex'), '89504e470d0a1a0a');
}
}
if (index) {
const rows = execFileSync('git', ['ls-files', '--stage', '-z', '--', 'docs/model-catalog-1.25.3-evidence'], { encoding: 'utf8', maxBuffer: 16 * 1024 * 1024, windowsHide: true }).split('\0').filter(Boolean);
const blobs = new Map(rows.map(row => { const [meta, file] = row.split('\t'); return [file, meta.split(' ')[1]]; }));
for (const name of [...names, 'manifest.json']) assert.equal(blobs.get('docs/model-catalog-1.25.3-evidence/' + name), blob(bytes(name)), name + ' indexed bytes');
}
console.log(JSON.stringify({ success: true, files: names.size, bytes: total, sourceCommit: freeze.sourceCommit, inputs: freeze.inputs.length,
independentInputs: build.inputCount, compiled: 22, fullSuitePassed: stats.pass, fullSuiteFailed: stats.fail,
sourceUIGroups: identity.sourceUIGroups, portableUIGroups: identity.portableUIGroups, isolatedInstallGroups: 12, installedUIGroups: 103,
firstFailuresPreserved: true, realModelRequests: 0, realCatalogRequests: 0, indexChecked: index }));
Binary file not shown.

After

Width:  |  Height:  |  Size: 129 KiB

File diff suppressed because one or more lines are too long
Binary file not shown.

After

Width:  |  Height:  |  Size: 127 KiB

Some files were not shown because too many files have changed in this diff Show More