fix: hand off complete changes and retain verified source context

This commit is contained in:
Agent Workbench Dev
2026-11-03 07:58:13 +08:00
parent 876df5f421
commit c1521e6cd2
35 changed files with 669 additions and 79 deletions
+2
View File
@@ -1,5 +1,7 @@
# Agent Workbench项目执行规范
当前优先G:用户要求先修完已总结问题。1.24.1已实现完整变更批次同轮结算后审查交接、必要小源码先保留和原生容量/剩余预算准确诊断;停止、新提案、冲突、未知与篡改不冒充完整交接,应用后仍显式续接。E原第7–12请求离线复现估算完全一致,修复前六文件保留6/5/5/4/2/2,修复后均6/6且原硬预算保持。正式仍1.24.0/PID36552/Provider,当前正在完成1.24.1冻结回归、两视口和发行安装收口,见docs/QA-DELIVERY-RELIABILITY-G.md。新真实模型/Purge收益未重测,旧失败不改写;用户最新优先修复覆盖下方历史“先补原生窗口”的顺序,不以旧Escape暂停阻塞源码工作。
当前F固定安装已完成:1.24.0/source c03860d/候选20261102152844546,原PID36552/Provider保持,原42项目/58会话/4953记录/10329事件/2850工件/公开模型中心精确保持,调试端口关闭。独立源码ZIP514输入/22编译完全一致、三归档、实际便携、12组隔离安装及安装副本16组成果/9组文件流程通过;69份/46795742字节证据离线通过。首候选安装管理器漏编译及原60秒QA超时保持失败,见docs/QA-ORIGINAL-PURGE-RELEASE-F.md。正式原生窗口回看被物理Escape停止,未验收;不能宣称本切片全部体验验收完成。先补该项,再按docs/LIVE-DELIVERY-EFFICIENCY-PLAN.md处理原小源码保留抖动和完整批次后的检视空转。不得以新的发行源码重写D/E原编译身份或收益结论。
E实测完成:同源修复载荷、真实claude-opus-5-5/Chat/32K六新例52请求/621486报告tokens/0未知;Python原6/6/编译/审查Git通过,React和四原版AB/BA未交付。两次原版实际apply与11开启/17关闭输入通过核对,不证明收益。889份证据见docs/LIVE-RETEST-E-RESULTS.md;46原会话/2143工件/原表每行与96原尝试保持,公开参数和原池恢复。F现已固定安装:21份执行服务/网页/桌面编译与E保持,一个安装管理器入口补齐已修正脚本的重新编译。不重跑旧失败,不擅自扩大预算,不复制Key。
+7
View File
@@ -1,5 +1,12 @@
# Agent Workbench 更新记录
## 1.24.1 · 审查交接与上下文可靠性修复
- 完整文件批次封存并结算同轮工具后立即进入审查,停止额外模型调用;应用后显式续接原任务。
- 优先稳定保留有原工具与SHA来源的小源码,下一页按实际保留量分配空间;保留上下文和累计预算硬限制。
- 原生引擎区分上下文窗口不足和剩余预算不足,保留旧失败与原用量。
- 原件离线复现、边界回归、实际编译HTTP/审查/检查与两视口验证分别记录,见docs/QA-DELIVERY-RELIABILITY-G.md;真实模型收益未重测。
## 1.24.0 · 已有成果续接与真实交付(已本地交付)
- 当前已有文件可经完整读取和SHA核对纳入成果,保存Task/Run、原文件快照、任务要求与显式采用来源。
+33
View File
@@ -0,0 +1,33 @@
# 1.24.1:真实案例确认问题修复
本切片先修复E原件中可复现的问题,暂停1.25新增能力。不续跑旧Task,不读取或复制Key,不将声明HTTP夹具当作真实模型交付或Purge收益。
## 改动
完整`prepare_change_set`成功后,先结算同轮所有工具,再核对当前Task、批次范围、原提案/快照、Git分支和HEAD、实际文件基线以及未知操作。只有覆盖所有待审查修改的完整未变批次进入`needs_review/change_set_review`,立即交还审查,不再调用模型。后续提案使批次不完整、失败封存、外部冲突、待核对操作和用户停止保持各自状态;单个只读工具失败保留原错误,不破坏完整批次的审查交接。文件仍须人工审查应用,应用后显式续接原Task进行实际检查。
请求规划先放入有原工具/SHA来源的小源码,再接纳可选历史。按长度和路径稳定排序,下一页实际保留量写入封存计划;不再用2048固定上限代表所有小源码。当前窗口、完整目标/权限/原结果/配对和逐请求剩余预算仍是硬约束,16K软目标允许容纳必要固定成本和已读源码。旧发送视图继续按原版本核对。
原生预算诊断按原`limitingReason`区分上下文容量和累计剩余额度。E中27424输入可放入27552可用窗口,但下一次29600预留超过28339剩余额度,应显示“原生引擎剩余预算不足”。缺少历史分项时保持“不足项待核对”,不推断旧记录。
## 开发证据与回顾
- 审查交接旧回归两例失败,边界五例通过;修改后的初轮9/9。补充未知结果及篡改指纹场景,最终结果另录。
- 原React第7–12请求使用原system、实际路由工具、历史、32K窗口和逐请求剩余额度离线重建。修复前输入估算逐项等于原件:15714/15873/15905/15779/16000/15999,六文件完整保留6/5/5/4/2/2。修复后六次均6/6,输入15714/16110/16383/16504/16993/16945,全部仍在原27552或21197硬预算内。
- 上下文初相关42项41通过/1失败:实现误将最新完整工具组也限制到软目标。已恢复“保留源码后,最新完整工具组可在硬预算内保留”的兼容行为,最终request-plan/working-set31/31。旧失败保留。
- 首编译缺新共享模块声明文件,补齐`.d.mts`后1.24.1构建通过。首HTTP相关使用旧dist,保留2失败;后续只按当前编译验证。
- 首UI在390px未关闭侧边检视区就点击后方续接按钮,脚本失败保留;按实际导航关闭检视区后继续。该失败不作为产品成功结果。
- 首全回归为开发混合轮,发现版本入口常量尚为1.24.0(package已1.24.1),已停止该轮,不作为最终完整整轮。另两失败为浏览审批等待和旧`resume`错误文案断言;独立复验浏览原场景通过,`needs_review`正确返回409“请先审查待处理改动”,原断言改核对该明确状态及提示。修正版本入口后重新构建,后续按最终来源验证受影响范围。
- 受影响15文件原轮131项130通过/1失败,唯一失败为旧断言要求容量错误含“预算”。已明确核对“上下文窗口可用容量”、原拒绝工件和`limitingReason=context_limit`,独立原场景1/1通过;原131轮不改写为全通过。后续18文件冻结回归另录。
- 已检查发行验证器的版本范围,为1.24.1保留REA、原生SDK、浏览器、受控工程、已有成果、实际安装及历史界面入口,避免新补丁版静默跳过原验收分支。旧证据验证器的固定版本与身份不改。
- 离线驱动初始化、SQLite保存和遗漏路由工具的开发失败保留;最终从路由variant取完整tools,才得到与原件逐项相同的修复前基线。不得用早期漏工具输出宣称修复收益。
## 收口状态
最终18文件冻结回归155/155通过,覆盖完整交接/所有边界、当前Task显式续接、实际Node检查、文件快照恢复、分页/源码保留、原生引擎/用量、Chat/Responses协议。1.24.1完整构建通过,12份产品输入在冻结回归与原件离线重放后SHA保持。没有把原131轮或中断混合轮改写成该最终结果。
最终1.24.1编译界面10组通过,桌面/390px的批次审查、刷新、整批应用、显式续接与实际Node检查均完成;实际原图已目视核对,两视口预算诊断可读、无横向溢出。额外11组批次界面验收通过,包括实际外部冲突、原字节回退、未知恢复、跨项目迟到响应保护,页面及请求错误均为0。
不可变源码独立重建、便携/安装和正式原状态核对正在执行。正式当前仍1.24.0/PID36552/Provider保留;全部会话以新只读快照为准,bootstrap仅显示其中48个,不能把它当作全部58个会话计数。正式原生窗口回看尚待补验。
可复用验收入口:`scripts/verify-live-context-retention.mjs`和`scripts/verify-delivery-reliability-ui.mjs`。各自要求新输出目录,保留既有失败,不覆盖旧证据。
@@ -0,0 +1,29 @@
{
"success": true,
"version": "1.24.1",
"testedAt": "2026-11-02T23:56:19.047Z",
"cases": [
"Desktop batch scope, per-file diffs, sealed single-write guard and pre-apply bytes",
"Three-file reviewed application preserves exact BOM/CRLF/LF and creates the new file",
"Explicit batch rollback restores exact originals and removes only the reviewed new file",
"A last-file owner conflict blocks the whole batch and shows the real recovery direction",
"Durable partial recovery reads actual hashes and preserves owner content during rollback",
"Changed observation fingerprints reject stale acknowledgement without releasing the project",
"Explicit owner-content acknowledgement ends recovery without claiming successful application",
"Switching projects clears the previous task notification while retaining its durable error and recovery history",
"390px step-limit review, same-task continuation entry, rollback confirmation and keyboard close fit the viewport",
"A delayed preparation response never opens a different project or changes its selected session",
"A failed batch response after its component unmounts cannot display an error in the newly selected project"
],
"pageErrors": [],
"requestFailures": [],
"screenshots": [
".qa/screenshots/delivery-reliability-batch-desktop-review.png",
".qa/screenshots/delivery-reliability-batch-desktop-conflict.png",
".qa/screenshots/delivery-reliability-batch-desktop-recovery.png",
".qa/screenshots/delivery-reliability-batch-mobile-review.png",
".qa/screenshots/delivery-reliability-batch-mobile-rollback.png"
],
"projectRoot": "D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\batch-ui-c32248b4-ae65-4150-aca2-a252ec93daff",
"scope": "Real independent demo host and durable batch states; UI fixtures include actual owner conflict/rollback results. No real model credentials or user project were used."
}
+3
View File
@@ -954,3 +954,6 @@ C实际渲染回看:1440/390px原截图均显示55,017上游报告、35,000运
C后续审查:首冻结147/147通过后,Responses独立首轮4响应+1unknown/reset,未自动重发,原失败保留;原夹具catch直接destroy且未保存原因。增加只收集error的诊断后新UUID五组通过,不能把未复现断开归因于已知断言。另修正每次重试/切池前的预留核对:45,100已报告/70,000预算时虽尚未超额,剩余24,900已低于校准预留,legacy和池都必须只发一次并保留503。最终425输入重新冻结,首轮147不算新来源149通过。D只读确认原插件默认autoRetry=true/retryMax3/hooksDenyBypass=true;完整原patch另含身份与redteam组件,必须隔离配置/加载闭包,不能直接改正式宿主。
D原版加载实证:31模块闭包还依赖dsh-llm,60包输入不足以代表全部Harness/原生闭包,已追加每场景335个实际JS/JSON加载身份。原插件自动更新/应用可关闭,但仍注册三项宿主操作工具和fullAccess钩子,需外部门控。Windows Node权限对不存在目录需显式通配;Harness启动向祖先查package元数据,JSONL落盘从盘根逐级stat且fsync被权限模型禁用。采用祖先实际stat元数据委托及受认证/真路径/dev/ino/size限制的真实父fsync,不跳过同步;三负例通过。native builtin loader需allow-addons,故仍不是任意插件OS沙箱。最终正常退出0和四全树退出已核对;原两项passedtrue漏退出断言已在报告明确不合格。工具错误会正常触发下一模型轮次,与模型失败重试不同,所有六桥输入均保留,不能隐藏第二轮或计作真实模型请求。
G修复开发:完整批次交接旧回归2例失败、边界5例通过;修改后9/9通过。React原件离线以原system/tools/历史/32K/逐请求剩余额度重建,修复前第7–12次估算全部与原件相同,六文件完整保留为6/5/5/4/2/2;稳定小源码优先后6次均6/6,实际输入15714–16993,仍在原硬预算内,16K软目标未当成硬上限。无需新模型请求,原件保持。QA驱动初始化/SQLite/漏传路由工具的原失败保留,最终补实际路由工具后实现精确复现。
+2 -2
View File
@@ -1,12 +1,12 @@
{
"name": "local-agent-workbench",
"version": "1.24.0",
"version": "1.24.1",
"lockfileVersion": 3,
"requires": true,
"packages": {
"": {
"name": "local-agent-workbench",
"version": "1.24.0",
"version": "1.24.1",
"dependencies": {
"@babel/parser": "7.29.11",
"@peculiar/x509": "3.1.0",
+1 -1
View File
@@ -1,6 +1,6 @@
{
"name": "local-agent-workbench",
"version": "1.24.0",
"version": "1.24.1",
"private": true,
"type": "module",
"description": "A local coding workbench with explicit tool execution, reviewable changes and recoverable sessions.",
+9
View File
@@ -1,5 +1,9 @@
# 统一执行进度
## G 已确认问题修复启动
用户明确优先修完已总结的问题。基线876df5f干净;先复现完整审查批次仍running和源码在16K软目标下逐份挤出。沿用既有规划与产品流程验收技能,不开子代理,不读取Key或重放旧实测Task;F正式已固定安装,旧Esc停止不作为当前源码修复的等待条件。首批读取超出输出上限,截断内容不作完整审查,后续缩小范围。
## F 固定安装与证据收口
最终候选20261102152844546/source c03860d已实际安装到正式versions目录,替代临时QA处理器,版本仍1.24.0。新源码ZIP提取514输入并离线实际重建,全部22编译与便携一致;D/E的21执行编译不变,修正安装管理器一个入口。三归档、实际便携链路、12组隔离安装及安装副本16组成果/9组文件流程通过。原60秒QA等待超时保持,最终驱动只把受控本机响应客户端期限设300秒、实际42382ms/两认证请求无重试,不修改产品时限。
@@ -1442,3 +1446,8 @@ C实现bb45dab;464份/19,086,976字节/425源码身份/446工件记录封存
D研究接续:实际原版入口/31模块及apply已在独立无Key Harness加载。最终original-purge-load-c6df89e7-8cd3-4e6c-b070-a8e6d02fec05四场景、六模型桥输入、外部工具拒绝、三项broker负例通过;正常原进程退出码0,四全树关闭。观察335模块/场景、338唯一路径;60原包文件与原目录/副本SHA保持。原始错误逐日志/UUID保留,早期两项passed=true因漏正常退出核对已明确不合格,未改原件。Windows JSONL祖先元数据及真实fsync通过有归属代理,不开放根目录读取;allow-addons不是OS沙箱,D生产接入/原生闭包仍待实现,不代表Purge收益。开发中一次rg误传PowerShell参数、两次猜测路径已纠正,未改变产品运行源码。真实请求0,正式宿主未动。
D实现7b4f7ee;496份/17,251,023字节/338观察加载路径/1,340加载记录离线通过;manifest SHA acabdce397139dcd59cbe7f66229a1353d4f95a5ceb7a555977e544b17ea1066。封存预检首次将自己的空输出log误当JSON收据,创建目标前拒绝;排除封存输出并原样保留失败log后完成,未改原模型/过程证据。继续记录生产接入范围,不以此研究驱动直接跑真实Purge收益。
G修复开发:完整批次交接旧回归2例失败、边界5例通过;修改后9/9通过。React原件离线以原system/tools/历史/32K/逐请求剩余额度重建,修复前第7–12次估算全部与原件相同,六文件完整保留为6/5/5/4/2/2;稳定小源码优先后6次均6/6,实际输入15714–16993,仍在原硬预算内,16K软目标未当成硬上限。无需新模型请求,原件保持。QA驱动初始化/SQLite/漏传路由工具的原失败保留,最终补实际路由工具后实现精确复现。
G冻结收口:18文件155/155相关回归、最终1.24.1完整构建、10组两视口交接与预算界面、11组批次恢复/冲突/回退通过。桌面交接与1440/390px预算原图目视核对通过。一次只读收据路径误猜为.qa,实际生成于docs,按验证器写入位置定位;未重跑流程。封存验证器在提交前收紧为严格155项。一次补丁因工具JSON转义误读拒绝,未改变源码。下一不可变打包、独立重建与保留正式状态部署。
+1
View File
@@ -19,6 +19,7 @@ const prefixes = Object.freeze({
'1.22.1': 'local-files',
'1.23.0': 'release-efficiency',
'1.24.0': 'existing-results',
'1.24.1': 'delivery-reliability',
});
export function releaseEvidencePrefix(version) {
const prefix = prefixes[version];
+1 -1
View File
@@ -64,7 +64,7 @@ export async function upgradeRunning({ url, pid, currentRoot, newRoot, handlerRo
if (!fs.existsSync(originalScript) || !fs.existsSync(upgradeScript) || !fs.existsSync(path.join(next, 'dist-client', 'index.html'))) throw new Error('Upgrade bundles are incomplete.');
if (!Number.isInteger(pid) || pid < 1 || pid === process.pid) throw new Error('Invalid target process.');
const before = await readUpgradeJson(origin + '/api/bootstrap', 5000, 'Read original Workbench');
if (!before.token || !['0.1.0', '1.0.0', '1.1.0', '1.2.0', '1.3.0', '1.4.0', '1.5.0', '1.5.1', '1.6.0', '1.6.1', '1.6.2', '1.7.0', '1.8.0', '1.9.0', '1.10.0', '1.10.1', '1.11.0', '1.12.0', '1.12.1', '1.13.0', '1.14.0', '1.15.0', '1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1', '1.20.2', '1.20.3', '1.21.0', '1.22.0', '1.22.1', '1.23.0', '1.24.0'].includes(before.version)) throw new Error('This migration accepts a supported existing Workbench.');
if (!before.token || !['0.1.0', '1.0.0', '1.1.0', '1.2.0', '1.3.0', '1.4.0', '1.5.0', '1.5.1', '1.6.0', '1.6.1', '1.6.2', '1.7.0', '1.8.0', '1.9.0', '1.10.0', '1.10.1', '1.11.0', '1.12.0', '1.12.1', '1.13.0', '1.14.0', '1.15.0', '1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1', '1.20.2', '1.20.3', '1.21.0', '1.22.0', '1.22.1', '1.23.0', '1.24.0', '1.24.1'].includes(before.version)) throw new Error('This migration accepts a supported existing Workbench.');
if (before.sessions.some(session => ['running', 'awaiting_approval', 'stopping'].includes(session.status))) throw new Error('Wait for existing tasks and approvals before upgrading.');
let protocol, socket, enabledHere = false, paused = false;
const breakpoints = [];
+20 -20
View File
@@ -9,36 +9,36 @@ import { randomUUID } from 'node:crypto';
if (process.argv.slice(2).some(value => value !== '--help')) throw new Error('Unsupported argument; use --help.');
if (process.argv.includes('--help')) { console.log('node scripts/verify-all-ui.mjs - run serial desktop/390px checks in independent local hosts; use explicit QA Python/Harness/Purge source paths for 1.15.'); process.exit(0); }
const version = JSON.parse(fs.readFileSync('package.json', 'utf8')).version;
assert.ok(['1.6.0', '1.6.1', '1.6.2', '1.7.0', '1.8.0', '1.9.0', '1.10.0', '1.10.1', '1.11.0', '1.12.0', '1.12.1', '1.13.0', '1.14.0', '1.15.0', '1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1', '1.21.0', '1.22.0', '1.22.1', '1.23.0', '1.24.0'].includes(version), 'Review evidence paths when introducing the next release.');
assert.ok(['1.6.0', '1.6.1', '1.6.2', '1.7.0', '1.8.0', '1.9.0', '1.10.0', '1.10.1', '1.11.0', '1.12.0', '1.12.1', '1.13.0', '1.14.0', '1.15.0', '1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1', '1.21.0', '1.22.0', '1.22.1', '1.23.0', '1.24.0', '1.24.1'].includes(version), 'Review evidence paths when introducing the next release.');
const prefix = releaseEvidencePrefix(version), resultFile = `.qa/${prefix}-all-ui-validation.json`, results = [];
const qaEnvironment = Object.fromEntries(['AGENT_WORKBENCH_TEST_PYTHON', 'AGENT_WORKBENCH_DEEPSEEK_ROOT', 'AGENT_WORKBENCH_PURGE_178_ROOT', 'AGENT_WORKBENCH_PURGE_181_ROOT']
.filter(name => process.env[name]).map(name => [name, process.env[name]]));
for (const [name, value] of Object.entries(qaEnvironment)) assert.ok(path.isAbsolute(value) && fs.existsSync(value), 'Use an existing absolute QA runtime/source path: ' + name);
const cases = [
...(version === '1.24.0' ? [['verify-file-results-ui.mjs', null]] : []),
...(['1.23.0', '1.24.0'].includes(version) ? [['verify-release-efficiency-ui.mjs', null]] : []),
...(['1.22.1', '1.23.0', '1.24.0'].includes(version) ? [['verify-local-files.mjs', null]] : []),
...(['1.22.0', '1.22.1', '1.23.0', '1.24.0'].includes(version)?[['verify-embedded-browser.mjs',null]]:[]),
...(['1.21.0', '1.22.0', '1.22.1', '1.23.0', '1.24.0'].includes(version)?[['verify-environment-ui.mjs',null],['verify-model-fidelity-ui.mjs',null],['verify-model-wait-ui.mjs',null],['verify-task-control-ui.mjs',null]]:[]),
...(['1.20.0', '1.20.1', '1.21.0', '1.22.0', '1.22.1', '1.23.0', '1.24.0'].includes(version) ? [['verify-evaluations-ui.mjs', null]] : []),
...(['1.19.0', '1.20.0', '1.20.1', '1.21.0', '1.22.0', '1.22.1', '1.23.0', '1.24.0'].includes(version) ? [['verify-project-cache-ui.mjs', null]] : []),
...(['1.18.0', '1.19.0', '1.20.0', '1.20.1', '1.21.0', '1.22.0', '1.22.1', '1.23.0', '1.24.0'].includes(version) ? [['verify-vision-ui.mjs', null]] : []),
...(['1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1', '1.21.0', '1.22.0', '1.22.1', '1.23.0', '1.24.0'].includes(version) ? [['verify-model-protocols-ui.mjs', null]] : []),
...(['1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1', '1.21.0', '1.22.0', '1.22.1', '1.23.0', '1.24.0'].includes(version) ? [['verify-companion-ui.mjs', null]] : []),
...(['1.14.0', '1.15.0', '1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1', '1.21.0', '1.22.0', '1.22.1', '1.23.0', '1.24.0'].includes(version) ? [['verify-controlled-ui.mjs', null], ['verify-purge-ui.mjs', null]] : []),
...(['1.13.0', '1.14.0', '1.15.0', '1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1', '1.21.0', '1.22.0', '1.22.1', '1.23.0', '1.24.0'].includes(version) ? [['verify-workers-ui.mjs', null]] : []),
...(['1.11.0', '1.12.0', '1.12.1', '1.13.0', '1.14.0', '1.15.0', '1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1', '1.21.0', '1.22.0', '1.22.1', '1.23.0', '1.24.0'].includes(version) ? [['verify-task-plans-ui.mjs', `docs/${prefix}-task-ui-validation.json`]] : []),
...(['1.10.0', '1.10.1', '1.11.0', '1.12.0', '1.12.1', '1.13.0', '1.14.0', '1.15.0', '1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1', '1.21.0', '1.22.0', '1.22.1', '1.23.0', '1.24.0'].includes(version) ? [['verify-browser-ui.mjs', `docs/${prefix}-browser-ui-validation.json`]] : []),
...(['1.24.0', '1.24.1'].includes(version) ? [['verify-file-results-ui.mjs', null]] : []),
...(['1.23.0', '1.24.0', '1.24.1'].includes(version) ? [['verify-release-efficiency-ui.mjs', null]] : []),
...(['1.22.1', '1.23.0', '1.24.0', '1.24.1'].includes(version) ? [['verify-local-files.mjs', null]] : []),
...(['1.22.0', '1.22.1', '1.23.0', '1.24.0', '1.24.1'].includes(version)?[['verify-embedded-browser.mjs',null]]:[]),
...(['1.21.0', '1.22.0', '1.22.1', '1.23.0', '1.24.0', '1.24.1'].includes(version)?[['verify-environment-ui.mjs',null],['verify-model-fidelity-ui.mjs',null],['verify-model-wait-ui.mjs',null],['verify-task-control-ui.mjs',null]]:[]),
...(['1.20.0', '1.20.1', '1.21.0', '1.22.0', '1.22.1', '1.23.0', '1.24.0', '1.24.1'].includes(version) ? [['verify-evaluations-ui.mjs', null]] : []),
...(['1.19.0', '1.20.0', '1.20.1', '1.21.0', '1.22.0', '1.22.1', '1.23.0', '1.24.0', '1.24.1'].includes(version) ? [['verify-project-cache-ui.mjs', null]] : []),
...(['1.18.0', '1.19.0', '1.20.0', '1.20.1', '1.21.0', '1.22.0', '1.22.1', '1.23.0', '1.24.0', '1.24.1'].includes(version) ? [['verify-vision-ui.mjs', null]] : []),
...(['1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1', '1.21.0', '1.22.0', '1.22.1', '1.23.0', '1.24.0', '1.24.1'].includes(version) ? [['verify-model-protocols-ui.mjs', null]] : []),
...(['1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1', '1.21.0', '1.22.0', '1.22.1', '1.23.0', '1.24.0', '1.24.1'].includes(version) ? [['verify-companion-ui.mjs', null]] : []),
...(['1.14.0', '1.15.0', '1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1', '1.21.0', '1.22.0', '1.22.1', '1.23.0', '1.24.0', '1.24.1'].includes(version) ? [['verify-controlled-ui.mjs', null], ['verify-purge-ui.mjs', null]] : []),
...(['1.13.0', '1.14.0', '1.15.0', '1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1', '1.21.0', '1.22.0', '1.22.1', '1.23.0', '1.24.0', '1.24.1'].includes(version) ? [['verify-workers-ui.mjs', null]] : []),
...(['1.11.0', '1.12.0', '1.12.1', '1.13.0', '1.14.0', '1.15.0', '1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1', '1.21.0', '1.22.0', '1.22.1', '1.23.0', '1.24.0', '1.24.1'].includes(version) ? [['verify-task-plans-ui.mjs', `docs/${prefix}-task-ui-validation.json`]] : []),
...(['1.10.0', '1.10.1', '1.11.0', '1.12.0', '1.12.1', '1.13.0', '1.14.0', '1.15.0', '1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1', '1.21.0', '1.22.0', '1.22.1', '1.23.0', '1.24.0', '1.24.1'].includes(version) ? [['verify-browser-ui.mjs', `docs/${prefix}-browser-ui-validation.json`]] : []),
['verify-ui.mjs', '.qa/ui-validation.json'], ['verify-diagnostics-ui.mjs', '.qa/diagnostics-ui-validation.json'],
['verify-execution-ui.mjs', '.qa/execution-ui-validation.json'], ['verify-delivery-ui.mjs', '.qa/delivery-ui-validation.json'],
['verify-index-ui.mjs', '.qa/index-ui-result.json'], ['verify-change-sets-ui.mjs', `docs/${prefix}-batch-ui-validation.json`],
['verify-context-ui.mjs', `.qa/${version === '1.6.0' ? 'workspaces-context' : prefix}-ui-validation.json`],
['verify-workspaces-ui.mjs', `docs/${prefix === 'workspaces' ? prefix : prefix + '-workspaces'}-ui-validation.json`],
...(['1.7.0', '1.8.0', '1.9.0', '1.10.0', '1.10.1', '1.11.0', '1.12.0', '1.12.1', '1.13.0', '1.14.0', '1.15.0', '1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1', '1.21.0', '1.22.0', '1.22.1', '1.23.0', '1.24.0'].includes(version) ? [['verify-python-ui.mjs', `docs/${prefix}-python-ui-validation.json`]] : []),
...(['1.7.0', '1.8.0', '1.9.0', '1.10.0', '1.10.1', '1.11.0', '1.12.0', '1.12.1', '1.13.0', '1.14.0', '1.15.0', '1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1', '1.21.0', '1.22.0', '1.22.1', '1.23.0', '1.24.0'].includes(version) ? [['verify-react-delivery.mjs', version === '1.7.0' ? 'docs/react-engine-delivery-validation.json' : `docs/${prefix}-react-delivery-validation.json`]] : []),
...(['1.8.0', '1.9.0', '1.10.0', '1.10.1', '1.11.0', '1.12.0', '1.12.1', '1.13.0', '1.14.0', '1.15.0', '1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1', '1.21.0', '1.22.0', '1.22.1', '1.23.0', '1.24.0'].includes(version) ? [['verify-research-ui.mjs', `docs/${prefix}-research-ui-validation.json`]] : []),
...(['1.9.0', '1.10.0', '1.10.1', '1.11.0', '1.12.0', '1.12.1', '1.13.0', '1.14.0', '1.15.0', '1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1', '1.21.0', '1.22.0', '1.22.1', '1.23.0', '1.24.0'].includes(version) ? [['verify-engines-ui.mjs', version === '1.9.0' ? 'docs/engines-ui-validation.json' : `docs/${prefix}-engines-ui-validation.json`]] : []),
...(['1.12.0', '1.12.1', '1.13.0', '1.14.0', '1.15.0', '1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1', '1.21.0', '1.22.0', '1.22.1', '1.23.0', '1.24.0'].includes(version) ? ['model-center', 'project-memory', 'task-checkpoints', 'skills', 'mcp', 'mcp-recovery'].map(name => [`verify-${name}-ui.mjs`, null]) : []),
...(['1.7.0', '1.8.0', '1.9.0', '1.10.0', '1.10.1', '1.11.0', '1.12.0', '1.12.1', '1.13.0', '1.14.0', '1.15.0', '1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1', '1.21.0', '1.22.0', '1.22.1', '1.23.0', '1.24.0', '1.24.1'].includes(version) ? [['verify-python-ui.mjs', `docs/${prefix}-python-ui-validation.json`]] : []),
...(['1.7.0', '1.8.0', '1.9.0', '1.10.0', '1.10.1', '1.11.0', '1.12.0', '1.12.1', '1.13.0', '1.14.0', '1.15.0', '1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1', '1.21.0', '1.22.0', '1.22.1', '1.23.0', '1.24.0', '1.24.1'].includes(version) ? [['verify-react-delivery.mjs', version === '1.7.0' ? 'docs/react-engine-delivery-validation.json' : `docs/${prefix}-react-delivery-validation.json`]] : []),
...(['1.8.0', '1.9.0', '1.10.0', '1.10.1', '1.11.0', '1.12.0', '1.12.1', '1.13.0', '1.14.0', '1.15.0', '1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1', '1.21.0', '1.22.0', '1.22.1', '1.23.0', '1.24.0', '1.24.1'].includes(version) ? [['verify-research-ui.mjs', `docs/${prefix}-research-ui-validation.json`]] : []),
...(['1.9.0', '1.10.0', '1.10.1', '1.11.0', '1.12.0', '1.12.1', '1.13.0', '1.14.0', '1.15.0', '1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1', '1.21.0', '1.22.0', '1.22.1', '1.23.0', '1.24.0', '1.24.1'].includes(version) ? [['verify-engines-ui.mjs', version === '1.9.0' ? 'docs/engines-ui-validation.json' : `docs/${prefix}-engines-ui-validation.json`]] : []),
...(['1.12.0', '1.12.1', '1.13.0', '1.14.0', '1.15.0', '1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1', '1.21.0', '1.22.0', '1.22.1', '1.23.0', '1.24.0', '1.24.1'].includes(version) ? ['model-center', 'project-memory', 'task-checkpoints', 'skills', 'mcp', 'mcp-recovery'].map(name => [`verify-${name}-ui.mjs`, null]) : []),
];
fs.mkdirSync('.qa', { recursive: true });
const externalHost = new Set(['verify-ui.mjs', 'verify-diagnostics-ui.mjs', 'verify-execution-ui.mjs', 'verify-index-ui.mjs']);
+1 -1
View File
@@ -22,7 +22,7 @@ export async function verifyControlledApi({ api, url, token, directory, appRoot,
const origin = new URL(url);
assert.equal(origin.hostname, '127.0.0.1'); assert.notEqual(origin.port, '4318');
const expectedVersion = JSON.parse(fs.readFileSync(path.join(appRoot, fs.existsSync(path.join(appRoot, 'MANIFEST.json')) ? 'MANIFEST.json' : 'package.json'), 'utf8')).version;
assert.ok(['1.14.0', '1.15.0', '1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1', '1.20.2', '1.20.3', '1.21.0', '1.22.0', '1.22.1', '1.23.0', '1.24.0'].includes(expectedVersion));
assert.ok(['1.14.0', '1.15.0', '1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1', '1.20.2', '1.20.3', '1.21.0', '1.22.0', '1.22.1', '1.23.0', '1.24.0', '1.24.1'].includes(expectedVersion));
const bootstrap = await api('/bootstrap'); assert.equal(bootstrap.version, expectedVersion);
assert.equal(bootstrap.config.mode, 'demo'); assert.equal(bootstrap.config.hasKey, false);
const root = path.join(directory, 'compiled-controlled-node'), hooks = path.join(directory, 'disabled-hooks');
@@ -0,0 +1,83 @@
import fs from 'node:fs';
import path from 'node:path';
import assert from 'node:assert/strict';
import { createHash } from 'node:crypto';
import { requestMetrics } from '../src/shared/contracts/model-request.mjs';
if (process.argv.includes('--help')) {
console.log('node scripts/verify-delivery-reliability-evidence.mjs — offline verify sealed G bytes, exact original/fixed context, frozen regression, UI/rebuild/installation receipts and preserved state. No model or host calls.');
process.exit(0);
}
assert.equal(process.argv.length, 2);
const root = path.resolve('docs/delivery-reliability-g-evidence');
const hash = bytes => createHash('sha256').update(bytes).digest('hex');
const read = file => JSON.parse(fs.readFileSync(path.join(root, file), 'utf8'));
const manifestBytes = fs.readFileSync(path.join(root, 'manifest.json')), manifest = JSON.parse(manifestBytes);
assert.equal(manifest.schemaVersion, 1); assert.equal(manifest.version, '1.24.1');
assert.match(manifest.sourceCommit, /^[a-f0-9]{40}$/);
const paths = new Set(); let bytes = 0;
for (const entry of manifest.entries) {
assert.ok(!paths.has(entry.path)); paths.add(entry.path);
const target = path.resolve(root, entry.path); assert.ok(target.startsWith(root + path.sep));
const raw = fs.readFileSync(target); assert.equal(raw.length, entry.bytes, entry.path); assert.equal(hash(raw), entry.sha256, entry.path); bytes += raw.length;
}
assert.equal(hash(fs.readFileSync('docs/live-original-purge-e-evidence/manifest.json')), manifest.originalEManifestSha256);
assert.equal(hash(fs.readFileSync('docs/original-purge-release-f-evidence/manifest.json')), manifest.originalFManifestSha256);
const tap = fs.readFileSync(path.join(root, manifest.checks.regression), 'utf8');
const total = Number(tap.match(/ℹ tests (\d+)/)?.[1]), passed = Number(tap.match(/ℹ pass (\d+)/)?.[1]), failed = Number(tap.match(/ℹ fail (\d+)/)?.[1]);
assert.equal(total, 155); assert.equal(total, passed); assert.equal(failed, 0);
function strings(value, output = [], depth = 0) {
if (depth > 12) return output;
if (typeof value === 'string') {
output.push(value);
for (const text of new Set([value, ...value.split('\n').filter(line => /^\s*[\[{]/.test(line))])) {
try { strings(JSON.parse(text), output, depth + 1); } catch {}
}
} else if (value && typeof value === 'object') for (const item of Object.values(value)) strings(item, output, depth + 1);
return output;
}
const before = read('context/before/result.json'), after = read('context/final/result.json');
assert.equal(before.manifestSha256, manifest.originalEManifestSha256); assert.equal(after.manifestSha256, before.manifestSha256);
assert.equal(after.realModelRequests, 0); assert.equal(before.realModelRequests, 0);
assert.deepEqual(before.reports.map(report => report.sources.filter(source => source.retained).length), [6, 5, 5, 4, 2, 2]);
for (const [index, report] of after.reports.entries()) {
assert.equal(report.sources.length, 6); assert.ok(report.sources.every(source => source.retained));
assert.equal(report.inputBudget, before.reports[index].inputBudget); assert.equal(report.softTarget, 16000);
const original = read('context/before/request-' + report.request + '.json');
const fixed = read('context/final/request-' + report.request + '.json');
assert.equal(requestMetrics(original.body).estimatedInputTokens, before.reports[index].originalInputTokens);
const metrics = requestMetrics(fixed.body); assert.equal(metrics.estimatedInputTokens, report.inputTokens); assert.ok(metrics.estimatedInputTokens <= report.inputBudget);
assert.equal(metrics.sha256, fixed.plan.requestSha256);
assert.deepEqual(fixed.body.messages[0], original.body.messages[0], 'Complete original fixed goal/context');
assert.deepEqual(fixed.body.variants.map(route => route.body.tools), original.body.variants.map(route => route.body.tools), 'Original complete tools');
for (const source of report.sources) {
const raw = fs.readFileSync(path.join('docs/live-original-purge-e-evidence/baselines/react', source.path), 'utf8');
assert.equal(raw.length, source.characters); assert.ok(strings(fixed.body.messages).includes(raw), source.path);
}
}
for (const key of ['sourceUI', 'portableUI']) {
const ui = read(manifest.checks[key]); assert.equal(ui.success, true); assert.equal(ui.version, manifest.version);
assert.equal(ui.realModelRequests, 0); assert.equal(ui.fixtureRequests, 5); assert.equal(ui.checks.length, 10); assert.deepEqual(ui.errors, []);
for (const file of ui.screenshots) assert.ok(paths.has(path.posix.join(path.posix.dirname(manifest.checks[key]), path.basename(file))));
}
const release = read(manifest.checks.release), rebuild = read(manifest.checks.rebuild), archives = read(manifest.checks.archives), portable = read(manifest.checks.portable), install = read(manifest.checks.installer);
assert.equal(release.documentation.sourceCommit, manifest.sourceCommit);
assert.equal(rebuild.success, true); assert.equal(rebuild.sourceCommit, manifest.sourceCommit); assert.equal(rebuild.sourceZip.sha256, release.source.sha256);
assert.equal(rebuild.compiled.length, 22); assert.deepEqual(rebuild.mismatches, []);
assert.equal(archives.success, true); assert.equal(archives.version, manifest.version); assert.equal(archives.archives.length, 3);
for (const archive of [release.portable, release.source, release.evidence]) assert.ok(archives.archives.some(record => record.sha256 === archive.sha256));
assert.equal(portable.success, true); assert.equal(portable.version, manifest.version); assert.equal(portable.realModel, false);
assert.ok(portable.actualRea.length > 0); assert.ok(portable.actualEngine); assert.ok(portable.actualControlled); assert.ok(portable.actualBrowser);
assert.equal(install.success, true); assert.equal(install.version, manifest.version); assert.equal(install.realModelRequests, 0);
assert.equal(install.results.length, 12); assert.ok(install.results.every(result => result.passed));
for (const key of ['formalInstall', 'formalFinal']) {
const record = read(manifest.checks[key]); assert.equal(record.success, true); assert.equal(record.version, manifest.version);
assert.equal(record.sourceCommit, manifest.sourceCommit); assert.equal(record.pid, 36552); assert.equal(record.hasKey, true);
assert.equal(record.retention.originalRows, 'exact'); assert.equal(record.retention.originalExports, 'exact');
assert.equal(record.retention.originalArtifactBytes, 'exact'); assert.equal(record.retention.originalModelCenter, 'exact');
assert.equal(record.operation.state, 'completed'); assert.equal(record.inspectorClosed, true); assert.equal(record.realModelRequests, 0);
}
console.log(JSON.stringify({ success: true, version: manifest.version, sourceCommit: manifest.sourceCommit,
files: paths.size, bytes, manifestSha256: hash(manifestBytes), regression: { total, passed, failed },
originalRetained: before.reports.map(report => report.sources.filter(source => source.retained).length), fixedRetained: after.reports.map(report => report.sources.filter(source => source.retained).length),
compiledFiles: rebuild.compiled.length, uiGroups: 20, realModelRequests: 0, scope: 'Offline sealed-byte and actual-receipt verification. No new model calls, task execution, or claim of real-model/Purge success.' }));
+128
View File
@@ -0,0 +1,128 @@
import fs from 'node:fs';
import path from 'node:path';
import assert from 'node:assert/strict';
import { chromium, expect } from 'playwright/test';
import { fixture, compatibleFixture, launchWorkbench, call, completion, until, assertToolPairs } from '../tests/helpers.mjs';
if (process.argv.includes('--help')) {
console.log('node scripts/verify-delivery-reliability-ui.mjs OUTPUT [APP_ROOT] — declared local HTTP model fixture, actual review/continue/check UI and saved budget cases; no real model. OUTPUT must be new.');
process.exit(0);
}
assert.ok(process.argv[2], 'Choose a new output directory.');
const output = path.resolve(process.argv[2]);
assert.ok(!fs.existsSync(output), 'Keep prior evidence; choose a new directory.');
fs.mkdirSync(output, { recursive: true });
const appRoot = path.resolve(process.argv[3] ?? '.');
const cleanups = [], t = { after: fn => cleanups.push(fn) };
const f = fixture(t), cases = {};
fs.mkdirSync(path.join(f.projectRoot, 'test'));
fs.writeFileSync(path.join(f.projectRoot, 'package.json'), JSON.stringify({ type: 'module', scripts: { test: 'node --test test/*.test.mjs' } }));
fs.writeFileSync(path.join(f.projectRoot, 'test/values.test.mjs'), "import test from 'node:test';import assert from 'node:assert/strict';import {left} from '../src/left.mjs';import {right} from '../src/right.mjs';test('left',()=>assert.equal(left,1));test('right',()=>assert.equal(right,2));\n");
for (const [name, limitingReason] of [['remaining', 'remaining_token_budget'], ['window', 'context_limit'], ['legacy', null]]) {
const root = path.join(f.root, name); fs.mkdirSync(root); fs.writeFileSync(path.join(root, 'README.md'), '# Declared saved-budget UI fixture\n');
const project = f.store.addProject(root), session = f.store.createSession(project.id, '预算诊断 ' + name);
const goal = '保留原任务和已完成工具;按原拒绝分项核对预算。';
const run = f.store.beginRun(session, goal, false, f.store.config);
run.budgetPause = { schema: 1, reason: 'engine_input_budget', ...(limitingReason ? { limitingReason } : {}),
requiredInputTokens: name === 'window' ? 28000 : 27424, inputBudget: 27552, contextLimit: 32000,
outputReservation: 2048, safetyTokens: 2400, runRemainingTokens: 28339, taskRemainingTokens: 28339,
remainingTokens: 28339, tokenReservation: { schemaVersion: 1, reservedTokens: 29600, routes: [] } };
session.status = 'paused'; f.store.finishRun(session, run, 'engine_input_budget'); cases[name] = { project, session };
}
f.store.close();
let requests = 0, browser, page;
const checks = [], errors = [], screenshots = [];
const mark = name => { checks.push(name); console.log('PASS ' + name); };
const screenshot = async name => { const file = path.join(output, name + '.png'); await page.screenshot({ path: file, fullPage: true }); screenshots.push(file); };
try {
const model = await compatibleFixture(t, (_route, input) => {
if (!input?.messages) return { data: [{ id: 'declared-ui-fixture' }] };
requests++; assertToolPairs(input.messages);
const values = input.messages.filter(message => message.role === 'tool').map(message => JSON.parse(message.content));
if (requests === 1) return completion([
call('ui-left', 'propose_file', { path: 'src/left.mjs', content: 'export const left = 1;\n', reason: 'Declared fixture left' }),
call('ui-right', 'propose_file', { path: 'src/right.mjs', content: 'export const right = 2;\n', reason: 'Declared fixture right' }),
]);
if (requests === 2) return completion([
call('ui-seal', 'prepare_change_set', { change_ids: values.map(value => value.change_id).filter(Boolean) }),
call('ui-same-turn-inspect', 'inspect_session', {}),
]);
if (requests === 3) return completion([call('ui-check-plan', 'plan_checks', { names: ['test'] })]);
if (requests === 4) return completion([call('ui-check-run', 'run_checks', { plan_id: values.findLast(value => value.checks && value.fingerprint).id })]);
return completion([], '两份模块已通过实际 Node 测试。');
});
const host = await launchWorkbench(t, f.data, {}, { appRoot });
assert.equal(host.bootstrap.version, JSON.parse(fs.readFileSync(path.join(appRoot, 'package.json'))).version, 'Compiled host and package must agree.');
await host.api('/config', { mode: 'compatible', baseUrl: model.baseUrl, model: 'declared-ui-fixture', apiKey: 'SYNTHETIC-QA-ONLY', maxSteps: 6 });
browser = await chromium.launch({ channel: 'msedge', headless: true });
page = await browser.newPage({ viewport: { width: 1440, height: 1000 } });
page.on('pageerror', error => errors.push(error.message));
page.on('requestfailed', request => { if (request.resourceType() !== 'eventsource') errors.push(request.url() + ': ' + request.failure()?.errorText); });
await page.goto(host.url); await page.waitForLoadState('networkidle');
async function choose(projectId) {
const selector = page.getByLabel('当前项目', { exact: true });
if (!await selector.isVisible()) await page.getByRole('button', { name: '展开项目栏', exact: true }).click();
await selector.selectOption(projectId);
if ((await page.viewportSize()).width <= 940 && await page.locator('.project-sidebar').isVisible()) await page.getByRole('button', { name: '收起项目栏', exact: true }).click();
}
await choose(f.project.id);
await page.getByLabel('任务内容', { exact: true }).fill('完成双模块工程,并在审查应用后继续实际测试。');
await page.getByRole('button', { name: '发送任务', exact: true }).click();
const route = '/sessions/' + f.session.id;
let reviewed;
await until(async () => { reviewed = await host.api(route); return reviewed.status === 'needs_review'; }, 30000);
assert.equal(requests, 2); assert.equal(reviewed.execution.run.stopReason, 'change_set_review');
assert.equal(reviewed.events.find(event => event.toolCallId === 'ui-same-turn-inspect').phase, 'completed');
assert.equal(fs.existsSync(path.join(f.projectRoot, 'src')), false);
mark('Actual compiled HTTP model loop settles both tools and hands off after exactly two fixture requests without writing files');
await page.getByRole('button', { name: '查看任务记录', exact: true }).click();
await expect(page.getByLabel('任务记录', { exact: true })).toContainText('完整文件批次已封存,等待审查与应用');
await screenshot('desktop-review-handoff');
await page.reload(); await page.waitForLoadState('networkidle');
assert.equal((await host.api(route)).execution.task.id, reviewed.execution.task.id); assert.equal(requests, 2);
mark('Review handoff persists across refresh without another model request');
await page.setViewportSize({ width: 390, height: 844 }); await choose(f.project.id);
await page.getByRole('button', { name: '查看改动', exact: true }).click();
await page.getByRole('button', { name: '审查变更集', exact: true }).click();
const dialog = page.getByRole('dialog', { name: '多文件变更集审查', exact: true });
await expect(dialog).toContainText('src/left.mjs'); await expect(dialog).toContainText('src/right.mjs');
await screenshot('mobile-batch-review');
await dialog.getByRole('button', { name: '确认应用全部 2 项', exact: true }).click();
await until(async () => (await host.api(route)).execution.changeSets.at(-1).status === 'completed', 30000);
await page.keyboard.press('Escape'); await expect(page.getByRole('button', { name: '继续原任务', exact: true })).toBeVisible();
await page.getByRole('button', { name: '关闭检视区', exact: true }).click();
assert.equal(requests, 2); mark('390px reviewed whole-batch application waits for explicit continuation');
await page.getByRole('button', { name: '继续原任务', exact: true }).click();
await until(async () => (await host.api(route)).approvals.some(approval => approval.status === 'pending'), 30000);
await screenshot('mobile-check-approval');
await page.getByRole('button', { name: '确认执行', exact: true }).click();
let done;
await until(async () => { done = await host.api(route); return done.status === 'completed'; }, 30000);
assert.equal(done.execution.task.id, reviewed.execution.task.id); assert.equal(done.execution.checks[0].results[0].tests.passed, 2);
assert.equal(requests, 5);
assert.equal(done.events.filter(event => ['propose_file', 'prepare_change_set'].includes(event.tool)).length, 3);
mark('Explicit same-Task continuation runs two actual Node assertions and never replays proposals or sealing');
for (const width of [1440, 390]) {
await page.setViewportSize({ width, height: width === 390 ? 844 : 1000 });
for (const [name, title] of [['remaining', '原生引擎剩余预算不足'], ['window', '原生引擎上下文窗口不足'], ['legacy', '原生引擎输入或剩余预算不足']]) {
await choose(cases[name].project.id);
await expect(page.locator('.session-item.selected')).toContainText('预算诊断 ' + name);
const panel = page.getByLabel('任务预算诊断', { exact: true });
await expect(panel.getByRole('heading', { name: title, exact: true })).toBeVisible();
if (name === 'remaining') { await expect(panel).toContainText('输入可容纳于当前上下文窗口'); await expect(panel).toContainText('29,600'); await expect(panel).toContainText('28,339'); }
assert.ok(await page.evaluate(() => document.documentElement.scrollWidth <= window.innerWidth + 2));
await screenshot(width + '-budget-' + name);
mark(width + 'px ' + name + ' diagnosis renders actual limit, numbers and historical fallback without overflow');
}
}
assert.deepEqual(errors, []);
const result = { success: true, version: (await host.api('/bootstrap')).version, appRoot, checks, errors, screenshots,
fixtureRequests: requests, realModelRequests: 0, taskId: done.execution.task.id, runs: done.execution.runs.map(run => ({ id: run.id, status: run.status, stopReason: run.stopReason })),
scope: 'Declared local HTTP model fixture with actual engineering tools/checks, persisted review state and seeded diagnostic cases. No real model or user project.', browser: browser.version() };
fs.writeFileSync(path.join(output, 'result.json'), JSON.stringify(result, null, 2) + '\n', { flag: 'wx' });
console.log(JSON.stringify({ success: true, checks: checks.length, output }));
} catch (error) {
if (page) await screenshot('failure').catch(() => {});
fs.writeFileSync(path.join(output, 'result.json'), JSON.stringify({ success: false, checks, errors, screenshots, error: error.stack, requests }, null, 2) + '\n', { flag: 'wx' });
throw error;
} finally { if (browser) await browser.close(); for (const cleanup of cleanups.reverse()) await cleanup(); }
+1 -1
View File
@@ -149,7 +149,7 @@ try {
const beforeRestart = snapshotPersistentData(dataRoot); await controller.stop(); const restarted = (await controller.start()).identity;
assert.notEqual(restarted.pid, original.pid); assert.deepEqual(snapshotPersistentData(dataRoot), beforeRestart);
await controller.api('/model-center/probe', { method: 'POST', input: { id: connectionId } }); assert.equal(authenticated, 3); record(stage, { originalPid: original.pid, restartedPid: restarted.pid, authenticatedRequests: authenticated });
if (version === '1.24.0') {
if (['1.24.0', '1.24.1'].includes(version)) {
stage = 'actual installed current results and saved HTML user paths';
const pointer = JSON.parse(fs.readFileSync(path.join(installRoot, 'current.json')));
const installedRoot = fs.realpathSync.native(path.join(installRoot, 'versions', pointer.active.directory)), proofs = [];
+96
View File
@@ -0,0 +1,96 @@
import fs from 'node:fs';
import path from 'node:path';
import assert from 'node:assert/strict';
import { createHash } from 'node:crypto';
import { Store } from '../src/server/store.mjs';
import { RequestPlanner, historyGroups, verifyPlanEvidence } from '../src/core/context/request-plan.mjs';
if (process.argv.includes('--help')) {
console.log('node scripts/verify-live-context-retention.mjs OUTPUT [retained|original] — offline replay of sealed E source/context, no model calls. OUTPUT must be a new directory.');
process.exit(0);
}
assert.ok(process.argv[2], 'Choose a new output directory.');
const destination = path.resolve(process.argv[2]);
assert.ok(!fs.existsSync(destination), 'Keep prior evidence; choose a new directory.');
fs.mkdirSync(destination, { recursive: true });
const root = path.resolve('docs/live-original-purge-e-evidence');
const sha = value => createHash('sha256').update(value).digest('hex');
const read = name => JSON.parse(fs.readFileSync(path.join(root, name)));
const manifest = read('manifest.json');
for (const entry of manifest.entries) {
const raw = fs.readFileSync(path.join(root, entry.path));
assert.equal(raw.length, entry.bytes); assert.equal(sha(raw), entry.sha256);
}
const sample = manifest.samples.find(sample => sample.name === 'react');
const exported = read('owned-exports/' + sample.sessionId + '.json');
const original = exported.session;
const rows = read('receipts/owned-records.json');
const originalCore = {};
for (const row of rows) (originalCore[row.kind] ??= new Map()).set(row.id, JSON.parse(row.payload));
const artifacts = new Map(manifest.artifacts.map(artifact => [artifact.id, artifact]));
function rawArtifact(id) { return fs.readFileSync(path.join(root, artifacts.get(id).file), 'utf8'); }
const plans = [...originalCore.context_plan.values()].filter(plan => plan.sessionId === sample.sessionId).sort((a, b) => a.requestIndex - b.requestIndex);
const reports = [];
for (const prior of plans.filter(plan => plan.requestIndex >= 7)) {
const store = new Store(path.join(destination, 'request-' + prior.requestIndex));
try {
for (const [kind, map] of Object.entries(originalCore)) store.core[kind] = structuredClone(map);
const readNew = store.artifactStore.read.bind(store.artifactStore);
store.artifactStore.read = artifact => artifacts.has(artifact.id) ? rawArtifact(artifact.id) : readNew(artifact);
const session = structuredClone(original);
store.sessions.set(session.id, session);
store.save = () => {}; // Scratch planner records only; never persist original sessions.
const project = store.project.bind(store);
store.project = id => id === exported.project.id ? exported.project : project(id);
session.providerMessages = session.providerMessages.slice(0, prior.originalMessages);
session.changes = session.changes.filter(change => change.createdAt <= prior.createdAt);
const run = store.core.run.get(prior.runId), task = store.core.task.get(prior.taskId);
// Restore this request's exact remaining credits, not the final run's usage.
for (const [record, remaining] of [[run, prior.budget.runRemainingTokens], [task, prior.budget.taskRemainingTokens]]) {
record.usage = { prompt_tokens: 0, completion_tokens: 0, total_tokens: 0, estimatedUnknownTokens: 0 };
record.budget.maxTokens = remaining ?? 0;
}
run.consumption = []; // Context selection only; no fabricated upstream usage.
const request = JSON.parse(rawArtifact(prior.requestArtifactId));
const wire = request.variants?.[0].body ?? request;
const config = { ...run.model, modelRouting: run.model.routing, streaming: wire.stream, nativeParameters: run.model.nativeParameters ?? {} };
const planner = new RequestPlanner(store);
const prepared = planner.prepare(session, run, config, { system: request.messages[0].content, tools: wire.tools });
assert.equal(prepared.record.budget.inputBudget, prior.budget.inputBudget);
assert.ok(prepared.record.estimatedInputTokens <= prepared.record.budget.inputBudget);
verifyPlanEvidence(store, session, prepared.record);
const sources = [];
for (const group of historyGroups(session.providerMessages)) for (const message of group.messages.filter(message => message.role === 'tool')) {
const call = group.messages[0].tool_calls.find(call => call.id === message.tool_call_id);
const value = JSON.parse(message.content);
if (call.function.name === 'read_file' && !value.truncated && value.start_line === 1 && value.end_line === value.total_lines) {
sources.push({ path: value.path, content: value.content });
}
}
function strings(value, out = []) {
if (typeof value === 'string') {
out.push(value);
for (const line of [value, ...value.split('\n').filter(line => /^[\[{]/.test(line))]) { try { strings(JSON.parse(line), out); } catch {} }
} else if (value && typeof value === 'object') for (const item of Object.values(value)) strings(item, out);
return out;
}
const sent = strings(prepared.messages);
const report = { request: prior.requestIndex, originalInputTokens: prior.estimatedInputTokens,
inputTokens: prepared.record.estimatedInputTokens, inputBudget: prior.budget.inputBudget,
softTarget: prior.budget.targetInputTokens,
sources: [...new Map(sources.map(source => [source.path, source])).values()].map(source => ({ path: source.path,
characters: source.content.length, retained: sent.some(text => text.includes(source.content)) })) };
reports.push(report);
fs.writeFileSync(path.join(destination, 'request-' + prior.requestIndex + '.json'), JSON.stringify({ report, plan: prepared.record, body: prepared.body }, null, 2), { flag: 'wx' });
} finally { store.close(); }
}
const result = { schemaVersion: 1, manifestSha256: sha(fs.readFileSync(path.join(root, 'manifest.json'))),
scope: 'Offline context-selection counterfactual using original E system/tools/history/source bytes and remaining credits. No model requests or task execution. Historical outcomes unchanged.',
realModelRequests: 0, reports };
if (process.argv[3] === 'retained') assert.ok(reports.every(report => report.sources.length === 6 && report.sources.every(source => source.retained)));
if (process.argv[3] === 'original') {
assert.ok(reports.every(report => report.inputTokens === report.originalInputTokens));
assert.deepEqual(reports.map(report => report.sources.filter(source => source.retained).length), [6, 5, 5, 4, 2, 2]);
}
fs.writeFileSync(path.join(destination, 'result.json'), JSON.stringify(result, null, 2) + '\n', { flag: 'wx' });
console.log(JSON.stringify(result));
+7 -7
View File
@@ -34,7 +34,7 @@ for (const name of Object.keys(env)) if (name.toLowerCase() === 'path') delete e
const windowsRoot = process.env.SystemRoot || 'C:\\Windows';
env.PATH = [windowsRoot, path.join(windowsRoot, 'System32'), path.join(windowsRoot, 'System32', 'WindowsPowerShell', 'v1.0')].join(path.delimiter);
let externalGit = null;
if (['1.14.0', '1.15.0', '1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1','1.21.0', '1.22.0', '1.22.1', '1.23.0', '1.24.0'].includes(manifest.version)) {
if (['1.14.0', '1.15.0', '1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1','1.21.0', '1.22.0', '1.22.1', '1.23.0', '1.24.0', '1.24.1'].includes(manifest.version)) {
// The controlled project requires external Git. Preserve its observed binary
// directory while keeping system Node/npm unavailable to the packaged host.
const binary = fs.realpathSync.native(execFileSync('where.exe', ['git.exe'], { env: gitEnvironment(), encoding: 'utf8', windowsHide: true, timeout: 10000 }).trim().split(/\r?\n/)[0]);
@@ -88,7 +88,7 @@ try {
console.log((await launch()).trim());
assert.equal(JSON.parse(fs.readFileSync(path.join(data, 'service-info.json'), 'utf8')).pid, firstInfo.pid);
const example = bootstrap.projects.find(project => project.id === bootstrap.exampleProjectId);
if (['1.15.0', '1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1','1.21.0', '1.22.0', '1.22.1', '1.23.0', '1.24.0'].includes(manifest.version)) {
if (['1.15.0', '1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1','1.21.0', '1.22.0', '1.22.1', '1.23.0', '1.24.0', '1.24.1'].includes(manifest.version)) {
assert.equal(path.resolve(example.root).toLowerCase(), path.join(data, 'projects', 'starter-project').toLowerCase());
const compareTemplate = relative => { for (const entry of fs.readdirSync(path.join(appRoot, 'examples/starter-project', relative), { withFileTypes: true })) {
const next = path.join(relative, entry.name); if (entry.isDirectory()) compareTemplate(next);
@@ -114,7 +114,7 @@ try {
const checkLog = await api('/artifacts/' + checked.results[0].logArtifactId + '?session=' + session.id); assert.match(checkLog.content, /tests 4/);
assert.ok(fs.existsSync(path.join(appRoot, 'runtime/node_modules/npm/bin/npm-cli.js')));
const actualRea = [];
if (['1.8.0', '1.9.0', '1.10.0', '1.10.1', '1.11.0', '1.12.0', '1.12.1', '1.13.0', '1.14.0', '1.15.0', '1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1','1.21.0', '1.22.0', '1.22.1', '1.23.0', '1.24.0'].includes(manifest.version)) {
if (['1.8.0', '1.9.0', '1.10.0', '1.10.1', '1.11.0', '1.12.0', '1.12.1', '1.13.0', '1.14.0', '1.15.0', '1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1','1.21.0', '1.22.0', '1.22.1', '1.23.0', '1.24.0', '1.24.1'].includes(manifest.version)) {
const installation = await api('/research/availability'); assert.equal(installation.status, 'available'); assert.equal(installation.version, '6.8.0');
assert.equal(fs.realpathSync.native(installation.packageRoot), fs.realpathSync.native(path.join(appRoot, 'runtime/node_modules/rea-agents')));
assert.equal(fs.realpathSync.native(installation.node.path), fs.realpathSync.native(path.join(appRoot, 'runtime/node.exe')));
@@ -146,7 +146,7 @@ try {
console.log('PASS actual contained REA: ASAR, PE imports and PE resources; zero target execution/model requests');
}
let actualEngine = null;
if (['1.9.0', '1.10.0', '1.10.1', '1.11.0', '1.12.0', '1.12.1', '1.13.0', '1.14.0', '1.15.0', '1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1','1.21.0', '1.22.0', '1.22.1', '1.23.0', '1.24.0'].includes(manifest.version) && process.argv.includes('--native-engines')) {
if (['1.9.0', '1.10.0', '1.10.1', '1.11.0', '1.12.0', '1.12.1', '1.13.0', '1.14.0', '1.15.0', '1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1','1.21.0', '1.22.0', '1.22.1', '1.23.0', '1.24.0', '1.24.1'].includes(manifest.version) && process.argv.includes('--native-engines')) {
const installation = await api('/engines/deepseek/availability'); assert.equal(installation.status, 'available', installation.message);
assert.equal(fs.realpathSync.native(installation.node.path), fs.realpathSync.native(path.join(appRoot, 'runtime/node.exe')));
const engineRoot = path.join(data, 'engine-project'); fs.mkdirSync(engineRoot); fs.writeFileSync(path.join(engineRoot, 'README.md'), '# Owned portable engine connection\n');
@@ -166,8 +166,8 @@ try {
assert.equal((await api('/research')).length, 4);
const page = await fetch(url); assert.equal(page.status, 200);
assert.match(await page.text(), /Agent Workbench/);
const actualBrowser = ['1.10.0', '1.10.1', '1.11.0', '1.12.0', '1.12.1', '1.13.0', '1.14.0', '1.15.0', '1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1','1.21.0', '1.22.0', '1.22.1', '1.23.0', '1.24.0'].includes(manifest.version) ? await verifyBrowserApi({ api, url, token: bootstrap.token, directory: data, appRoot, contained: true }) : null;
const actualControlled = ['1.14.0', '1.15.0', '1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1','1.21.0', '1.22.0', '1.22.1', '1.23.0', '1.24.0'].includes(manifest.version) ? await verifyControlledApi({ api, url, token: bootstrap.token, directory: data, appRoot, contained: true }) : null;
const actualBrowser = ['1.10.0', '1.10.1', '1.11.0', '1.12.0', '1.12.1', '1.13.0', '1.14.0', '1.15.0', '1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1','1.21.0', '1.22.0', '1.22.1', '1.23.0', '1.24.0', '1.24.1'].includes(manifest.version) ? await verifyBrowserApi({ api, url, token: bootstrap.token, directory: data, appRoot, contained: true }) : null;
const actualControlled = ['1.14.0', '1.15.0', '1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1','1.21.0', '1.22.0', '1.22.1', '1.23.0', '1.24.0', '1.24.1'].includes(manifest.version) ? await verifyControlledApi({ api, url, token: bootstrap.token, directory: data, appRoot, contained: true }) : null;
console.log((await stop()).trim()); started = false;
await until(async () => { try { await fetch(url + '/api/health'); return false; } catch { return true; } });
for (const file of manifest.files) {
@@ -176,7 +176,7 @@ try {
}
const result = { success: true, date: new Date().toISOString(), appRoot, version: manifest.version, nodeVersion: manifest.nodeVersion,
verifiedFiles: manifest.files.length, url, reusedPid: firstInfo.pid, checks: { jobId, status: checked.status, tests: checked.results[0].tests }, actualRea, actualEngine, actualBrowser, actualControlled, externalGit,
steps: ['manifest hashes', 'bundled Node launch', 'same-service reuse', 'packaged example and static page', ...(['1.15.0', '1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1','1.21.0', '1.22.0', '1.22.1', '1.23.0', '1.24.0'].includes(manifest.version) ? ['starter project copied to owned user data; original template bytes matched'] : []), 'explicit command consent', 'actual bundled Node version', 'actual bundled npm check without system Node/npm', 'check log artifact', ...(actualRea.length ? ['contained REA installation and Node identity', 'actual ASAR/imports/resources', 'original Bundle and target hash binding', 'verified reuse without producer replay'] : []), 'report availability', 'graceful stop', 'every original packaged file remains unchanged'], realModel: false };
steps: ['manifest hashes', 'bundled Node launch', 'same-service reuse', 'packaged example and static page', ...(['1.15.0', '1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1','1.21.0', '1.22.0', '1.22.1', '1.23.0', '1.24.0', '1.24.1'].includes(manifest.version) ? ['starter project copied to owned user data; original template bytes matched'] : []), 'explicit command consent', 'actual bundled Node version', 'actual bundled npm check without system Node/npm', 'check log artifact', ...(actualRea.length ? ['contained REA installation and Node identity', 'actual ASAR/imports/resources', 'original Bundle and target hash binding', 'verified reuse without producer replay'] : []), 'report availability', 'graceful stop', 'every original packaged file remains unchanged'], realModel: false };
fs.writeFileSync('.qa/release-validation.json', JSON.stringify(result, null, 2));
console.log(JSON.stringify(result, null, 2));
} finally { if (started) { try { await stop(); } catch {} } }
+20 -20
View File
@@ -19,7 +19,7 @@ assert.ok(fs.existsSync(npmCli), 'Use Node with its bundled npm.');
const files = (root, relative = '') => fs.readdirSync(path.join(root, relative), { withFileTypes: true }).sort((a, b) => a.name.localeCompare(b.name))
.flatMap(entry => entry.isDirectory() ? files(root, path.join(relative, entry.name)) : [path.join(relative, entry.name)]);
const digest = file => createHash('sha256').update(fs.readFileSync(file)).digest('hex');
const buildFolders = ['dist-client', 'dist-server', ...(['1.15.0', '1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1','1.21.0', '1.22.0', '1.23.0', '1.24.0'].includes(evidence.version) ? ['dist-desktop'] : [])];
const buildFolders = ['dist-client', 'dist-server', ...(['1.15.0', '1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1','1.21.0', '1.22.0', '1.23.0', '1.24.0', '1.24.1'].includes(evidence.version) ? ['dist-desktop'] : [])];
const expected = Object.fromEntries(buildFolders.flatMap(folder => files(path.join(workspace, folder)).map(relative => [path.join(folder, relative), digest(path.join(workspace, folder, relative))])));
const allowedEnv = /^(path|pathext|systemroot|systemdrive|windir|comspec|temp|tmp|userprofile|homedrive|homepath|home|appdata|localappdata|programfiles(?:\(x86\))?|commonprogramfiles(?:\(x86\))?|lang|lc_all|AGENT_WORKBENCH_TEST_PYTHON|AGENT_WORKBENCH_DEEPSEEK_ROOT|AGENT_WORKBENCH_PURGE_ROOT)$/i;
const env = Object.fromEntries(Object.entries(process.env).filter(([key]) => allowedEnv.test(key)));
@@ -30,7 +30,7 @@ const directory = path.join(sourceRoot, '.qa', 'package-validation-' + randomUUI
fs.mkdirSync(directory, { recursive: true });
const inputFolders = ['src', 'tests', 'scripts', ...['benchmarks', 'public', 'examples', 'research'].filter(folder => fs.existsSync(path.join(sourceRoot, folder)))];
const inputNames = [...inputFolders.flatMap(folder => files(path.join(sourceRoot, folder)).map(relative => path.join(folder, relative))),
'package.json', 'package-lock.json', 'vite.config.ts', 'tsconfig.json', 'index.html', ...(['1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1','1.21.0', '1.22.0', '1.23.0', '1.24.0'].includes(evidence.version) ? ['mobile.html'] : []), ...['.gitattributes', '.gitignore', 'Start-Desktop.cmd', 'Start-Workbench.cmd', 'Stop-Workbench.cmd'].filter(name => fs.existsSync(path.join(sourceRoot, name)))].sort();
'package.json', 'package-lock.json', 'vite.config.ts', 'tsconfig.json', 'index.html', ...(['1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1','1.21.0', '1.22.0', '1.23.0', '1.24.0', '1.24.1'].includes(evidence.version) ? ['mobile.html'] : []), ...['.gitattributes', '.gitignore', 'Start-Desktop.cmd', 'Start-Workbench.cmd', 'Stop-Workbench.cmd'].filter(name => fs.existsSync(path.join(sourceRoot, name)))].sort();
const snapshot = () => inputNames.map(relative => ({ path: relative.replaceAll('\\', '/'), bytes: fs.statSync(path.join(sourceRoot, relative)).size, sha256: digest(path.join(sourceRoot, relative)) }));
const inputs = snapshot(); fs.writeFileSync(path.join(directory, 'inputs.json'), JSON.stringify(inputs, null, 2));
let stage = 'environment preflight', python = null;
@@ -83,7 +83,7 @@ try {
assert.deepEqual(actualFiles.slice().sort(), Object.keys(expected).sort());
for (const [relative, sha256] of Object.entries(expected)) assert.equal(digest(path.join(sourceRoot, relative)), sha256, relative);
let actualRea = null;
if (['1.8.0', '1.9.0', '1.10.0', '1.10.1', '1.11.0', '1.12.0', '1.12.1', '1.13.0', '1.14.0', '1.15.0', '1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1','1.21.0', '1.22.0', '1.23.0', '1.24.0'].includes(evidence.version)) {
if (['1.8.0', '1.9.0', '1.10.0', '1.10.1', '1.11.0', '1.12.0', '1.12.1', '1.13.0', '1.14.0', '1.15.0', '1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1','1.21.0', '1.22.0', '1.23.0', '1.24.0', '1.24.1'].includes(evidence.version)) {
stage = 'contained REA';
const reaOutput = await run(['scripts/verify-rea.mjs', '--app-root', sourceRoot, '--contained'], 'rea.log');
actualRea = JSON.parse(reaOutput.trim().split('\n').at(-1));
@@ -91,7 +91,7 @@ try {
assert.ok(actualRea.results.every(item => item.rootExited === true));
}
let actualEngine = null;
if (['1.9.0', '1.10.0', '1.10.1', '1.11.0', '1.12.0', '1.12.1', '1.13.0', '1.14.0', '1.15.0', '1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1','1.21.0', '1.22.0', '1.23.0', '1.24.0'].includes(evidence.version) && process.argv.includes('--native-engines')) {
if (['1.9.0', '1.10.0', '1.10.1', '1.11.0', '1.12.0', '1.12.1', '1.13.0', '1.14.0', '1.15.0', '1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1','1.21.0', '1.22.0', '1.23.0', '1.24.0', '1.24.1'].includes(evidence.version) && process.argv.includes('--native-engines')) {
stage = 'native DeepSeek SDK';
const output = await run(['scripts/verify-engines.mjs'], 'engines.log');
actualEngine = JSON.parse(output.trim().split('\n').at(-1));
@@ -99,7 +99,7 @@ try {
assert.equal(actualEngine.modelRequests, 0); assert.equal(actualEngine.rootExited, true); assert.equal(actualEngine.nativeMethods.length, 6);
}
let actualBrowser = null;
if (['1.10.0', '1.10.1', '1.11.0', '1.12.0', '1.12.1', '1.13.0', '1.14.0', '1.15.0', '1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1','1.21.0', '1.22.0', '1.23.0', '1.24.0'].includes(evidence.version)) {
if (['1.10.0', '1.10.1', '1.11.0', '1.12.0', '1.12.1', '1.13.0', '1.14.0', '1.15.0', '1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1','1.21.0', '1.22.0', '1.23.0', '1.24.0', '1.24.1'].includes(evidence.version)) {
stage = 'contained browser runtime';
const output = await run(['scripts/verify-browser.mjs', '--app-root', sourceRoot, '--contained'], 'browser.log');
actualBrowser = JSON.parse(output.trim().split('\n').at(-1));
@@ -109,7 +109,7 @@ try {
assert.equal(actualBrowser.modelRequests, 0);
}
let actualControlled = null;
if (['1.14.0', '1.15.0', '1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1','1.21.0', '1.22.0', '1.23.0', '1.24.0'].includes(evidence.version)) {
if (['1.14.0', '1.15.0', '1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1','1.21.0', '1.22.0', '1.23.0', '1.24.0', '1.24.1'].includes(evidence.version)) {
stage = 'compiled controlled-engine delivery';
const output = await run(['scripts/verify-controlled-release.mjs', '--app-root', sourceRoot], 'controlled-engine.log');
actualControlled = JSON.parse(output.trim().split('\n').at(-1));
@@ -117,7 +117,7 @@ try {
assert.equal(actualControlled.realModelRequests, 0); assert.equal(actualControlled.cases, 4); assert.ok(actualControlled.fixtureModelRequests > 0);
}
let actualCompanion = null;
if (['1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1','1.21.0', '1.22.0', '1.23.0', '1.24.0'].includes(evidence.version)) {
if (['1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1','1.21.0', '1.22.0', '1.23.0', '1.24.0', '1.24.1'].includes(evidence.version)) {
stage = 'contained companion gateway and UI';
const output = await run(['scripts/verify-companion-ui.mjs', '--app-root', sourceRoot], 'companion-ui.log');
const location = [...output.matchAll(/^EVIDENCE (.+)\r?$/gm)].at(-1)?.[1]?.trim();
@@ -128,7 +128,7 @@ try {
assert.equal(actualCompanion.passed.length, 7); assert.deepEqual(actualCompanion.errors, []); assert.equal(actualCompanion.realModelRequests, 0);
}
let actualProtocols = null;
if (['1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1','1.21.0', '1.22.0', '1.23.0', '1.24.0'].includes(evidence.version)) {
if (['1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1','1.21.0', '1.22.0', '1.23.0', '1.24.0', '1.24.1'].includes(evidence.version)) {
stage = 'contained model protocols and mobile streams';
const output = await run(['scripts/verify-model-protocols-ui.mjs', '--app-root', sourceRoot], 'model-protocols-ui.log');
const location = [...output.matchAll(/^EVIDENCE (.+)\r?$/gm)].at(-1)?.[1]?.trim();
@@ -139,7 +139,7 @@ try {
assert.equal(actualProtocols.passed.length, 5); assert.deepEqual(actualProtocols.pageErrors, []); assert.deepEqual(actualProtocols.requestFailures, []);
}
let actualVision = null;
if (['1.18.0', '1.19.0', '1.20.0', '1.20.1','1.21.0', '1.22.0', '1.23.0', '1.24.0'].includes(evidence.version)) {
if (['1.18.0', '1.19.0', '1.20.0', '1.20.1','1.21.0', '1.22.0', '1.23.0', '1.24.0', '1.24.1'].includes(evidence.version)) {
stage = 'contained screenshot inputs and exact Task budget resume';
const output = await run(['scripts/verify-vision-ui.mjs', '--app-root', sourceRoot], 'vision-ui.log');
const location = [...output.matchAll(/^EVIDENCE (.+)\r?$/gm)].at(-1)?.[1]?.trim();
@@ -150,7 +150,7 @@ try {
assert.equal(actualVision.passed.length, 5); assert.deepEqual(actualVision.pageErrors, []); assert.equal(actualVision.realModelRequests, 0);
}
let actualProjectCache = null;
if (['1.19.0', '1.20.0', '1.20.1','1.21.0', '1.22.0', '1.23.0', '1.24.0'].includes(evidence.version)) {
if (['1.19.0', '1.20.0', '1.20.1','1.21.0', '1.22.0', '1.23.0', '1.24.0', '1.24.1'].includes(evidence.version)) {
stage = 'contained source-verified parsing cache controls';
const output = await run(['scripts/verify-project-cache-ui.mjs', '--app-root', sourceRoot], 'project-cache-ui.log');
const location = [...output.matchAll(/^EVIDENCE (.+)\r?$/gm)].at(-1)?.[1]?.trim();
@@ -161,7 +161,7 @@ try {
assert.equal(actualProjectCache.realModelRequests, 0); assert.deepEqual(actualProjectCache.pageErrors, []);
}
let actualEnvironment = null, actualEnvironmentEngineering = null;
if (['1.21.0','1.22.0', '1.23.0', '1.24.0'].includes(evidence.version)) {
if (['1.21.0','1.22.0', '1.23.0', '1.24.0', '1.24.1'].includes(evidence.version)) {
stage = 'contained environment approval, stop and recovery UI';
const output = await run(['scripts/verify-environment-ui.mjs', '--app-root', sourceRoot], 'environment-ui.log');
const summary = JSON.parse(output.trim().split(/\r?\n/).findLast(line => line.startsWith('{"success":')));
@@ -179,7 +179,7 @@ try {
assert.equal(actualEnvironmentEngineering.pythonTests, 2); assert.equal(actualEnvironmentEngineering.realModelRequests, 0);
}
let actualEmbeddedBrowser = null;
if (['1.22.0', '1.23.0', '1.24.0'].includes(evidence.version)) {
if (['1.22.0', '1.23.0', '1.24.0', '1.24.1'].includes(evidence.version)) {
stage = 'actual embedded Electron, Vite and original mobile observation';
await run(['scripts/verify-embedded-browser.mjs', '--app-root', sourceRoot], 'embedded-browser.log');
actualEmbeddedBrowser = JSON.parse(fs.readFileSync(path.join(sourceRoot, '.qa/embedded-browser-latest.json'), 'utf8'));
@@ -189,7 +189,7 @@ try {
assert.equal(actualEmbeddedBrowser.realModelRequests, 0); assert.deepEqual(actualEmbeddedBrowser.errors, []);
}
let actualReleaseEfficiency = null;
if (['1.23.0', '1.24.0'].includes(evidence.version)) {
if (['1.23.0', '1.24.0', '1.24.1'].includes(evidence.version)) {
stage = 'actual update status Edge/Electron and basic saved HTML flows';
const proofs = [];
for (const script of ['verify-release-efficiency-ui.mjs', 'verify-local-files.mjs']) {
@@ -203,7 +203,7 @@ try {
actualReleaseEfficiency = proofs;
}
let actualFileResults = null;
if (evidence.version === '1.24.0') {
if (['1.24.0', '1.24.1'].includes(evidence.version)) {
stage = 'current file results Edge, Electron and paired mobile observation';
await run(['scripts/verify-file-results-ui.mjs'], 'file-results-ui.log');
actualFileResults = JSON.parse(fs.readFileSync(path.join(sourceRoot, '.qa/file-results-ui-latest.json'), 'utf8'));
@@ -243,14 +243,14 @@ try {
for (const key of ['fail', 'cancelled', 'skipped', 'todo']) assert.equal(groupStats[key], 0);
}
let actualTaskControl = null;
if (['1.20.1', '1.21.0', '1.22.0', '1.23.0', '1.24.0'].includes(evidence.version)) {
if (['1.20.1', '1.21.0', '1.22.0', '1.23.0', '1.24.0', '1.24.1'].includes(evidence.version)) {
stage = 'task control desktop and 390px';
await run(['scripts/verify-task-control-ui.mjs', '--app-root', sourceRoot], 'task-control-ui.log');
actualTaskControl = JSON.parse(fs.readFileSync(path.join(sourceRoot, '.qa/task-control-ui-validation.json'), 'utf8'));
assert.equal(actualTaskControl.success, true); assert.equal(actualTaskControl.passed.length, 5); assert.equal(actualTaskControl.realModelRequests, 0);
}
let actualDesktop = null;
if (['1.15.0', '1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1','1.21.0', '1.22.0', '1.23.0', '1.24.0'].includes(evidence.version)) {
if (['1.15.0', '1.16.0', '1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1','1.21.0', '1.22.0', '1.23.0', '1.24.0', '1.24.1'].includes(evidence.version)) {
stage = 'actual contained Electron window'; await run(['scripts/verify-desktop-window.mjs'], 'desktop-window.log');
const window = JSON.parse(fs.readFileSync(path.join(sourceRoot, '.qa/desktop-window-latest.json'), 'utf8'));
assert.equal(window.success, true); assert.equal(path.resolve(window.appRoot), sourceRoot); assert.equal(window.results.length, 7);
@@ -264,7 +264,7 @@ try {
actualDesktop = { window, ui, projects, scope: 'Independently rebuilt source and its shipped Electron binary; original window/engine/engineering assertions. Own credentials/HTTP fixtures; Windows native picker and complete host crash/stop have separate actual evidence.' };
}
let actualProtocolEngineering = null;
if (['1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1','1.21.0', '1.22.0', '1.23.0', '1.24.0'].includes(evidence.version)) {
if (['1.17.0', '1.18.0', '1.19.0', '1.20.0', '1.20.1','1.21.0', '1.22.0', '1.23.0', '1.24.0', '1.24.1'].includes(evidence.version)) {
stage = 'Responses SSE actual Node React Python engineering';
await run(['scripts/verify-controlled-projects.mjs', '--responses'], 'responses-engineering.log');
const location = JSON.parse(fs.readFileSync(path.join(sourceRoot, '.qa/responses-engineering-location.json'), 'utf8'));
@@ -274,7 +274,7 @@ try {
assert.deepEqual(actualProtocolEngineering.results.map(value => value.parentCheck.results[0].tests.passed), [8, 9, 10]);
}
let actualVisionEngineering = null;
if (['1.18.0', '1.19.0', '1.20.0', '1.20.1','1.21.0', '1.22.0', '1.23.0', '1.24.0'].includes(evidence.version)) {
if (['1.18.0', '1.19.0', '1.20.0', '1.20.1','1.21.0', '1.22.0', '1.23.0', '1.24.0', '1.24.1'].includes(evidence.version)) {
stage = 'screenshot Task Node React Python engineering';
await run(['scripts/verify-vision-projects.mjs'], 'vision-engineering.log');
const location = JSON.parse(fs.readFileSync(path.join(sourceRoot, '.qa/vision-engineering-location.json'), 'utf8'));
@@ -284,7 +284,7 @@ try {
assert.deepEqual(actualVisionEngineering.results.map(value => value.check.results[0].tests.passed), [8, 9, 10]);
}
let actualCacheEngineering = null;
if (['1.19.0', '1.20.0', '1.20.1','1.21.0', '1.22.0', '1.23.0', '1.24.0'].includes(evidence.version)) {
if (['1.19.0', '1.20.0', '1.20.1','1.21.0', '1.22.0', '1.23.0', '1.24.0', '1.24.1'].includes(evidence.version)) {
stage = 'cached index original requests and Node React Python engineering';
await run(['scripts/verify-project-cache-projects.mjs'], 'project-cache-engineering.log');
const location = JSON.parse(fs.readFileSync(path.join(sourceRoot, '.qa/project-cache-engineering-location.json'), 'utf8'));
@@ -294,7 +294,7 @@ try {
assert.deepEqual(actualCacheEngineering.results.map(value => value.check.results[0].tests.passed), [8, 9, 10]);
}
let actualEvaluations = null, actualEvaluationEngineering = null;
if (['1.20.0', '1.20.1','1.21.0', '1.22.0', '1.23.0', '1.24.0'].includes(evidence.version)) {
if (['1.20.0', '1.20.1','1.21.0', '1.22.0', '1.23.0', '1.24.0', '1.24.1'].includes(evidence.version)) {
stage = 'compiled fixed engineering evaluations';
const output = await run(['scripts/verify-evaluations-release.mjs', '--app-root', sourceRoot], 'evaluations-release.log');
const location = [...output.matchAll(/^EVIDENCE (.+)\r?$/gm)].at(-1)?.[1]?.trim();
+5 -3
View File
@@ -4,10 +4,10 @@ import { api } from './api';
import { Modal } from './components';
import type { Config,Session } from './types';
import './input-attachments.css';
import { budgetPausePresentation } from '../shared/budget-presentation.mjs';
type Fields={contextLimit:number;maxOutputTokens:number;maxRunTokens:number;maxTaskTokens:number;maxSteps:number;maxRunDurationMs:number};
type Preview={id:string;sha256:string;taskId:string;runId:string;config:Fields&{model:string};eligibleConnections:{id:string;name:string;model:string}[];budget:{inputBudget:number;contextLimit:number};completedToolIds:string[];unknownToolIds:string[];attachmentIds:string[];note:string};
const reasons:Record<string,string>={context_limit:'上下文窗口不足',engine_input_budget:'原生引擎输入或剩余预算不足',token_limit:'本次运行预算不足',task_token_limit:'任务累计预算不足',time_limit:'本次运行时间已用完'};
const labels:Record<string,string>={transportAndImageFramingTokens:'请求格式与图片说明',goalAndContextTextTokens:'目标、全部要求及工程上下文',toolDefinitionTokens:'工具定义',historyAndEvidenceTokens:'必需历史及实际依据',imageReservationTokens:'图像本地预留',requestOverheadTokens:'请求格式',fixedInstructionTokens:'完整目标与原生提示',retainedHistoryAndViewTokens:'保留历史与工具结果'};
export function BudgetPausePanel({session,config,onSession,onRecovery}:{session:Session;config:Config;onSession:(value:Session)=>void;onRecovery?:()=>void}){
@@ -16,13 +16,15 @@ export function BudgetPausePanel({session,config,onSession,onRecovery}:{session:
maxRunTokens:config.maxRunTokens,maxTaskTokens:session.execution?.task?.budget.maxTokens??config.maxTaskTokens,maxSteps:config.maxSteps,maxRunDurationMs:config.maxRunDurationMs});
const [selection,setSelection]=useState(''),[options,setOptions]=useState<{value:string;name:string}[]>([]);
const run=session.execution?.run,pause=run?.budgetPause,reason=pause?.reason??run?.stopReason??'';
const diagnosis=budgetPausePresentation(pause,reason);
const controlled=session.execution?.task?.origin==='isolated-worker'||run?.engine==='deepseek-workbench';
const calibrationSources=new Set(pause?.tokenReservation?.routes.flatMap(route=>route.sources.map(source=>source.attemptId))??[]).size;
if(!['paused','failed'].includes(session.status)||!reasons[reason])return null;
if(!['paused','failed'].includes(session.status)||!diagnosis)return null;
async function start(){setOpen(true);setError('');setPlan(null);try{const center=await api<{selected:{type:string;id:string};connections:{id:string;name:string;contextLimit:number}[];pools:{id:string;name:string}[]}>('/model-center');setSelection(center.selected.type+':'+center.selected.id);setOptions([...center.connections.map(row=>({value:(row.id==='legacy-default'?'legacy':'connection')+':'+row.id,name:row.name+' · '+row.contextLimit.toLocaleString()+' 上下文'})),...center.pools.map(row=>({value:'pool:'+row.id,name:row.name+' · 按成员最小容量'}))]);}catch(cause){setError((cause as Error).message);}}
async function preview(){setBusy(true);setError('');setPlan(null);try{const [type,id]=selection.split(':');const value=await api<Preview>('/sessions/'+session.id+'/resume-preview',{...fields,...(selection?{selection:{type,id}}:{})});setPlan(value);onSession(await api<Session>('/sessions/'+session.id));}catch(cause){setError((cause as Error).message);}finally{setBusy(false);}}
async function confirm(){if(!plan)return;setBusy(true);setError('');try{onSession(await api<Session>('/sessions/'+session.id+'/resume-budget',{planId:plan.id,sha256:plan.sha256}));setOpen(false);}catch(cause){setError((cause as Error).message);}finally{setBusy(false);}}
return <aside className="budget-pause" aria-label="任务预算诊断"><h3>{reasons[reason]}</h3>{pause?.requiredInputTokens!==undefined?<><p>完整要求与必需输入估算 <strong>{pause.requiredInputTokens.toLocaleString()}</strong> tokens;当前可用输入 <strong>{pause.inputBudget?.toLocaleString()}</strong>。请求已暂停,原任务和工具结果保留。</p><dl className="budget-facts">{Object.entries(pause.breakdown??{}).map(([key,value])=><div key={key} style={{display:'contents'}}><dt>{labels[key]??key}</dt><dd>{value.toLocaleString()}</dd></div>)}<dt>所选上下文窗口</dt><dd>{pause.contextLimit?.toLocaleString()}</dd><dt>响应预留 / 安全余量</dt><dd>{pause.outputReservation?.toLocaleString()} / {pause.safetyTokens?.toLocaleString()}</dd><dt>运行 / 任务剩余</dt><dd>{pause.runRemainingTokens?.toLocaleString()??'不限'} / {pause.taskRemainingTokens?.toLocaleString()??'不限'}</dd></dl></>:<p>原 Task 已保存。可核对剩余预算和模型容量,再准备同一任务的新 Run;旧记录没有的预算分解不补造。</p>}
return <aside className="budget-pause" aria-label="任务预算诊断"><h3>{diagnosis.title}</h3>{pause?.requiredInputTokens!==undefined?<><p>完整要求与必需输入估算 <strong>{pause.requiredInputTokens.toLocaleString()}</strong> tokens;{diagnosis.inputLabel} <strong>{pause.inputBudget?.toLocaleString()}</strong>。请求已暂停,原任务和工具结果保留。</p><dl className="budget-facts">{Object.entries(pause.breakdown??{}).map(([key,value])=><div key={key} style={{display:'contents'}}><dt>{labels[key]??key}</dt><dd>{value.toLocaleString()}</dd></div>)}<dt>所选上下文窗口</dt><dd>{pause.contextLimit?.toLocaleString()}</dd><dt>响应预留 / 安全余量</dt><dd>{pause.outputReservation?.toLocaleString()} / {pause.safetyTokens?.toLocaleString()}</dd><dt>运行 / 任务剩余</dt><dd>{pause.runRemainingTokens?.toLocaleString()??'不限'} / {pause.taskRemainingTokens?.toLocaleString()??'不限'}</dd></dl></>:<p>原 Task 已保存。可核对剩余预算和模型容量,再准备同一任务的新 Run;旧记录没有的预算分解不补造。</p>}
{diagnosis.guidance&&<p>{diagnosis.guidance}</p>}
{pause?.originalInputTokens!==undefined&&<p>原生完整历史估算 {pause.originalInputTokens.toLocaleString()} tokens。原始输入、历史索引与拒绝正文已保存为任务工件,本次拒绝没有发送模型请求。</p>}
{!!pause?.transportReservationTokens&&<p>准备原生任务时另留 {pause.transportReservationTokens.toLocaleString()} tokens 给引擎提示和消息封装;实际原生请求仍单独核对容量。</p>}
{!!pause?.imageReservationTokens&&<p>图像使用单独的本地预留,属于容量估算;实际计费与服务报告用量分列。</p>}
+1
View File
@@ -18,6 +18,7 @@ import { FileResultsPanel } from './FileResultsPanel';
const stopReasons: Record<string, string> = { no_progress: '重复读取未推进任务,请核对后继续', context_limit: '完整任务要求超过请求上下文预算', time_limit: '到达运行时间预算', token_limit: '到达本次 token 预算', task_token_limit: '到达任务累计预算', step_limit: '到达工具轮次上限', cancelled: '用户停止', model_error: '模型请求失败', interrupted: '进程中断' };
const number = (value: number | undefined) => (value ?? 0).toLocaleString('zh-CN');
stopReasons.git_review = 'Git预览已保存,等待审查与确认';
stopReasons.change_set_review = '完整文件批次已封存,等待审查与应用';
stopReasons.git_review_declined = 'Git预览已撤回,文件和检查记录保留';
stopReasons.task_plan_incomplete = '执行计划仍有未核验或失效步骤,请核对后继续';
stopReasons.diagnostic_answered = '诊断答复已完成;原任务的目标与交付状态保持';
+1
View File
@@ -47,6 +47,7 @@ export interface InputAttachment { id:string;name:string;status:string;mediaType
export interface BudgetPause { reason:string;requiredInputTokens?:number;contextLimit?:number;inputBudget?:number;outputReservation?:number;safetyTokens?:number;runRemainingTokens?:number|null;taskRemainingTokens?:number|null;imageReservationTokens?:number;breakdown?:Record<string,number> }
export interface Run { budgetPause?:BudgetPause }
export interface BudgetPause { originalInputTokens?:number;transportReservationTokens?:number }
export interface BudgetPause { limitingReason?:string }
export interface BudgetPause { modelFailure?:RuntimeFailure }
export interface TokenBudgetCounter { maximumTokens:number;reportedTokens:number;unknownRequests:number;estimatedUnknownTokens:number;accountedTokens:number;remainingTokens:number|null;exceededTokens:number }
export interface BudgetPause { remainingTokens?:number;tokenReservation?:{schemaVersion:number;reservedTokens:number;routes:{sources:{runId:string;attemptId:string}[]}[]};tokenAccounting?:{schemaVersion:number;singleInFlightResponseMayExceed:boolean;run:TokenBudgetCounter;task:TokenBudgetCounter} }
+20 -14
View File
@@ -181,6 +181,7 @@ export class RequestPlanner {
let omitted = [], reference = null, workingReference = null, latestReference = null;
const latestDocument = latestToolDocument(this.store, session, run, groups);
const entries = [];
let retainedSources = [];
const metricsFor = (indices, includeNotice = true, currentEntries = entries, includeLatest = true) => {
const missing = groups.filter(group => !indices.has(group.index));
const messages = [fixed, ...(missing.length && includeNotice ? [notice({ id: PLACEHOLDER_ID, sha256: PLACEHOLDER_SHA }, missing, descriptions, evidence)] : []),
@@ -194,22 +195,25 @@ export class RequestPlanner {
const mandatory = metricsFor(selected);
if (mandatory.estimatedInputTokens > budget.inputBudget) throw pause(budget, mandatory, breakdown);
const observed = workingCandidates(this.store, session, run, groups);
// Reserve other small source files before admitting a new result group.
// Historical locator pages are optional; they cannot evict actual code.
// Admit complete small source observations before optional raw history.
// A reservation alone does not keep code in the actual sent request.
const newest = groups.at(-1);
const newestFiles = new Set((newest?.messages[0].tool_calls ?? []).filter(call => call.function.name === 'read_file').map(call => {
try { return JSON.parse(call.function.arguments).path; } catch { return null; }
}));
const pinned = observed.filter(candidate => candidate.observation.kind === 'file' && candidate.observation.content.length <= 2000
&& !newestFiles.has(candidate.observation.path)).slice(0, 12).map(candidate => workingEntry(candidate, candidate.observation.content.length));
const retention = Math.min(2048, Math.max(0, metricsFor(selected, true, pinned).estimatedInputTokens - mandatory.estimatedInputTokens));
const retentionTarget = budget.inputBudget;
for (const candidate of observed.filter(candidate => candidate.observation.kind === 'file' && candidate.observation.content.length <= 2000)) {
if (entries.length >= 12) break;
const entry = workingEntry(candidate, candidate.observation.content.length);
if (metricsFor(selected, true, [...entries, entry]).estimatedInputTokens <= retentionTarget) entries.push(entry);
}
retainedSources = [...entries];
const newestUseful = newest?.messages.every(message => {
if (message.role !== 'tool') return true;
try { const result = JSON.parse(message.content); return !result.kind?.startsWith('context-') && result.kind !== 'tool-result'; } catch { return true; }
});
if (newestUseful && !selected.has(newest.index) && newest.contributionTokens <= budget.inputBudget) {
selected.add(newest.index);
if (metricsFor(selected).estimatedInputTokens + retention > budget.inputBudget) selected.delete(newest.index);
const remaining = entries.filter(entry => entry.source.group_index !== newest.index);
if (metricsFor(selected, true, remaining).estimatedInputTokens > budget.inputBudget) selected.delete(newest.index);
else entries.splice(0, entries.length, ...remaining);
}
const target = Math.max(budget.targetInputTokens, metricsFor(selected).estimatedInputTokens);
const priority = [...new Set(evidence.map(value => value.group_index))].map(index => groups[index]);
@@ -220,7 +224,8 @@ export class RequestPlanner {
};
// A small result in a mixed large group must remain usable. It is a
// separate, provenance-checked observation, never a broken tool pair.
const candidates = observed.filter(candidate => !selected.has(candidate.source.group_index));
const candidates = observed.filter(candidate => !selected.has(candidate.source.group_index)
&& !entries.some(entry => entry.source.call_id === candidate.source.call_id));
for (const [index, candidate] of candidates.entries()) {
if (entries.length >= 12) break;
const remaining = target - metricsFor(selected).estimatedInputTokens;
@@ -325,9 +330,9 @@ export class RequestPlanner {
// The next result replaces the current latest-result feedback. Do not
// reserve both views when sizing that result's actual protocol page.
nextReadFixedInputTokens: metricsFor(new Set(lastUser ? [lastUser.index] : []), true, [], false).estimatedInputTokens,
workingRetentionTokens: Math.min(2048, Math.max(0, metricsFor(new Set(lastUser ? [lastUser.index] : []), true,
entries.filter(entry => entry.observation.kind === 'file' && !entry.truncated && entry.total_content_characters <= 2000)).estimatedInputTokens
- metricsFor(new Set(lastUser ? [lastUser.index] : []), true, []).estimatedInputTokens)),
workingRetentionTokens: Math.max(0, metricsFor(new Set(lastUser ? [lastUser.index] : []), true, retainedSources).estimatedInputTokens
- metricsFor(new Set(lastUser ? [lastUser.index] : []), true, []).estimatedInputTokens),
workingRetentionPolicyVersion: 1,
originalMessages: session.providerMessages.length, retainedMessages: messages.length - 1 - (reference ? 1 : 0) - (workingReference ? 1 : 0) - (latestReference ? 1 : 0) - (memoryReference ? 1 : 0) - (skillReference ? 1 : 0) - (mcpReference ? 1 : 0),
archivedGroups: omitted.length, retainedGroups: groups.filter(group => selected.has(group.index)).map(group => group.index),
historyIndexArtifactId: reference?.id ?? null, historyIndexSha256: reference?.sha256 ?? null, actualUsage: null,
@@ -400,7 +405,8 @@ export function verifyPlanEvidence(store, session, plan, cache = new Map()) {
|| (frozen.workingSetSha256 ?? null) !== (plan.workingSetSha256 ?? null)
|| (frozen.workingSetEntries ?? 0) !== (plan.workingSetEntries ?? 0)) throw new Error('工作依据与发送前的封存记录不匹配。');
if ((frozen.workingSetViewVersion ?? 1) !== (plan.workingSetViewVersion ?? 1)
|| (frozen.workingRetentionTokens ?? 0) !== (plan.workingRetentionTokens ?? 0)) throw new Error('工作依据视图或保留预算与封存记录不匹配。');
|| (frozen.workingRetentionTokens ?? 0) !== (plan.workingRetentionTokens ?? 0)
|| (frozen.workingRetentionPolicyVersion ?? 0) !== (plan.workingRetentionPolicyVersion ?? 0)) throw new Error('工作依据视图或保留预算与封存记录不匹配。');
if ((frozen.nextReadFixedInputTokens ?? null) !== (plan.nextReadFixedInputTokens ?? null)) throw new Error('下一工具页的固定预算与封存记录不匹配。');
if ((frozen.latestToolsArtifactId ?? null) !== (plan.latestToolsArtifactId ?? null)
|| (frozen.latestToolsSha256 ?? null) !== (plan.latestToolsSha256 ?? null)
+2 -1
View File
@@ -81,7 +81,8 @@ export function workingCandidates(store, session, run, groups) {
}
// Keep complete small modules/requirements before larger evidence excerpts.
// Every selected page still identifies the exact invocation that read it.
primary.sort((a, b) => Number(a.observation.content.length > 2000) - Number(b.observation.content.length > 2000));
primary.sort((a, b) => Number(a.observation.content.length > 2000) - Number(b.observation.content.length > 2000)
|| a.observation.content.length - b.observation.content.length || a.observation.path.localeCompare(b.observation.path));
const checks = candidates.filter(candidate => ['check-results', 'command-result'].includes(candidate.observation.kind));
const artifacts = candidates.filter(candidate => candidate.observation.kind === 'artifact-page');
return [...primary, ...checks, ...artifacts, ...secondary].slice(0, 24);
+3 -1
View File
@@ -226,7 +226,9 @@ class EngineRunDriver {
const rejected=this.store.artifactStore.write(this.session,this.run.record,
JSON.parse(this.provider.redact(JSON.stringify({schemaVersion:1,owner,view,body,budget:contextBudget,dispatched:false}))),{kind:'engine-model-rejection'});
this.store.core.artifact.set(rejected.id,rejected);(this.job.inputRejections??=[]).push({artifactId:rejected.id,sha256:rejected.sha256,...contextBudget});this.save('input_rejected');
const error=new BudgetExceeded('engine_input_budget','原引擎必需目标、最新结果或剩余尝试超出封存预算,未发送模型请求。原输入和容量分项已保存;请核对上下文/预算后显式继续。');
const error=new BudgetExceeded('engine_input_budget',contextBudget.limitingReason==='context_limit'
?'原引擎必需输入超过上下文窗口可用容量,未发送模型请求。原输入和容量分项已保存;请核对模型容量后显式继续。'
:'原引擎输入可容纳于上下文窗口,但下一次请求的用量预留超过剩余预算,未发送模型请求。原输入和容量分项已保存;请核对运行和任务预算后显式继续。');
error.contextBudget={...contextBudget,rejectionArtifactId:rejected.id,rejectionSha256:rejected.sha256};throw error;
}
const selected = config.modelRouting ? { ...config, modelRouting: { ...config.modelRouting, maxAttempts: attempts } } : config;
+17
View File
@@ -130,6 +130,23 @@ export class ChangeSetManager {
for (let index = 0; index < set.files.length; index++) set.files[index].state = observations[index].state;
return observations;
}
async reviewReady(session, id, run) {
const set = this.own(session, id, null, { current: true });
if (session.runId !== run.id || session.taskId !== run.taskId || set.status !== 'prepared' || set.attemptedAt) return false;
const pending = session.changes.filter(change => change.taskId === run.taskId && ['pending', 'conflict'].includes(change.status));
if (pending.length !== set.files.length || pending.some(change => change.status !== 'pending'
|| !set.files.some(file => file.changeId === change.id))) return false;
if ([...this.store.core.tool.values()].some(tool => tool.sessionId === session.id && tool.taskId === run.taskId && tool.requiresVerification)) return false;
assertMcpResolved(this.store, session.projectId);
assertWorkersResolved(this.store, session.projectId);
this.snapshots(session, set, { pending: true });
const project = this.store.project(session.projectId);
await verifyWorkspaceProject(this.store, project);
const inspection = await inspectProject(project);
return session.runId === run.id && session.taskId === run.taskId && set.status === 'prepared'
&& JSON.stringify(gitBaseline(inspection)) === JSON.stringify(set.baseline)
&& this.inspect(session, set).every(value => value.state === 'before');
}
async prepare(session, changeIds, { source = 'user', collaborationBinding = null } = {}) {
if (!Array.isArray(changeIds) || !changeIds.length || changeIds.length > 32 || new Set(changeIds).size !== changeIds.length) throw conflict('变更集需要1到32个不同的提案ID。');
this.requireIdle(session, { source });
+1 -1
View File
@@ -1,6 +1,6 @@
import { randomUUID, createHash } from 'node:crypto';
export const APP_VERSION = '1.24.0';
export const APP_VERSION = '1.24.1';
export function executionEnvironment() {
const windows = process.platform === 'win32';
+18
View File
@@ -498,6 +498,7 @@ export class Runtime {
this.store.save(session); return;
}
let deliveryReview = null, mcpReview = null;
const preparedSets = new Set();
for (const call of calls) {
let result;
if (deliveryReview || mcpReview) {
@@ -513,6 +514,7 @@ export class Runtime {
else if (signal.aborted) result = { status: 'interrupted', message: '用户停止了任务,这次调用没有执行。' };
else result = await this.executeTool(session, call, run);
session.providerMessages.push({ role: 'tool', tool_call_id: call.id, content: JSON.stringify(result) });
if (call.function.name === 'prepare_change_set' && result.status === 'awaiting_review') preparedSets.add(result.change_set_id);
if (call.function.name === 'call_registered_mcp' && result.requires_verification) mcpReview = result.mcp_job_id;
if (call.function.name === 'plan_delivery' && result.status === 'awaiting_review') {
deliveryReview = this.store.owned('delivery_plan', result.plan_id, session.id);
@@ -534,6 +536,22 @@ export class Runtime {
this.store.save(session, { type: 'delivery.awaiting_review', detail: { planId: deliveryReview.id } });
return;
}
// Settle the whole model turn first: later proposals can make a sealed
// set partial, and later tools may fail or need verification/stop.
for (const id of preparedSets) {
let ready = false;
try { ready = await this.changeSetManager().reviewReady(session, id, run.record); }
catch (error) {
this.store.event(session, { kind: 'system', source: 'workbench', label: '变更批次交接预检未通过', phase: 'failed',
output: this.provider.redact(error.message ?? String(error)) });
}
signal.throwIfAborted();
if (!ready) continue;
session.status = 'needs_review'; run.stopReason = 'change_set_review';
this.store.message(session, 'system', '当前任务的完整变更批次已封存,同轮工具结果已保存。请审查并应用或撤销批次;等待审查期间不继续请求模型。应用后可显式续接原任务进行检查。');
this.store.save(session, { type: 'change_set.awaiting_review', detail: { changeSetId: id } });
return;
}
if (this.store.core.task.get(run.taskId).readProgress?.pause) throw new BudgetExceeded('no_progress',
'模型连续重复读取工件或核对相同的工程状态,未推进当前任务。实际结果和原任务已保存,请核对后继续。');
}
+4
View File
@@ -0,0 +1,4 @@
export interface BudgetPauseSource { reason?: string; limitingReason?: string }
export function budgetPausePresentation(pause?: BudgetPauseSource, stopReason?: string): {
title: string; inputLabel: string; guidance: string;
} | null;
+18
View File
@@ -0,0 +1,18 @@
const reasons = {
context_limit: '上下文窗口不足', engine_input_budget: '原生引擎输入或剩余预算不足',
token_limit: '本次运行预算不足', task_token_limit: '任务累计预算不足', time_limit: '本次运行时间已用完',
};
export function budgetPausePresentation(pause, stopReason = '') {
const reason = pause?.reason ?? stopReason;
if (!reasons[reason]) return null;
if (reason === 'engine_input_budget' && pause?.limitingReason === 'remaining_token_budget') return {
title: '原生引擎剩余预算不足', inputLabel: '当前窗口可用输入',
guidance: '输入可容纳于当前上下文窗口;下一次请求的用量预留超过剩余预算。请核对运行和任务预算。',
};
if (reason === 'engine_input_budget' && pause?.limitingReason === 'context_limit') return {
title: '原生引擎上下文窗口不足', inputLabel: '当前窗口可用输入',
guidance: '必需输入超过当前上下文窗口可用容量。请核对模型容量和输入范围。',
};
return { title: reasons[reason], inputLabel: '当前可用输入', guidance: '' };
}
+9
View File
@@ -1,5 +1,14 @@
# Agent Workbench:核心代理能力研究与产品计划
## 当前优先 G:修完原件已确认问题
用户要求先把已总结问题修完;原生窗口回看不阻塞本次源码修复。沿用planning-with-files及product-ui-ux的真实流程验收,不启动子代理。
1. [complete] E原第7–12请求估算逐项离线重现,小源码保留6/5/5/4/2/2;无真实模型请求或原件改写。完整批次旧回归明确失败,已固定原日志。
2. [complete] 完整批次同轮结算后预检并交接审查,部分/失败/冲突/新提案/未知/篡改/停止维持各自语义。实际编译HTTP两请求停审查,显式续接同Task后两项实际Node检查通过,提案不重放。
3. [complete] 小源码先进入实际请求,按原来源稳定保留及实际空间分配下一页;第7–12请求均6/6源码、原硬预算保持。原生窗口和累计剩余额度文案已区分,历史未知分项不补造。
4. [in_progress] 最终受影响回归、1.24.1编译、正式版本两视口、独立源包重建及保留状态本地升级。开发旧dist/旧文案/版本入口与移动检视区导航失败保持;不把中断混合全轮记最终通过。封存新结果,本地Git提交。
## 当前接续 D:原版 Purge 生产策略
接续研究/封存 7b4f7ee / 1408e7a,A/B/C 不重做;独立端口和自有夹具,正式宿主与 Key 不操作。继续沿用 planning-with-files 与 product-ui-ux;本轮不启动子代理。
+20
View File
@@ -0,0 +1,20 @@
import test from 'node:test';
import assert from 'node:assert/strict';
import { budgetPausePresentation } from '../src/shared/budget-presentation.mjs';
test('input that fits the engine window but exceeds remaining credit diagnoses the credit limit', () => {
const pause = { reason: 'engine_input_budget', limitingReason: 'remaining_token_budget', requiredInputTokens: 27424,
inputBudget: 27552, remainingTokens: 28339, tokenReservation: { reservedTokens: 29600 } };
const diagnosis = budgetPausePresentation(pause);
assert.equal(diagnosis.title, '原生引擎剩余预算不足');
assert.match(diagnosis.guidance, /输入可容纳/); assert.match(diagnosis.guidance, /预留超过剩余预算/);
assert.ok(pause.requiredInputTokens < pause.inputBudget); assert.ok(pause.tokenReservation.reservedTokens > pause.remainingTokens);
});
test('window overflow and historical pauses retain distinct and honest diagnoses', () => {
assert.equal(budgetPausePresentation({ reason: 'engine_input_budget', limitingReason: 'context_limit' }).title, '原生引擎上下文窗口不足');
assert.equal(budgetPausePresentation({ reason: 'engine_input_budget' }).title, '原生引擎输入或剩余预算不足');
assert.equal(budgetPausePresentation(undefined, 'task_token_limit').title, '任务累计预算不足');
assert.equal(budgetPausePresentation(undefined, 'token_limit').title, '本次运行预算不足');
assert.equal(budgetPausePresentation(undefined, 'change_set_review'), null);
});
+6 -5
View File
@@ -67,11 +67,12 @@ test('a compatible model can seal proposals through the real HTTP loop while app
assert.equal(session.execution.changeSets[0].files.length, 2); assert.equal(fs.existsSync(path.join(f.projectRoot, 'src')), false);
assert.equal(fs.existsSync(path.join(f.projectRoot, 'docs')), false);
assert.equal(session.events.find(event => event.tool === 'prepare_change_set').phase, 'completed');
assert.equal(count, 2); assert.equal(session.execution.run.stopReason, 'change_set_review');
const tools = provider.requests.find(request => request.input?.tools)?.input.tools.map(tool => tool.function.name);
assert.ok(tools.includes('prepare_change_set')); assert.ok(!tools.includes('apply_change_set')); assert.ok(!tools.includes('rollback_change_set'));
});
test('a step-limited sealed batch applies and continues the same Task into actual checks without replaying saved tools', { timeout: 60000 }, async t => {
test('a sealed batch at the step limit hands off and continues the same Task into actual checks without replaying saved tools', { timeout: 60000 }, async t => {
const f = fixture(t);
fs.mkdirSync(path.join(f.projectRoot, 'test'));
const acceptance = "import test from 'node:test'; import assert from 'node:assert/strict'; import { left } from '../src/left.mjs'; import { right } from '../src/right.mjs';\ntest('left value', () => assert.equal(left, 1)); test('right value', () => assert.equal(right, 2));\n";
@@ -95,13 +96,13 @@ test('a step-limited sealed batch applies and continues the same Task into actua
await host.api('/config', { mode: 'compatible', baseUrl: model.baseUrl, model: 'batch-continuation-fixture', maxSteps: 2 });
await host.api(route + '/messages', { content: '完成原始双模块工程目标并运行测试' });
let paused;
await until(async () => { paused = await host.api(route); return paused.status === 'paused'; }, 30000);
assert.equal(paused.execution.run.stopReason, 'step_limit'); assert.equal(count, 2);
await until(async () => { paused = await host.api(route); return paused.status === 'needs_review'; }, 30000);
assert.equal(paused.execution.run.stopReason, 'change_set_review'); assert.equal(count, 2);
const set = paused.execution.changeSets.at(-1), firstRun = structuredClone(paused.execution.run);
assert.equal(set.status, 'prepared'); assert.equal(fs.existsSync(path.join(f.projectRoot, 'src')), false);
const savedMessages = (await host.api(route + '/export')).session.providerMessages;
const savedResults = savedMessages.filter(message => message.role === 'tool'); assert.equal(savedResults.length, 3);
await assert.rejects(host.api(route + '/resume', {}), /变更集/);
await assert.rejects(host.api(route + '/resume', {}), { status: 409, message: /请先审查待处理改动/ });
await assert.rejects(host.api(route + '/continue-reviewed', {}), { status: 409 });
await host.api(route + '/change-sets/' + set.id + '/apply', { fingerprint: set.fingerprint });
await until(async () => (await host.api(route)).execution.changeSets.at(-1).status === 'completed', 30000);
@@ -116,7 +117,7 @@ test('a step-limited sealed batch applies and continues the same Task into actua
assert.equal(completed.execution.task.goal, '完成原始双模块工程目标并运行测试');
assert.equal(completed.execution.run.resumeOf, firstRun.id); assert.equal(completed.execution.task.verification.status, 'passed');
assert.equal(completed.execution.checks.length, 1); assert.equal(completed.execution.checks[0].results[0].tests.passed, 2);
assert.equal(completed.execution.runs.find(run => run.id === firstRun.id).status, 'paused');
assert.equal(completed.execution.runs.find(run => run.id === firstRun.id).status, 'needs_review');
assert.deepEqual(completed.execution.runs.find(run => run.id === firstRun.id).usage, firstRun.usage);
assert.equal(completed.execution.changeSets[0].runId, firstRun.id); assert.equal(count, 5);
const exported = (await host.api(route + '/export')).session.providerMessages;
+71
View File
@@ -0,0 +1,71 @@
import test from 'node:test';
import assert from 'node:assert/strict';
import fs from 'node:fs';
import path from 'node:path';
import { Runtime } from '../src/server/runtime.mjs';
import { Provider } from '../src/server/provider.mjs';
import { fixture, call, assertToolPairs } from './helpers.mjs';
async function exercise(t, mode) {
const f = fixture(t), provider = new Provider(), runtime = new Runtime(f.store, provider);
f.onCleanup(() => runtime.stopAll());
f.store.saveConfig({ maxSteps: 2 });
let requests = 0, stopped, stopScheduled = false;
provider.complete = async () => {
requests++;
const calls = requests === 1 ? [
call('left', 'propose_file', { path: 'left.mjs', content: 'export const left = 1;\n', reason: 'Left module' }),
call('right', 'propose_file', { path: 'right.mjs', content: 'export const right = 2;\n', reason: 'Right module' }),
] : [call('seal', 'prepare_change_set', { change_ids: mode === 'failed' ? ['missing']
: f.session.changes.slice(0, mode === 'partial' ? 1 : 2).map(change => change.id) }),
...(mode === 'new-proposal' ? [call('third', 'propose_file', { path: 'third.mjs', content: 'export const third = 3;\n', reason: 'Later same-turn proposal' })]
: [call('after-seal', 'inspect_session', {})]),
...(mode === 'later-error' ? [call('bad-read', 'read_file', { path: 'missing.mjs' })] : []),
];
return { message: { role: 'assistant', content: 'Declared model fixture.', tool_calls: calls }, usage: null };
};
f.store.on('update', () => {
if (mode === 'cancelled' && !stopScheduled && f.session.events.some(event => event.toolCallId === 'seal' && event.phase === 'completed')) {
stopScheduled = true; setImmediate(() => { stopped = runtime.stop(f.session.id); });
}
if (mode === 'conflict' && f.session.events.some(event => event.toolCallId === 'after-seal' && event.phase === 'completed')) {
fs.writeFileSync(path.join(f.projectRoot, 'left.mjs'), 'Concurrent owner bytes\n');
}
if (mode === 'unknown') {
const event = f.session.events.find(event => event.toolCallId === 'after-seal' && event.phase === 'completed');
if (event) f.store.core.tool.get(event.invocationId).requiresVerification = true;
}
if (mode === 'tampered') {
const set = [...f.store.core.change_set.values()][0];
if (set) set.fingerprint = '0'.repeat(64);
}
});
runtime.start(f.session.id, 'Create and review two modules; do not apply them automatically.');
await runtime.active.get(f.session.id).promise;
if (mode === 'cancelled') { await new Promise(resolve => setImmediate(resolve)); await stopped; }
const run = f.store.core.run.get(f.session.runId);
assert.equal(requests, 2); assertToolPairs(f.session.providerMessages);
assert.equal([...f.store.core.tool.values()].filter(tool => tool.sessionId === f.session.id).length, mode === 'later-error' ? 5 : 4);
assert.ok(f.session.changes.every(change => change.status === 'pending'));
assert.equal(fs.existsSync(path.join(f.projectRoot, 'right.mjs')), false);
if (mode === 'conflict') assert.equal(fs.readFileSync(path.join(f.projectRoot, 'left.mjs'), 'utf8'), 'Concurrent owner bytes\n');
else assert.equal(fs.existsSync(path.join(f.projectRoot, 'left.mjs')), false);
return { ...f, run };
}
test('complete sealed batch hands off after all tools settle with no next model request', async t => {
for (const mode of ['complete', 'later-error']) await t.test(mode, async t => {
const f = await exercise(t, mode);
assert.equal(f.session.status, 'needs_review'); assert.equal(f.run.stopReason, 'change_set_review');
assert.equal(f.session.events.find(event => event.toolCallId === 'after-seal').phase, 'completed');
if (mode === 'later-error') assert.equal(f.session.events.find(event => event.toolCallId === 'bad-read').phase, 'failed');
});
});
test('partial, changed, failed or cancelled batches cannot claim complete review handoff', async t => {
for (const mode of ['partial', 'new-proposal', 'failed', 'conflict', 'cancelled', 'unknown', 'tampered']) await t.test(mode, async t => {
const f = await exercise(t, mode);
assert.equal(f.session.status, 'paused'); assert.equal(f.run.stopReason, mode === 'cancelled' ? 'cancelled' : 'step_limit');
if (mode !== 'failed' && mode !== 'cancelled') assert.equal([...f.store.core.change_set.values()][0].status, 'prepared');
});
});
+1 -1
View File
@@ -95,7 +95,7 @@ test('native input exceeding the context budget is retained and sends no model r
const f = fixture(t), context = controlledWorker(f, { sessionFactory: () => new Oversized() }), group = await prepare(f, context);
await (await context.manager.start(f.session, group.id, group.fingerprint)).promise;
assert.equal(context.requests.length, 0); assert.equal(jobs(f, group)[0].requests.length, 0);
assert.equal(f.store.core.task.get(group.children[0].taskId).usage.requests, 0); assert.match(jobs(f, group)[0].error, /预算/);
assert.equal(f.store.core.task.get(group.children[0].taskId).usage.requests, 0); assert.match(jobs(f, group)[0].error, /上下文窗口可用容量/);
const job=jobs(f,group)[0],rejection=job.inputRejections[0],artifact=f.store.owned('artifact',rejection.artifactId,job.sessionId);
const receipt=JSON.parse(f.store.artifactStore.read(artifact));
assert.equal(artifact.kind,'engine-model-rejection');assert.equal(artifact.sha256,rejection.sha256);assert.equal(receipt.dispatched,false);
+28
View File
@@ -228,6 +228,34 @@ test('a later narrow page of the same file revision cannot evict the wider actua
verifyPlanEvidence(f.store, f.session, plan.record);
});
test('six small source files stay complete across metadata turns even when mandatory evidence exceeds the soft target', t => {
const f = setup(t, { contextLimit: 32000, maxOutputTokens: 2048 });
const files = [['src/App.tsx', 1991], ['test/domain.test.mjs', 1715], ['src/transition.ts', 167],
['src/query.ts', 142], ['src/types.ts', 82], ['src/summary.ts', 155]].map(([name, size]) => read(name,
('// ' + name + '\n').padEnd(size, 'x')));
group(f, [...files, huge()]);
const system = 'Complete original Task constraints and required permission boundaries. '.repeat(620);
const original = structuredClone(f.session.providerMessages);
let exceededSoftTarget = false;
for (let index = 0; index < 4; index++) {
group(f, [{ tool: 'inspect_session', args: {}, result: { session: { status: 'running' },
execution: { progress: { completedTools: index + 7 } }, metadata: 'Historical metadata '.repeat(1800) } }]);
const prepared = f.planner.prepare(f.session, f.run, f.config, { system, tools: toolDefinitions });
const set = setFrom(prepared.messages);
for (const file of files) {
const entry = set.entries.find(entry => entry.observation.path === file.args.path);
assert.equal(entry?.content, file.result.content); assert.equal(entry.truncated, false);
}
assert.equal(prepared.record.budget.targetInputTokens, 16000);
assert.ok(prepared.record.estimatedInputTokens <= prepared.record.budget.inputBudget);
assert.ok(prepared.record.workingRetentionTokens > 2048);
exceededSoftTarget ||= prepared.record.estimatedInputTokens > 16000;
assertToolPairs(prepared.messages); verifyPlanEvidence(f.store, f.session, prepared.record);
}
assert.ok(exceededSoftTarget, 'The regression must exercise the soft-target boundary.');
assert.deepEqual(f.session.providerMessages.slice(0, original.length), original);
});
test('historical read-artifact navigation points to the original artifact and next page, never its result wrapper', t => {
const f = setup(t), text = 'An actual saved failure log.\n';
const artifact = f.store.artifactStore.write(f.session, f.run, text, { kind: 'check-log' }); f.store.core.artifact.set(artifact.id, artifact);