diff --git a/AGENTS.md b/AGENTS.md index b36eb3b3..83aa947c 100644 --- a/AGENTS.md +++ b/AGENTS.md @@ -1,5 +1,7 @@ # Agent Workbench项目执行规范 +2026-11-06 官方优先全系统审查已形成当前接续:见 [全层审查](docs/SYSTEM-AUDIT-OFFICIAL-FIRST-2026-11-06.md)、[开发路线](docs/DEVELOPMENT-ROADMAP-OFFICIAL-FIRST.md)、[机器矩阵](docs/official-first-system-audit-2026-11-06.json)。基线4da2550b,715清单输入SHA一致,30能力区域/38事项(含缺陷、缺项、未部署、风险与未知),21编号观察及O18-C单独纠正,1280/390自有页面实看;不是逐行人工审计或全球真实兼容全过。正式只读health在13:08Z为1.25.12/PID71628/ready,仍沿F源码6867cd5a;目录刷新ae5123c5、wire诊断341f5f69、官方容量2bd7b24a三项已推送未部署。本轮只审查/规划/纠正文档,不读真实Key、不请求真实模型、不改正式数据、发行或重启。下一M1修原生嵌套参数/stream_options、精确官方推理模式、自动选模与残留MCP/检查点/手机/凭据登记上限及权限说明,再M2官方注册表/缓存状态、M3独立发行、M4新真实基线;原网络/unknown/重复读/Purge收益unknown保持。官方直连以官方为准,代理声明与实际结果分开,用户完整透传和持久权限不恢复本地额度;下方历史“当前”均按各自时点理解。 + Claude官方容量已纠正并修复源码:当前官方表四精确ID(Fable5.1/Opus5.5/Sonnet5.5/Haiku5.5)同步128000/上下文1000000已公开,来源与目标正文SHA见docs/claude-official-capacity-2026-11-06.json。2bd7b24a增加显式原生→相符目录→精确官方→未知连接声明,保持本地透传/用户配置/原凭据;模型中心和参数注明来源,Chat/Responses不添加可选上限。独立相关67/67、build-2与网页4组/1280及390实看通过;原失败与上轮未知调查状态保持,见docs/QA-CLAUDE-OFFICIAL-CAPACITY.md。尚未部署/发行/重启,正式仍沿下方F旧源码身份;不能把官方已知规格当真实代理兼容通过。源码/纠正/只读协调已普通增量推送至1b2ea607,远端逐SHA一致,回执docs/gitatlas-claude-capacity-push.json另随完成状态提交;随后独立发行含目录/诊断/官方容量三修复;不续跑原工程/旧实验、不委派/触真实Key,历史网络/unknown/Purge unknown保持。 New API中文学习已覆盖305正文/201嵌入接口,API与Plugins全量审读与来源SHA/Git blob核对通过;见docs/NEWAPI-DOCUMENTATION-LEARNING.md及逐篇review清单。两处独立源码修复:ae5123c5目录刷新只更新上游事实、保持用户字段/凭据绑定与原Run;341f5f69截断诊断按实际wire字段说明,Responses不同incomplete原因分开。源码build/独立候选相关126/126通过,原夹具失败保持,见docs/QA-NEWAPI-COMPATIBILITY-REVIEW.md。来源/审读/两修复/验收5提交已普通增量推送至0dd9c50d90a8666575f934b7f70d2ca097462cd0、远端逐SHA一致,回执docs/gitatlas-newapi-compatibility-push.json另随收尾提交推送。尚未发行/正式安装或重启,正式身份沿下方F记录;Claude旧2048仍是必填声明而非已知最大值。本轮0真实模型/不触真实Key或正式数据、不委派,默认开放策略不变;Gemini原生/媒体/服务端工具/插件宿主缺项及真实兼容、缓存/角色/重复读取普遍改善/Purge收益unknown均保留。后续按需独立发行再另冻真实样本,不续跑旧或用户工程。 diff --git a/docs/CAPABILITIES.md b/docs/CAPABILITIES.md index 25c6d72b..a09437a9 100644 --- a/docs/CAPABILITIES.md +++ b/docs/CAPABILITIES.md @@ -1,5 +1,7 @@ # Agent Workbench 能力目标与验收矩阵 +当前入口更新(2026-11-06):正式只读health观察为1.25.12/PID71628,F角色路由已交付;后续目录/wire诊断/官方容量三源码修复尚未部署。最新[全层审查](SYSTEM-AUDIT-OFFICIAL-FIRST-2026-11-06.md)按30区域/38事项分开已实现、已验、已部署、缺陷、未实现与未知,含21编号观察及O18-C。下一按[官方优先路线](DEVELOPMENT-ROADMAP-OFFICIAL-FIRST.md)M1正确性/残留上限、M2官方契约/缓存状态推进;真实工程原成功、网络失败、unknown与Purge收益unknown保持。下方D/E等“当前”属于历史时点,不再作为最新状态。 + 当前正式Orven1.25.11已交付E:可从原报告/检查/工具事实生成来源绑定草稿,保留失败与unknown;默认不使用,用户可修订/采用/撤销/拒绝/记录反例/归档。完整修订和已发送快照保持,来源/源码变化停止采用,可选经验故障不阻断原工程,原手写记忆/开放权限/完整透传保持。独立源码/程序/安装/跨版与正式网页/桌面见[E验收](QA-EXPERIENCE-DRAFTS-E.md)。经验结论和真实收益未证明,F角色路由/Mnemo原生导出/入站委托仍未实现,历史网络/unknown/Purge收益unknown保留。下方为历史时点。 当前已正式交付Orven1.25.10:A要求证据/B单轮观察/C工具与前缀、Orven界面及AH/AI工程修复之后,新增默认关闭的进展提醒和可核对的实际回读。提醒保持模型额度/历史/权限,后续工具进展不冒称业务完成或因果收益。源码/随包/跨版安装与正式两端结果见[D验收](QA-REMINDER-READBACK-D.md)。E经验草稿/F路由及跨产品入站服务仍未实现;真实AD/AE结果、原失败和Purge收益unknown按各自证据保留。下方各“最新”为历史时点。 diff --git a/docs/DEVELOPMENT-ROADMAP-OFFICIAL-FIRST.md b/docs/DEVELOPMENT-ROADMAP-OFFICIAL-FIRST.md new file mode 100644 index 00000000..6f204d50 --- /dev/null +++ b/docs/DEVELOPMENT-ROADMAP-OFFICIAL-FIRST.md @@ -0,0 +1,145 @@ +# 官方优先开发路线(2026-11-06) + +本路线以[全系统审查](SYSTEM-AUDIT-OFFICIAL-FIRST-2026-11-06.md)的38项事项为输入。目标是官方直连优先、代理/兼容同样保真,用户选择的模型和权限持久生效,AI能完成实际工程并提供真实证据。当前审查只完成梳理与文档纠正,没有开始下面的产品改动或正式发行。 + +**立即下一批是 M1:修参数丢失、官方推理选项、自动选模错误,以及开放模式仍生效的 MCP/检查点/手机固定上限,统一真实权限说明。** 完成核心修复后进入M2供应商契约与缓存/状态,冻结一次M3发行;不要为每个小修反复打包,也不要先堆新界面或新“支持”标签。 + +## 执行原则与完成标准 + +- 官方协议/schema/SDK与精确型号事实为准;代理声明和实际观察独立保存。未取得官方来源的能力显示未知,允许自定义兼容与显式覆盖,不猜型号。 +- 保留用户完整内容、原生参数、上游原响应和真实用量。默认unrestricted/passthrough/preserve;初始批准模式ask,用户改auto/full后持久沿用。自动适配是选择合适模型的能力,不能重新成为全局额度/权限拦截。 +- 官方必填字段由用户显式值或可靠官方/上游值提供。Chat/Responses可选输出字段继续默认不附加;Messages必填max_tokens不能省略或发送Infinity;模型最大容量和本地不限分开展示。 +- 每批交付必须包含:最终来源/必要官方资料、实际改动、相关确定性核对、可运行核心路径、原失败/unknown、回顾和限定本地提交。测试只覆盖实质行为和边界,已冻结的无关大套结果不反复重跑。 +- 源码包只发布源码及构建所需输入;程序包另有固定来源和编译身份。数据保持核对属于安装/正式升级阶段,不让用户项目混入源码打包。 +- 真实模型实验另冻新ID、输入、宿主/模型/连接身份与开关。历史W/AD/AE/AG、X及其他旧实验不续跑、不改原失败,不把合成夹具和官方文档当真实交付。 + +## 分批顺序 + +| 阶段 | 交付 | 结束依据 | 依赖 | +| --- | --- | --- | --- | +| M0 本次审查 | 分层矩阵、来源账本、历史异常、开发与验收路线;I30当前指针纠正 | 原源码715输入SHA、21观察+O18-C、隔离1280/390、文档/JSON/证据封存核对 | 已完成审查;不发行 | +| M1 正确性与开放执行 | I01–I08、I33/I38修复;实际wire/权限/模式一致 | 非目标参数不丢、官方模式可用、跨旧次数/大小/登记边界执行并可恢复;网页/桌面核心路径 | M0 | +| M2 官方契约与缓存连续性 | I09–I15、I27/I28;版本化官方能力/请求合同、完整统计/账本、CI | 来源明确、模式/版本/端点可核对、会话连续、真实缓存字段保留、完整分页;相关确定性核对 | M1;CI可先完成基础 | +| M3 独立发行 | 原三项修复+M1/M2最终冻结来源,单次源码/程序发行与正式安装 | 编译/包逐输入身份、实际跨版/安装、两端核心入口、正式after核对 | M1/M2范围冻结 | +| M4 新真实兼容与交付基线 | 官方OpenAI/Claude/DeepSeek优先、当前代理另列;真实Node/React/Python | 发现/参数/流式/工具/原错/usage逐项,工程原验收与Git真实完成;失败仍可封存 | M3;启用线路身份固定 | +| M5 全球原生适配与模型能力 | Gemini/云平台/其余官方厂商、服务端工具/多模态/候选/资源 | 每协议/供应商一份官方契约、实际wire/UI/结果工件/生命周期证据,再另冻真实样本 | M2;分小批发布,不等全球一次做完 | +| M6 工程宿主与协作 | 交互PTY、电脑操作、Worker技能/MCP与逐要求质量 | 真实输入/停止/恢复/跨应用动作/子任务工具/检查证据闭环;full无额外审批 | M1/M2;可不等待全部M5厂商 | +| M7 规模与生态 | 测量后增量存储、跨产品原生契约服务、真实手机 | 长历史原件/时延/RSS、委托撤销/归属、物理手机断线与恢复 | 稳定来源和对应集成契约 | +| M8 收益验证 | 缓存/重复读/经验/角色分别对照,再新原版Purge AB/BA | 固定可比样本、工程质量与reported/unknown成本、边界触发/加载证据,原失败保留 | 目标线路M4成功基线及实际机制;无需等全球厂商全部接完 | + +### M1:先把已有路径做正确 + +原生对象编辑只改用户选中的键,保留reasoning.summary、output_config.format、thinking.display及其他未知但合法字段。Chat流式不能整对象替换stream_options。默认重置操作明确作用范围,保存/重新打开/实际请求保持一致。 + +Claude按精确ID、API协议和官方条件显示模式/effort。明确不支持与没有目录字段分开;全否定模式列表不能变成“全部支持”。对代理保留明确覆盖入口与实际原错。此处先修已确认的当前型号行为,M2再建立可更新完整注册表。 + +自动模型池根据任务和已知能力选候选,记录原因;手选模型或显式强制尝试继续透传。不要把自动选择修复做成所有请求都先行拒发。 + +MCP、检查点和手机账本改为完整分页/分段记录,跨31次操作、31个检查点、2001工具引用、64MiB结果、2001手机操作等旧边界仍可工作;校验原SHA、幂等、停止和重启unknown。MCP合法JSON Schema支持与服务器协议相符,不能用65属性之类固定额度代替真实适配。 + +本机加密凭据登记超过101项的整体拒加载门槛也要去掉,以版本和逐条有效性恢复。验收用自有合成/测试凭据跨102项保存重启,保留原绑定和原metadata,不读取用户真实Key。 + +真实权限/身份策略同时进入UI和模型工具说明。ask/auto/full的“请求策略、实际生效策略、动作结果”一致,不能用MCP提示“先确认”诱发不必要的停顿。MCP注册45000ms只是旧配置范围,开放执行本来已0超时,不重复修成另一层限制。 + +### M2:建立官方与兼容共同可追溯的模型层 + +以 `provider + protocol + API version + exact model ID` 建能力记录;至少包含来源URL/SDK commit或schema SHA、观测时间、原始目录字段、上下文/输出规格、推理模式与条件、工具/媒体/流式/缓存/结构化能力、弃用状态,以及本地实现/实际验证状态。 + +自动获取和分页选择模型,不要求普通用户手填所有型号。供应商目录有完整信息时保留原字段;目录没有容量/功能时由有版本的官方表补充。显式参数优先;上游声明与官方规格冲突时显示双方来源和实际线路范围,不静默覆盖用户原设置/凭据绑定。 + +请求合同分别管理认证、端点、API version/beta/供应商头与原生参数;连接定义、模型选择和账户凭据绑定分开。固定版本不是靠修改一处通用字符串就适配全球;旧API版本的有效性由证据决定。 + +缓存工作包含Messages块/工具cache_control保真、Kimi等读/写/未缓存归一、会话/Task亲和、原生状态跨轮续接和缓存相关参数指纹。旧ordered-prefix观察继续保留并注明只观察提示字段;不能据same_prompt声称缓存不受影响。改变型号/连接/原生状态时记录原因;DeepSeek带tools的所有历史reasoning_content需按官方契约完整处理。 + +请求账本改完整游标/导出和逐请求写入,保存量与文案一致。先搭GitAtlas可重复CI与锁定环境的协议/构建核对,真实服务实验保持独立来源和账本;Python/浏览器缺失等首环境失败不能被改绿。 + +### M3:一次冻结、一次发行 + +冻结范围后确定下一正式版本和源码commit,带上ae5123c5、341f5f69、2bd7b24a及本阶段修复。源码包、程序包、依赖锁、编译输出、安装器都有独立清单;原用户项目不参与源码打包。 + +实际隔离安装、跨正式旧版升级、关键协议wire和网页/桌面核心入口通过后才正式激活。正式安装核对原数据/配置/工件/凭据存在状态,不导出或读取真实Key;升级/回退原错误保留。此路线本身不启动安装/重启。 + +### M4:先取得新的官方真实基线 + +届时使用用户当前启用、明确选择的连接,另冻新样本;不沿用X/AD/AE/AG或旧工程。官方直连和代理样本分开;目录、身份声明、推理/输出参数、流式/工具、用量和错误都分别判定,不能用一个“连接成功”覆盖所有功能。 + +Node/React/Python从自有冻结工程开始,保留原测试/验收、真实源码差异、检查、Git范围与交付报告;不给实现答案,也不为过关改变原断言或默认开放设置。模型只读/未交付、网络断连、unknown、协议缺项均是可封存结果,不自动重放失败请求。 + +R/W预留v2边界未触达继续作为历史缺口保留。默认透传时不为了验证旧限制重新打开本地预算,不把这项老缺口伪装成已经真实验证通过。 + +### M5:逐官方协议扩大覆盖 + +先补Gemini原生生成/流式/发现/分页/内容/状态,再按官方云平台实现Bedrock、Vertex/Foundry等认证、区域/部署与结果;OpenAI兼容方式和原生方式分别登记。Qwen/Z.ai/Kimi等已经取得的官方定义用于各自契约;Mistral/xAI等来源未取得时先明确来源缺口,不能泛称全部支持。 + +分别实现供应商服务端工具、停止/暂停/继续、结构化结果验证、用户tool_choice、多候选选择,以及PDF/音视频/原生Files。随后补Batches/后台会话/资源轮询与取消、图像/音频/向量等需要的完整生命周期。字段能透传与功能闭环分开验收;不需要复制New API的账户、渠道计费和运营后台来完成开发工作台。 + +每种原生返回类型保留raw及使用来源,未知类型明确标为本地适配未实现,不能冒称上游模型服务报错。工具结果/模型声明/实际电脑动作与工程交付不能互相替代。 + +### M6:把宿主做成实用工程代理 + +建立交互PTY工具和持续终端会话:stdin、resize、长命令、停止、重新附着、退出码与现有Git证据连接。增加截图、键盘/鼠标/窗口/应用操作以实际控制电脑;仅有DOM点击和文件对话框不能算通用computer use。 + +Worker按角色和任务装配skills/MCP,保留真实工具表、资源身份与未装配原因。逐要求关联代码/检查/观察,模型语义评价与确定性验收分别报告;not_evaluated不能冒充已满足。角色/多代理模式的收益另做同题实验,不默认添加审查者或每步批准。 + +### M7:测量规模并完成生态闭环 + +先用自有数据测全量加载、会话/请求写入与长响应RSS和时延曲线,再增量事件/分页/流式工件。规模问题不能用截断历史、删除原错或新固定额度解决。 + +Mnemo/其他产品从双方实际原件契约实现导出、入站委托、归属/撤销/失败;多Run Task和单Run包不能混同。手机以当前物理设备跑完整配对/选择/批准/停止/断线/撤销,不能以本机模拟替代。 + +### M8:真实收益单独证明 + +缓存指标取上游实际read/write/uncached;区分新增文件读取、内容变化后的必要回读与同SHA/同范围重复读取。提醒、经验、角色分别固定来源和开关,比较工程成功率/质量、tokens、时延及原失败,不把工具进展观察当业务交付。 + +原版Purge另冻AB-off/AB-on/BA-on/BA-off;固定插件源码/宿主/模型/连接身份、同题输入、权限、顺序及预先声明判定方法。开启组必须有真实加载和相关边界证据;完整成功对照不足则收益继续unknown。AG三个ETIMEDOUT、三个unknown及唯一ba-off成功原样保留,事后TLS通过不能倒改原失败。只在目标线路基线具备可比性后执行,不需要等待全部厂商或其他生态扩展做完。 + +## 38项逐项验收入口 + +以下全部事项都有明确阶段;“未测收益”也有验证交付,不冒充产品源码修复。具体源码位置、官方来源和原观察在[审查矩阵](official-first-system-audit-2026-11-06.json)。 + +| ID / 阶段 | 下一项改动或验证 | 完成证据 | +| --- | --- | --- | +| I01 / M1 | 只修改目标键;删除目标值也不删除同对象其他字段;明确默认重置的范围。 | 三协议往返、JSON/快捷控件/保存/重新打开,非目标字段逐项保持。 | +| I02 / M1 | 透传显式原生设置;默认建议与用户值分开,不隐式覆盖。 | 实际本机 HTTP wire 与用户对象一致,缺失字段才按明确策略补充。 | +| I03 / M1 | 以精确型号、协议、官方版本解析模式/effort/条件,区分未知与明确不支持。 | 官方新旧型号矩阵、目录全否定和缺失情况通过;UI/保存/wire 一致;兼容显式覆盖仍可用。 | +| I04 / M1 | 自动候选按能力与任务排序,显式手选/强制尝试与自动适配分别记录,不恢复全局拒发。 | 自动选择工具支持成员;未知有明确策略;用户手选仍可透传;记录选择依据。 | +| I05 / M1 | 移除开放模式次数门槛,完整账本用游标/分页;保留每次真实身份、结果和停止。 | 同任务第31次及更长序列可执行且无旧记录丢失;停止/重启 unknown 保留。 | +| I06 / M1 | 支持官方/服务器声明的 JSON Schema 或按协议原样转交;不用固定属性数假冒模型能力。 | union/ref/嵌套/65以上属性工具真实握手后可选择;错误与原 schema 可核对。 | +| I07 / M1 | 采用分段索引、流式校验与分页恢复,去掉开放模式固定额度。 | 跨各旧边界建立和恢复检查点,旧工具/结果原 SHA 与归属完整。 | +| I08 / M1 | 展示声明能力、本地实现、实际权限/身份策略与可选覆盖;同步模型提示。 | ask/auto/full 与完整参数/身份原件在网页和桌面一致;不恢复强制拦截。 | +| I09 / M2 | 增加原生块/工具元数据保留及供应商缓存控制,按官方层级/TTL处理。 | 工具/系统/用户块 cache_control 经过请求与历史往返不变,缓存读写由上游 usage 证明。 | +| I10 / M2 | 提供带供应商语义/来源的读、写、未缓存、推理 token 统计,不补造费用。 | 三个互斥输入类别与总量相符;流式最终用量/unknown/raw保持。 | +| I11 / M2 | 会话/Task级可配置亲和和供应商状态绑定,用户换模显式记录影响。 | 同会话续接、停止恢复、失效换线、手选换模分别核对模型/历史/缓存与状态。 | +| I12 / M2 | 保留原有序提示观察,另记录缓存相关原生参数/型号/版本/工具/TTL指纹。 | 参数变化单列原因;不把字节共前缀或相同指纹当缓存命中。 | +| I13 / M2 | 完整游标检索/导出和准确保存说明;逐请求存储替代整数组回写。 | 所有历史尝试可分页/导出且归属、usage、unknown、raw SHA不变。 | +| I14 / M2 | 供应商、传输协议、官方API版本和代理覆盖分别配置并记录实际wire。 | 官方示例契约与版本/beta场景可复现;自定义端点原值与原错误保持。 | +| I15 / M2 | 自动发现+官方版本化能力表+上游声明+显式覆盖,逐字段记录来源和冲突。 | 多页/多型号/失效/无元数据/代理差异不覆盖用户字段;不从型号子串猜能力。 | +| I16 / M5 | 另建官方Gemini适配器,取得可读官方schema后固定契约来源。 | 发现、分页、认证、流式、工具、多模态、状态、usage、原错误均有隔离与真实证据。 | +| I17 / M5 | 按官方云平台分别实现身份提供器/区域/部署/签名/续期/错误,不复制存储真实Key到研究材料。 | 每平台固定官方契约与独立真实样本;代理转换路线单独标识。 | +| I18 / M5 | 原生服务端工具、结果块、签名、停止原因和继续请求独立适配,保留原响应。 | 官方块/SSE事件/暂停续接/usage/错误完整,不把未知块或工具结果当本地执行成功。 | +| I19 / M5 | 区分工作台工具与供应商原生工具,提供明确选择和实际调用记录。 | 用户选择在wire准确出现;本地/远端结果归属分开;禁止盲目并行应用多个候选。 | +| I20 / M5 | 按供应商内容格式实现附件/原文件/媒体/引用/返回工件,不把PNG通路当所有视觉支持。 | 原格式/用途/上传删除/尺寸/流式输出/错误来源与真实模型验证逐项完整。 | +| I21 / M5 | 定义完整保留与候选选择工作流;工程执行一次只用明确选中的完整候选。 | n>1两模式原结果保持、用户/自动选择一致;不盲目执行各候选的工具。 | +| I22 / M6 | 增加会话PTY、stdin、resize、进程恢复与工具接口,沿用户权限直接执行。 | 长命令、交互输入、停止、重新附着、退出码与Git证据真实可查。 | +| I23 / M6 | 支持截图、坐标/语义动作、键盘/窗口/应用操作,与选定权限和任务归属连接。 | 真实多应用闭环及取消/恢复可观察;不能用本地浏览器DOM测试冒充OS控制。 | +| I24 / M6 | 逐要求关联实际差异/测试/观察,语义评估与确定性结果分列。 | 未测/失败/证据不足不通过;真实同题角色对照才证明收益。 | +| I25 / M6 | 角色/子任务按需求装配继承资源、身份与作用范围,不默认要求额外审查轮。 | 子任务实际工具表/来源/SHA与请求相符,未装配原因显示,full流程无额外确认。 | +| I26 / M7 | 先测量,再增量事件/分页/流式工件,避免用截断历史或固定额度解决规模问题。 | 可复现增长曲线、RSS/时延/恢复和原字节核对;报告机器与数据规模。 | +| I27 / M2 | 在GitAtlas支持环境建立可重复的源码/契约/构建/安装验证任务;真实供应商另受控运行。 | 干净环境按锁定依赖完整执行并保留首失败/版本/source SHA;不将需浏览器/Python环境失败改绿。 | +| I28 / M2 | 官方示例/schema驱动契约夹具,版本差异/弃用更新进入能力注册表与回归。 | 固定来源/原字节/协议/schema/wire/真实批次相互关联;本机夹具不当真实服务验收。 | +| I29 / M3 | 完成首批确认修复后冻结独立发行,避免逐小修反复打包;安装再做数据保持核对。 | 候选源码/build/wire/UI/安装身份一一对应,再部署与正式回看;源码包不包含用户项目。 | +| I30 / M0 | 本轮统一当前指针,保留日期明确的历史,不把规划等于完成。 | AGENTS/CAPABILITIES/NEXT-STEPS/PRODUCT-PLAN/RESEARCH-ROADMAP当前入口一致可点击。 | +| I31 / M7 | 按实际双方契约实现原字节导出、单Run/多Run归属与委托状态,不自动继承用户授权。 | 双方合成原件和独立真实服务验证;失败、容量缺项、撤销和旧原件完整。 | +| I32 / M7 | 独立配对/查看/选择权限/批准/停止/断线/撤销真实设备样本。 | 保存设备/宿主/浏览器版本和实际动作,无任务或凭据跨错归属。 | +| I33 / M1 | 分页持久化幂等操作,移除固定总数拒绝和load格式上界,不丢旧记录。 | 跨2000边界继续操作,幂等/unknown/断线恢复与原行保持。 | +| I34 / M4 | 按正式新版本另冻官方优先的供应商样本与代理样本;旧样本不重写/续跑。 | 每连接发现/参数/流式/工具/媒体/usage/错误/真实工程独立分项,未知保留。 | +| I35 / M8 | 分别定义同题基线/干预、稳定线路/模型/权限、真实raw用量与质量指标。 | 成功率、重复读取、cache read/write、tokens/时延和失败原因同时报告;不得以提醒触发当业务完成。 | +| I36 / M8 | 供应商兼容与基线真实交付稳定后,新冻结原版Purge AB/BA,身份和输入严格固定。 | 插件确实加载/边界触发/对照可比/工程质量及reported和unknown单列;旧AG失败原样保留。 | +| I37 / M5 | 按工程优先级通过供应商适配器逐类扩展,显示实际已实现能力;无需复制网关账户/计费后台。 | 资源创建/使用/检索/取消/清理、返回状态/工件与费用来源可核对,异步响应不误当工具完成。 | +| I38 / M1 | 按版本/逐记录有效性和分页加载凭据登记,移除固定连接总数格式门槛。 | 102以上自有合成/加密测试登记保存重启恢复;无旧条目损失、无真实Key访问,来源绑定仍可核对。 | + +## 当前接续 + +当前只完成M0审查、文档和证据封存。本地提交/推送不代表M1源码修复、M3正式发行或M4/M8真实实验已经执行;以后接续须继续报告各自来源与状态。 + +下一轮从M1的参数保真和官方推理模式开始,完成相关实际wire/表单核对及回顾后本地提交,再推进固定上限和说明一致性。后续授权范围内普通增量推送;每个真实实验另有冻结计划,历史异常始终保留。 diff --git a/docs/NEXT-STEPS.md b/docs/NEXT-STEPS.md index c9c6fddd..29d07e10 100644 --- a/docs/NEXT-STEPS.md +++ b/docs/NEXT-STEPS.md @@ -1,5 +1,7 @@ # 后续执行计划 +当前入口更新(2026-11-06):[全系统审查](SYSTEM-AUDIT-OFFICIAL-FIRST-2026-11-06.md)已梳理30区域/38事项,F已正式1.25.12,三后续修复仅源码态。立即下一批是[官方优先路线](DEVELOPMENT-ROADMAP-OFFICIAL-FIRST.md)M1:参数保真、官方推理模式、自动选模、MCP/检查点/手机/凭据登记残留上限和说明一致性;然后M2契约/缓存与M3独立发行,再另冻真实供应商/工程/Purge样本。此次审查不发行或重启,不续旧实验,不改历史失败/unknown。下方旧接续按历史时点保留。 + 当前1.25.10的D已正式交付并保留原PID/Provider和所有原数据,见[D验收](QA-REMINDER-READBACK-D.md)。下一先做[E经验草稿](EXPERIENCE-DRAFTS-E-PLAN.md)的来源与采用分离,不自动注入未核对经验;其后才按前置推进F。Mnemo新证据包仅只读反馈,自动导出/入站委托另切;不续跑历史真实样本,网络异常、unknown用量和Purge收益unknown保持。下方旧“当前”按历史时点理解。 ## 当前:V已封存,Tianshu-harness源码研究 diff --git a/docs/OFFICIAL-FIRST-SYSTEM-AUDIT-PLAN.md b/docs/OFFICIAL-FIRST-SYSTEM-AUDIT-PLAN.md new file mode 100644 index 00000000..4098b08a --- /dev/null +++ b/docs/OFFICIAL-FIRST-SYSTEM-AUDIT-PLAN.md @@ -0,0 +1,29 @@ +# 官方优先的全系统审查计划(2026-11-06) + +用户要求:面向全球 AI 大模型,官方直连和官方协议为主,兼容接口及代理同时支持;把系统现状、问题、未开发能力和后续顺序梳理清楚。 + +审查基线:`4da2550becdf4750320cedd5ee30a7e38baba2f9`,`feat/task-agreement-1.25`,起点工作树干净且与上游跟踪分支一致。正式版本暂沿 F 的 1.25.12 记录,须重新只读确认;浏览器旧查询串不是版本证据。 + +## 阶段与交付 + +1. [complete] 恢复上下文与建立覆盖清单:AGENTS、TIANSHU、1.25 计划、实际源码/测试/入口/既有 QA;保存范围和证据等级。源码 272 文件 / 31738 行,属于清单及关键路径审查,不等于逐行人工审计。 +2. [complete] 官方协议与能力审查:取得 Claude 既有原件、OpenAI 固定官方 SDK 类型以及 DeepSeek / Qwen / Z.ai / Kimi / Bedrock / Azure 公共正文;读取范围和未取得来源分别记录,不声称全球官方文档全部读完。 +3. [complete] Harness 与产品全层审查:任务/权限、读写/Git/终端/浏览器、模型路由/协作、记忆/提醒/Purge、恢复/存储/发行、界面与外部集成。静态规模风险没有性能实测。 +4. [complete] 关键疑点隔离核对和矩阵完成:21 个编号观察及 O18 单独纠正,真实自有网页 1280 / 390 核对;30 个能力区域、38 个事项和官方来源账本已形成。均未发真实模型请求,不能作为实际供应商验收。 +5. [in_progress] M0–M8开发顺序及38项验收已形成,当前入口已统一;继续封存证据、检查文档/JSON/链接/diff,定向本地提交及已授权增量推送。 + +正式服务只读,不请求真实模型、不读取真实 Key、不续跑旧实验或用户工程。审查不自动发行或重启。默认开放/透传与用户已选权限保持;既有失败、网络异常、未知用量、重复读取及 Purge 收益未知不改写。 + +复用 New API 的 305 篇正文 / 201 份接口定义审读及 Claude 官方容量原件,不重复整个抓取/验收。其他官方来源记录请求 URL、最终 URL、状态、内容类型、读取范围和 SHA;403、地区不可用 HTML 和未取得正文只记来源缺口。 + +## 调查错误与处理 + +- 接续前尝试的 OpenAI skill 普通路径不存在;实际路径为 `C:/Users/wangy/.codex/skills/.system/openai-docs/SKILL.md`,已读正确文件。 +- OpenAI Responses / Models / Prompt caching 的三个 developers.openai.com 页面首次公共读取均 HTTP 403;未把失败当正文或用记忆填补,继续核对官方可取得入口。 +- 大历史文档合并输出发生截断,调整为小范围分段读取和清单;不据截断输出宣称逐项审完。 +- 公共页面调查中 Google / Mistral / xAI 初次读取 WinError 10060(仅公共文档,非模型请求);改用有总截止时间的现有 Node 网络层核对,不循环重发模型。 +- 公共抓取脚本第一次相对 import 多上一层而未启动;修正为实际仓库路径,首错误不记为任何供应商失败。 +- 接续中仍有合并历史输出截断和两个猜测源码路径不存在,已按 rg 实际路径缩小读取;不计入产品失败或覆盖清单完成率。 +- O18 首次源码推断把 Chat 流式误说为选第一个候选。原脚本/结果保留;单独 O18-C 用内存 SSE 确认:非流式保留多候选但 Harness 只消费 choices[0],流式遇到非 0 候选会本地拒绝。没有真实模型调用。 +- MCP 45000ms 是注册配置字段上界;unrestricted 执行超时已为 0。不能将它与仍生效的工具次数上限混为一项运行限制。 +- 数据组装首轮图像源码锚点错误(protocols没有image分支)被严格检查拒绝;原assemble-round-1.log保留,改以真实input-attachments入口,第二轮与新增O21后的第三轮通过。是审查组装失败,不是产品或模型测试失败。 diff --git a/docs/PLAN-1.25.md b/docs/PLAN-1.25.md index 05016130..11d6c6de 100644 --- a/docs/PLAN-1.25.md +++ b/docs/PLAN-1.25.md @@ -1,5 +1,7 @@ # 1.25 计划:任务要求与交付复核 +当前入口更新(2026-11-06):要求证据、单轮观察及后续Tianshu A–F机制已有各自实现/发行;正式为F/1.25.12,后续目录刷新/wire诊断/官方容量修复已推送未部署。最新[全系统审查](SYSTEM-AUDIT-OFFICIAL-FIRST-2026-11-06.md)按30区域/38事项列出现状与缺口,接续[官方优先路线](DEVELOPMENT-ROADMAP-OFFICIAL-FIRST.md)M1/M2。默认开放/透传、历史真实交付/网络/unknown/重复读取/Purgeunknown保留,旧预算/审批设计按历史时点理解。本轮只审查规划,未发行或重启。 + 容量后备来源已补齐:Opus5.5官方同步128000/上下文1000000公开已知,源码2bd7b24a按精确型号采用,保留显式参数与目录优先;独立67/67/网页4组通过,[官方容量验收](QA-CLAUDE-OFFICIAL-CAPACITY.md)。未部署,不把原调查未知续写为官方未公布,也不把官方声明当真实代理或Purge收益通过。 F后容量/兼容调查已有独立源码进展:New API中文305正文/201接口审读与SHA/blob核对完成;目录刷新ae5123c5、实际输出字段诊断341f5f69及独立候选126/126/build通过。见[兼容复核](QA-NEWAPI-COMPATIBILITY-REVIEW.md);尚未发行/正式升级,不将源码修复或本机夹具当真实Claude容量/供应商兼容、角色收益或Purge收益通过。历史失败与unknown保持,默认开放策略不回退。 diff --git a/docs/PRODUCT-PLAN.md b/docs/PRODUCT-PLAN.md index 9a469a41..edbc1d6d 100644 --- a/docs/PRODUCT-PLAN.md +++ b/docs/PRODUCT-PLAN.md @@ -1,5 +1,7 @@ # Agent Workbench 后续产品与执行开发方案 +当前入口更新(2026-11-06):官方直连优先,代理/兼容按各自原生协议与真实结果适配;自动发现模型与版本化官方能力、用户覆盖分别管理。[全系统审查](SYSTEM-AUDIT-OFFICIAL-FIRST-2026-11-06.md)明确正式F/1.25.12、后续三源码修复未部署及30区域/38事项。[开发路线](DEVELOPMENT-ROADMAP-OFFICIAL-FIRST.md)以正确性/残留上限、官方契约/缓存、独立发行、新真实交付及收益验证为顺序;全球原生、多模态、服务端工具、交互PTY/电脑操作和生态按批补齐。本轮审查不修改产品源码或部署;下方旧版本与预算设计按历史理解,不恢复默认额度/逐步审批。 + 1.20.1已完成真实案例修复及保留状态本地升级:原任务目标/无需测试约定、只读诊断、系统审查来源、完整无参协议兼容、失败请求用量、无进展保护和实际模型展示均已修复,并补齐暂停封存及Windows进程退出竞态。独立源码626/626、桌面/390px核心5组、相关历史UI5类、同载荷安装/重启/卸载7组、正式26页和实际Electron两视口通过。原PID36552/Provider及模型连接保持,17项目/30全部会话/3144记录/5986事件、全部表每行/全导出/模型中心及1824工件记录/1721去重文件原字节保持,9229关闭。本轮真实模型请求0,未续跑原用户任务或改写历史用量。 见[验收](QA-1.20.1.md)、[回顾](RETROSPECTIVE-1.20.1.md)和[发行身份](task-control-release-identity.json)。下一[1.21方案](PLAN-1.21.md)保持计划。 ## 1. 产品目标 diff --git a/docs/RESEARCH-ROADMAP.md b/docs/RESEARCH-ROADMAP.md index 32524918..9e89103e 100644 --- a/docs/RESEARCH-ROADMAP.md +++ b/docs/RESEARCH-ROADMAP.md @@ -1,5 +1,7 @@ # REA与代理机制研究路线 +当前入口更新(2026-11-06):[官方优先全系统审查](SYSTEM-AUDIT-OFFICIAL-FIRST-2026-11-06.md)和[开发路线](DEVELOPMENT-ROADMAP-OFFICIAL-FIRST.md)为最新执行入口。已有固定DeepSeek Harness/Purge来源及真实AD/AE交付;Codex/Claude Code等完整引擎、全球原生协议、缓存/角色/经验普遍收益仍缺完整证明。官方资料、网关文档、客户端机制来源和实际模型/工程结果分开;本轮21编号观察/官方公共来源只做审查,不代替真实兼容或Purge收益验证。下方2026-10-31描述保留原研究时点。 + 更新:2026-10-31。本文记录研究与落地路线;1.8已将REA三项静态分析接入工作台,1.9已完成固定DeepSeek Harness六个SDK方法探针的连接核验,缺少session/cancel的方法错误保持原值。1.12.1[里程碑审查](MILESTONE-AUDIT.md)及1.13隔离协作的独立/便携/保留状态发行已完成。五款客户端完整机制、受控外部引擎工程执行及Purge增强收益按[PLAN-1.14](PLAN-1.14.md)与[NEXT-STEPS](NEXT-STEPS.md)继续研究和实测。 来源身份登记见 [research/targets.json](../research/targets.json)。产品目标见 [PRODUCT-PLAN.md](PRODUCT-PLAN.md),能力ID见 [CAPABILITIES.md](CAPABILITIES.md)。 diff --git a/docs/SYSTEM-AUDIT-OFFICIAL-FIRST-2026-11-06.md b/docs/SYSTEM-AUDIT-OFFICIAL-FIRST-2026-11-06.md new file mode 100644 index 00000000..a34e291a --- /dev/null +++ b/docs/SYSTEM-AUDIT-OFFICIAL-FIRST-2026-11-06.md @@ -0,0 +1,155 @@ +# 官方优先全系统审查(2026-11-06) + +系统已经有 Node、React、Python 自主工程交付的真实成功样本,但仍未达到“全球模型官方能力完整对接”的目标。主要问题是请求参数保真、官方型号能力适配、残留执行上限和功能说明与行为不一致;一些高级模型能力及完整外部引擎尚未开发。缓存、角色、经验、重复读取改善和 Purge 的普遍收益也没有完整证明。 + +本次记录 **30 个主要能力区域、38 个问题/缺项/未知事项**,其中 **8 项 P0、25 项 P1、5 项 P2**。这不是“38 个全部复现的产品故障”:每项分别标注确认缺陷、未实现、源码未部署、静态风险或验证未知。I30 的当前文档入口在本轮纠正,其余事项没有在本轮修改产品源码。逐项开发与验收顺序见[开发路线](DEVELOPMENT-ROADMAP-OFFICIAL-FIRST.md),机器可读矩阵见[审查数据](official-first-system-audit-2026-11-06.json)。 + +## 当前真实状态 + +- 审查基线:`4da2550becdf4750320cedd5ee30a7e38baba2f9`,分支 `feat/task-agreement-1.25`;审查开始工作树干净并与远端一致。 +- 新只读观察:`2026-11-06T13:08:18Z`,正式 `/api/health` 返回 **1.25.12 / PID 71628 / ready=true**。浏览器 `?release=1.25.6` 是旧查询串,不能作为正式程序身份。 +- 正式源码仍沿 F 发行 `6867cd5a3094c133b09a7c50384ae844ef34c038`。F 的数据、Provider、原实例和凭据存在状态保持验收见[原F报告](QA-ROLE-ROUTING-F.md);本轮 health 不代表这些数据再次逐项核对。 +- **三项修复已提交并推送、尚未部署**:`ae5123c5` 已有目录元数据刷新、`341f5f69` 按实际 wire 说明截断、`2bd7b24a` 精确官方 Claude 容量后备。各有独立验收,见[New API兼容复核](QA-NEWAPI-COMPATIBILITY-REVIEW.md)和[Claude官方容量](QA-CLAUDE-OFFICIAL-CAPACITY.md)。 +- 本轮没有真实模型请求、真实Key读取、用户项目/正式数据写入、发行、安装或重启。实际界面复核使用自有临时数据与已核对未改的独立候选。 + +## 审查依据与范围 + +以官方文档、官方SDK/schema和精确模型ID为协议及能力依据;New API 是网关/兼容参考。代理返回的型号声明、目录元数据、真实接口行为和官方规格分别保存,型号相同不能认证底层模型。 + +已核对当前 `src` 272文件/31738行、`tests` 171文件/20981行、`scripts` 272文件/24917行,715个清单输入SHA一致。覆盖供应商、请求/流式、任务与权限、工程工具、协作、上下文/缓存、记忆/提醒、恢复/存储、浏览器/桌面/手机、集成及发行等全层关键入口。这是清单与关键路径审查,**不是每行代码都经过人工安全审计,也不是全球每个官方网站全部读完**。 + +新离线核对形成 O01–O21 共21个编号观察,另有 O18-C 纠正;脚本退出0表示完成了现状复现,不能记成“21项产品正确性测试全过”。实际自有网页在1280和390检查模型表单、控件及保存;0页面错误、0真实模型请求。首次390截图只有表头,随后另做真实控件/完整表单观察;这些不能代替正式Electron验收。O21只使用合成元数据复现保存登记总数门槛,没有读取或解密真实Key,也没有删除原metadata。 + +复用此前 New API 的305篇正文/201份嵌入接口定义完整审读及 Claude 容量原件。新/复用官方公共来源共24条可取得记录;17条失败读取记录原样保存。失败次数不是供应商数,也不是模型请求失败。来源URL、最终URL、SHA、读取范围、复用或不可取得状态见[来源账本](official-first-sources-2026-11-06.json)。 + +| 官方来源 | 这次依据 | 已确认的协议要求与边界 | +| --- | --- | --- | +| OpenAI | 固定官方 `openai/openai-python` commit `c6fc264689d384d5caec150fafa1a9567c4a3f56` 的Chat/Responses/Models/Tools类型 | Chat的max_tokens/max_completion_tokens、Responses的max_output_tokens是可选字段;原生reasoning/text/stream_options/工具/后台状态各有语义。Models类型只有ID/时间/归属/关闭日期,没有完整容量与能力。网页指南403,不能据SDK类型补造当前所有型号容量/缓存TTL。 | +| Claude | 已保存的官方导出,抽取11篇完整至下一篇边界的文章做关键段落审查;整个导出只取得前缀 | 当前四精确ID的同步输出128000/上下文1000000已公开;300K属于Batches beta,不能混为同步。Opus5.5不接受旧enabled/disabled/between_tools;Sonnet5.5有带effort条件的between_tools;Haiku5.5禁用也有条件。Messages max_tokens是必填。官方示例API version为2026-10-21,Bearer推荐且旧x-api-key仍支持;没有证明2023-06-01必然失效。 | +| DeepSeek | create/thinking/cache官方正文 | 工具请求要保留所有历史reasoning_content,包括没调用工具的轮次;思考模式、effort、采样/输出语义按型号/协议变化。缓存为best effort,官方不保证100%命中。 | +| Qwen / DashScope | 官方OpenAI兼容正文 | 兼容请求仍有供应商自己的思考/内容/模型范围;通用OpenAI表单不能代表完整官方特性。 | +| Z.ai | 官方Chat schema正文 | thinking、工具、用量等有自己的请求约束;已读定义不等于真实服务已验收。 | +| Kimi | 官方Chat正文(moonshot官方入口重定向kimi) | prompt_cache_options的implicit/5m/1h;cached、cache_write和uncached互斥且合为输入总量。当前raw保留了写缓存,但归一展示没完整接上。 | +| AWS Bedrock / Microsoft Azure | 官方Converse及端点切换正文 | 云平台原生路由、部署和认证需要独立适配;部分OpenAI兼容入口与完整原生支持分开。 | +| Google Gemini / Mistral / xAI | 本轮公共页/索引超时原记录 | 不冒称读完或已验证。Gemini原生缺项由当前只支持三协议的源码确认;其他供应商原生契约仍须取得。 | + +“官方为准”意味着模型发现、精确型号能力、协议版本、用户显式值和实际结果都有来源。并非所有上游 `/models` 都返回完整规格,因此实现应是**自动获取目录 + 有版本的官方能力注册表 + 连接实际声明/观察 + 用户覆盖**,不要求普通用户逐个手填,也不从型号子串猜完整能力。 + +本地默认继续透传、不裁剪历史、不先行挡预算。官方必填字段按用户明确值或可靠官方/上游规格发送;例如Messages必填max_tokens不可以发Infinity或省略。界面应分别显示模型最大能力与“本地不限”,不能把本地2048当模型最大值。上游原错、适配缺项、本地读取/执行错误和unknown用量分别显示并保留原件。 + +## 能力分层状态 + +下面的“已部署”沿已有发行证据,“本轮观察”指这次新的隔离核对,不能互相代替。原成功和原失败并存。 + +| 能力区域 | 实现与验证 | 正式状态 / 剩余事项 | +| --- | --- | --- | +| C01 模型发现、多连接、多模型池 | 已有目录/绑定/池;本机与历史真实样本 | 1.25.12;刷新修复仍源码态,自动能力选择/全能力注册表缺口 I04/I13/I15/I29 | +| C02 原生参数、推理开关与强度 | 已有控件/JSON/wire;本轮复现参数丢失和模式错误 | I01/I02/I03/I08未修 | +| C03 默认透传与官方容量 | 开放/不限已部署;精确官方容量源码修复有67项相关验收 | 官方容量修复未部署 I29 | +| C04 官方认证与原生协议 | Chat/Responses/Messages、Bearer/x-api-key | 云平台认证、API版本策略、Gemini原生缺项 I14/I16/I17 | +| C05 JSON/SSE、错误原文与用量 | 已有完整来源/工件/unknown;历史网络原错保持 | 复杂块/候选、缓存归一、完整契约覆盖仍缺 I02/I10/I18/I21/I28 | +| C06 三种批准模式/完整访问 | ask为初始模式;用户切换持久;full文件/命令/Git自动执行 | MCP/检查点/手机仍有固定次数,部分提示不实 I05/I06/I07/I08/I33 | +| C07 型号身份记录 | 记录请求、报告型号、映射及原响应 | 不是权重认证;开放实际策略与文案不符 I08 | +| C08 自动适配与连续性 | 现有池/Run亲和 | 缺会话级状态/缓存亲和,自动选模缺陷 I04/I11/I15 | +| C09 多模态与文件 | 当前截图/PNG基本视觉已有验收 | PDF/音视频/原生文件生命周期缺项 I20 | +| C10 服务端工具/后台/候选 | 原生参数部分能发;完整工作流未实现 | 本地仍拒未知服务端块/暂停,tool_choice不可原生配置 I18/I19/I21/I37 | +| C11 文件/改动集/Git/回滚 | 已实现;真实Node/React/Python有成功交付 | 已部署;不能据工具流程证明所有新任务都会成功 | +| C12 环境/依赖/检查 | 已实现;AH解释器修复及真实工程原测试 | 已部署;原Python误选失败仍保留 | +| C13 终端/日志/停止/Git证据 | 所属命令与AI终态证据已实现 | 已部署;pipes不等于交互PTY I22 | +| C14 要求证据/单轮观察/终态 | A/B/AI已实现并验收 | 语义质量与普遍收益仍需另测 I24/I35 | +| C15 工具表/前缀/重复读取观察 | C/D已实现,实际wire来源可核对 | 观察不能证明缓存命中或重复读因果改善 I12/I35 | +| C16 缓存控制/统计 | 基础上游cached usage及原raw已保留 | 块/工具注解、write归一、会话亲和/指纹缺项 I09–I12 | +| C17 提醒及实际回读 | D默认关闭、持久开关/原收据已验收 | 1.25.10起部署;提醒出现不是业务完成/收益证明 I35 | +| C18 经验草稿/采用/撤销 | E来源绑定/修订/采用分离已验收 | 1.25.11起部署;真实语义收益未知 I35 | +| C19 Worker角色/模型/成果回收 | F已实现实际派发与来源绑定 | 1.25.12;要求not_evaluated、skills/MCP未装配 I24/I25 | +| C20 Skills/MCP | 本地来源注册/stdio工具已有实现 | Schema子集、次数和子任务装配仍缺 I05/I06/I08/I25 | +| C21 浏览器/文件预览/DOM动作 | 已实现,基础文件保存预览和两端曾验收 | 固定浏览动作不能代表任意电脑控制 I23 | +| C22 桌面桥/本机凭据 | Windows文件对话框、安装、DPAPI可选加密保存已实现 | 临时Key退出丢失是明确内存策略;不能说系统没有保存能力。通用OS控制未实现 I23;超过101登记拒加载 I38 | +| C23 外部引擎 | 固定DeepSeek Harness桥及原版来源已集成 | 完整Codex/Claude Code引擎未实现;使用它们的机制/模型API不会自动获得完整产品能力 | +| C24 原版Purge | 来源锁/加载/隔离链路已有证据 | 默认关闭;AG不具可比成功对照,收益unknown I36 | +| C25 REA | 静态分析job、来源与工件已实现 | 不等于五款客户端全机制研究/完整引擎接入完成 | +| C26 手机Companion | 查看/批准/停止契约已实现 | 当前新真实设备流程未验,操作总量2000上限 I32/I33 | +| C27 跨产品/Mnemo | 原件只读消费反馈与映射已有 | 原生导出/入站委托/撤销服务缺项 I31 | +| C28 SQLite/事件/工件/恢复 | 已实现完整原件与恢复归属 | 固定检查点上限与全量读写规模风险 I07/I13/I26 | +| C29 源码发行与安装回退 | 原源码包/程序包/输入重建/跨版回退有独立验收 | 源码打包和用户数据已经分离;数据保持核对属于部署。CI及新源码发行待办 I27/I28/I29 | +| C30 Orven品牌与界面 | 三配色与网页/桌面已交付;本轮自有表单1280/390实看 | 模型能力/权限说明需纠正 I03/I08;当前规划入口I30本轮已纠正 | + +## 逐项问题与缺项 + +P0 是应先处理的正确性、开放执行和正式版本落差;P1是核心兼容/工程能力或验证闭环;P2是扩展能力与测量后推进的规模工作。完整“下一改动、验收条件、依赖、观察与源码SHA”在机器可读矩阵中;下面保留具体影响及源码入口。 + +未归为真实故障的历史残留也已核对:内部demo配置和旧连接/测试接口仍有源码,但正常提交入口会要求选择真实模型并打开模型中心,不会据此自动生成假交付;旧UI入口移除与内部兼容清理分开。MCP当前是本地stdio工具子集,远程传输/更完整协议生态须按实际官方契约扩展,不能泛称全生态已经接入。71个静态搜索候选不是71个确认缺陷。 + +当前完整性解释仍会把未支持内容块/上游截断等标为本地协议诊断;移除本地额度没有让上游length消失。后续I18/I21要把原返回、停止原因、可续接状态与实际工具执行分开,保留完整原文,不能笼统再要求用户改本地2048。 + +| ID / 优先级 | 事项与证据等级 | 实际影响 / 入口 | +| --- | --- | --- | +| I01 P0 | **推理快捷编辑丢失嵌套原生字段**
确认缺陷;O01、O02 | 修改 effort 或 thinking 会删除 summary、output_config.format、thinking.display,用户请求未如实保留。
[src/shared/contracts/model-parameter-profiles.mjs](../src/shared/contracts/model-parameter-profiles.mjs) 第54行;[src/shared/contracts/model-parameter-profiles.mjs](../src/shared/contracts/model-parameter-profiles.mjs) 第57行 | +| I02 P0 | **Chat 流式覆盖用户 stream_options**
确认缺陷;O17 | include_usage=false 与其他上游 stream_options 被整对象替换。
[src/shared/contracts/model-request.mjs](../src/shared/contracts/model-request.mjs) 第22行 | +| I03 P0 | **Claude 型号推理能力与官方未对齐**
确认缺陷;O03、O04 | Opus5.5 仍能生成官方拒绝的 enabled/disabled;between_tools 未解析;全否定目录又回退全部选项。
[src/shared/contracts/model-catalog.mjs](../src/shared/contracts/model-catalog.mjs) 第38行;[src/shared/contracts/model-parameter-profiles.mjs](../src/shared/contracts/model-parameter-profiles.mjs) 第27行 | +| I04 P0 | **自动模型池可选明确不支持工具的成员**
自动选择缺陷;O08 | 需要工程工具时,自动模式仍可把请求发给明确不支持工具的模型。
[src/core/models/model-hub.mjs](../src/core/models/model-hub.mjs) 第321行 | +| I05 P0 | **开放模式仍有每任务 30 次 MCP 操作上限**
确认残留上限;O14 | 长工程在已有开放权限下仍被固定次数挡住。
[src/core/tools/mcp-registry.mjs](../src/core/tools/mcp-registry.mjs) 第121行 | +| I06 P0 | **MCP JSON Schema 子集拒绝常见合法工具**
确认适配缺项;O13 | anyOf 和65属性等普通 schema 无法接入,与开放工具能力目标不符。
[src/core/tools/mcp-contract.mjs](../src/core/tools/mcp-contract.mjs) 第24行 | +| I07 P0 | **检查点仍受 30/2000/64MiB 固定上限**
确认残留上限;O15 | 工具引用2000项、检查点30个或结果64MiB可阻断开放工程恢复。
[src/core/tasks/task-checkpoint-manager.mjs](../src/core/tasks/task-checkpoint-manager.mjs) 第15行 | +| I08 P1 | **能力/权限说明与实际行为不一致**
确认说明不实;O05、O16、O20 | 结构化参数被说成拒绝、MCP仍提示确认、require-match显示暂停但开放模式继续。
[src/shared/contracts/model-parameter-profiles.mjs](../src/shared/contracts/model-parameter-profiles.mjs) 第81行;[src/core/tools/mcp-contract.mjs](../src/core/tools/mcp-contract.mjs) 第74行 | +| I09 P1 | **Messages 块/工具缓存注解丢失**
确认适配缺项;O06 | 顶层 cache_control 能透传,但系统/内容块和工具上的缓存边界被转换丢掉。
[src/shared/contracts/anthropic-messages.mjs](../src/shared/contracts/anthropic-messages.mjs) 第48行 | +| I10 P1 | **Kimi cache-write 未归一展示**
确认统计缺项;O07 | 原 usage 还在,但归一统计缺 cache_write_tokens/uncached,不能正确比较缓存成本。
[src/core/models/protocols.mjs](../src/core/models/protocols.mjs) 第19行 | +| I11 P1 | **亲和仅绑定 Run,缺会话缓存与状态连续性**
确认路由缺项;O09 | 续接新Run可换连接/型号;跨型号历史与DeepSeek reasoning_content不能视为可无损互换。
[src/core/models/model-hub.mjs](../src/core/models/model-hub.mjs) 第326行 | +| I12 P1 | **前缀观察没有供应商缓存相关参数指纹**
确认观察缺项;O19 | 改effort仍显示same_prompt;这个观察不能解释供应商实际缓存命中。
[src/core/context/prefix-audit.mjs](../src/core/context/prefix-audit.mjs) 第30行 | +| I13 P1 | **请求账本显示100/宣称1000,无法完整分页**
确认可观察性缺项;O10 | 开放模式已保存1001以上尝试,但视图仅100且宣称保留1000,历史难核查。
[src/core/models/model-hub.mjs](../src/core/models/model-hub.mjs) 第148行 | +| I14 P1 | **官方认证/版本/端点策略未形成完整契约**
部分实现;源码/既有QA | 仅Bearer/x-api-key;anthropic-version写死2023-06-01,缺版本/beta/额外头策略。旧版本失效未被证明。
[src/server/provider.mjs](../src/server/provider.mjs) 第186行 | +| I15 P1 | **模型发现不等于完整官方能力注册表**
部分实现;O12 | 目录容量/功能字段不完整;OpenAI Models类型无容量,Gemini nextPageToken未识别。
[src/shared/contracts/model-catalog.mjs](../src/shared/contracts/model-catalog.mjs) 第19行;[src/shared/contracts/model-capabilities.mjs](../src/shared/contracts/model-capabilities.mjs) 第1行 | +| I16 P1 | **缺 Gemini 原生协议**
未实现;源码/既有QA | 能解析类似目录不等于支持generateContent/streamGenerateContent和原生内容/思考状态。
[src/shared/contracts/model-capabilities.mjs](../src/shared/contracts/model-capabilities.mjs) 第1行 | +| I17 P1 | **缺云平台原生认证与路由**
未实现;源码/既有QA | 不能把Bedrock/Vertex/Foundry各种认证或部署ID当普通Bearer/v1。部分Azure兼容端点可能可用但未验。
[src/server/provider.mjs](../src/server/provider.mjs) 第189行 | +| I18 P1 | **服务端工具与官方停止/续接语义缺项**
未实现且本地拒绝;O11 | Claude server_tool_use/pause_turn和Responses多种服务端输出被本地协议拒绝。
[src/core/models/protocols.mjs](../src/core/models/protocols.mjs) 第120行;[src/core/models/messages-stream.mjs](../src/core/models/messages-stream.mjs) 第11行 | +| I19 P1 | **原生工具选择与用户服务端工具不可配置**
未实现;源码/既有QA | tools/tool_choice为托管字段,普通原生JSON无法开启官方web/computer等能力或选择官方tool_choice。
[src/shared/contracts/native-parameters.mjs](../src/shared/contracts/native-parameters.mjs) 第10行 | +| I20 P1 | **PDF/音频/视频/文件等原生内容缺项**
基础图像之外未实现;源码/既有QA | 当前截图/PNG基本视觉不是全球模型全多模态;JPEG被转PNG,缺官方原生文件生命周期。
[src/core/context/input-attachments.mjs](../src/core/context/input-attachments.mjs) 第3行 | +| I21 P1 | **Chat 多候选非流式/流式处理不一致**
确认契约缺项;O18、O18-C | 非流式留raw多choices但只消费第一个;流式非0候选本地拒绝。O18首推断已单独纠正。
[src/core/models/model-stream.mjs](../src/core/models/model-stream.mjs) 第86行;[src/server/provider.mjs](../src/server/provider.mjs) 第386行 | +| I22 P1 | **终端缺交互PTY/输入会话**
未实现;源码/既有QA | 已有所属命令、完整日志/停止/Git证据,但pipes一次性输入不能替代交互终端。
[src/core/owned-command.mjs](../src/core/owned-command.mjs) 第13行 | +| I23 P2 | **缺通用电脑操作能力**
未实现;源码/既有QA | 已有文件对话框/浏览器动作,不是任意Windows应用/全桌面computer use。
[src/core/desktop/desktop-bridge.mjs](../src/core/desktop/desktop-bridge.mjs) 第9行 | +| I24 P1 | **Worker要求语义仍未评估**
部分实现/收益未知;源码/既有QA | 角色/模型/成果与检查身份已绑定,但每条requirements仍为not_evaluated,不能认证角色质量。
[src/core/tasks/worker-evidence.mjs](../src/core/tasks/worker-evidence.mjs) 第49行 | +| I25 P1 | **Worker没有skills/MCP能力装配**
确认装配缺项;源码/既有QA | 父代理目录不能自动带给子任务,某些专业任务实际没有必要工具。
[src/server/runtime.mjs](../src/server/runtime.mjs) 第461行 | +| I26 P2 | **长历史读写与响应内存规模风险**
静态风险/未实测;源码/既有QA | 启动全量加载,会话/尝试整份回写,长raw响应内存累积;未测到确定故障。
[src/core/storage/state-database.mjs](../src/core/storage/state-database.mjs) 第55行;[src/core/models/model-hub.mjs](../src/core/models/model-hub.mjs) 第40行 | +| I27 P1 | **缺仓库持续集成工作流**
当前Git文件未发现;源码/既有QA | 本地QA丰富但版本/适配器/包校验不能稳定由远端流水线复现。
当前Git文件/文档入口清单 | +| I28 P1 | **供应商契约测试覆盖与官方更新闭环缺口**
验证缺口;源码/既有QA | 当前测试未覆盖这次发现的嵌套丢失/新推理模式/覆盖等官方契约。
[docs/QA-NEWAPI-COMPATIBILITY-REVIEW.md](QA-NEWAPI-COMPATIBILITY-REVIEW.md);[docs/QA-CLAUDE-OFFICIAL-CAPACITY.md](QA-CLAUDE-OFFICIAL-CAPACITY.md) | +| I29 P0 | **三项已推送修复尚未正式发行**
源码已修/未部署;源码/既有QA | 正式1.25.12仍是F旧源码;目录刷新/wire截断说明/官方Claude容量没有进入正式程序。
[docs/QA-ROLE-ROUTING-F.md](QA-ROLE-ROUTING-F.md);[docs/QA-NEWAPI-COMPATIBILITY-REVIEW.md](QA-NEWAPI-COMPATIBILITY-REVIEW.md) | +| I30 P1 | **当前规划/能力文档入口停留D/E旧时点**
文档已在本轮纠正;源码/既有QA | F已交付却仍有当前页说未实现,读者难判断源码和部署状态。
当前Git文件/文档入口清单 | +| I31 P2 | **跨产品原生导出/入站委托服务缺项**
未实现;源码/既有QA | Mnemo目前只有来源包只读核对与反馈;没有运行中的导出/委托/撤销闭环。
[docs/THREE-SYSTEM-INTERFACE-FEEDBACK-2026-11-06.md](THREE-SYSTEM-INTERFACE-FEEDBACK-2026-11-06.md) | +| I32 P2 | **新物理手机完整流程未验收**
真实设备未验;源码/既有QA | 已有Companion实现与本地契约验收不能代替当前真实手机端到端。
[src/core/companion/device-registry.mjs](../src/core/companion/device-registry.mjs) 第12行 | +| I33 P1 | **Companion操作账本达到2000后拒绝**
源码确认上限/未真机触发;源码/既有QA | 累计操作达到2000后手机操作409,并非开放模式动态账本。
[src/core/companion/device-registry.mjs](../src/core/companion/device-registry.mjs) 第124行 | +| I34 P1 | **全球官方与代理真实兼容没有完整证明**
已有样本之外未验;源码/既有QA | AD/AE有DeepSeek实际成功,也有Claude400/目录断连;不能据此宣称全球全部可用。
[docs/QA-LIVE-VALIDATION-AD.md](QA-LIVE-VALIDATION-AD.md);[docs/QA-LIVE-ENGINEERING-AE.md](QA-LIVE-ENGINEERING-AE.md) | +| I35 P1 | **重复读取/缓存/经验/角色普遍收益未证明**
收益未证实;源码/既有QA | D/E/F机制已部署,单例缓存与工具事实不证明广泛因果收益。
[docs/QA-REMINDER-READBACK-D.md](QA-REMINDER-READBACK-D.md);[docs/QA-EXPERIENCE-DRAFTS-E.md](QA-EXPERIENCE-DRAFTS-E.md) | +| I36 P1 | **Purge AB/BA收益仍unknown**
网络混杂/收益未知;源码/既有QA | AG四组只有ba-off自主完成,其余三个ETIMEDOUT/unknown,不能做收益比较。
[docs/QA-LIVE-PURGE-AG.md](QA-LIVE-PURGE-AG.md) | +| I37 P2 | **批处理/后台会话/文件与其他模型资源API缺项**
完整生命周期未实现;源码/既有QA | 原生字段能部分转交不等于支持异步任务轮询/取消、批次/文件/向量/音视频图像资源全生命周期。
[src/shared/contracts/model-capabilities.mjs](../src/shared/contracts/model-capabilities.mjs) 第1行;[src/core/models/protocols.mjs](../src/core/models/protocols.mjs) 第12行 | +| I38 P1 | **加密凭据元数据超过101项被整体拒加载**
合成元数据复现加载总数上限;O21 | 102项保存登记会标记unavailable,开放连接数与本机保存加载范围不一致;原metadata不删除。
[src/core/models/credential-vault.mjs](../src/core/models/credential-vault.mjs) 第20行 | + +## 保留的真实样本与未验证结论 + +| 样本 | 原结果 | 当前可以下的结论 | +| --- | --- | --- | +| [W](QA-LIVE-DELIVERY-W.md) | 6次型号matched,71044tokens,只读、0提案,no_progress停止,没有交付 | 当时模型只读行为保持;没有走到R第7次预留卡点,v2真实孤立异常剔除边界未验 | +| [AD](QA-LIVE-VALIDATION-AD.md) | 全批21请求/199250reported+2unknown;10预检8过;Claude两次400缺max_tokens;DeepSeek Node11请求完成原9tests/build/Git | Node自主交付已有真实成功;Claude原错误与目录断连保留;不能推出所有供应商兼容通过 | +| [AE](QA-LIVE-ENGINEERING-AE.md) | React/Python各10请求,合计20请求/367365reported/0unknown,原测试/构建/限定Git通过 | 两工程已有真实成功;当时Python内置解释器误选及终端报告缺口保持,AH/AI后来独立修复并部署 | +| [AG](QA-LIVE-PURGE-AG.md) | 11请求/109888reported+3unknown;ab-off/ab-on/ba-on为connect ETIMEDOUT,ba-off自主交付;开启组原版插件确实加载 | 四组不等成功,Purge收益仍unknown;事后DNS/TLS通过不能倒改原模型请求网络失败 | + +AD的122624缓存输入tokens是那个样本的上游报告;D/E/F的机制验收和新O19前缀观察都不能升级为“缓存永不掉、重复读取普遍减少、角色/经验一定有收益”。官方自身的TTL、驱逐、型号/参数变化和网络环境仍会影响缓存,应尽量减少工作台造成的前缀变化并准确报告实际命中。 + +## 下一步的明确顺序 + +1. **首批修正确性和开放模式残留**:I01–I08、I33/I38,补供应商契约夹具/实际wire核对;先完成改动、验收与回顾,再本地提交。 +2. **建立官方能力与协议闭环**:I14/I15/I28,明确官方/代理身份、型号、版本、来源、用户覆盖;补缓存元数据和使用统计I09–I13。 +3. **冻结一次独立发行**:带上现有三修复及首批修复,程序/源码包分别核验;安装保持核对只在正式部署阶段做。此次审查没有打包或重启。 +4. **分批扩官方与工程能力**:Gemini/云平台、服务端工具/媒体,再交互终端/Worker能力;每批有固定官方契约、实际核心路径和独立提交,不能只新增一个“支持”标签。 +5. **另开真实验证**:官方直连优先,代理样本独立;先兼容和Node/React/Python自主交付,再分别验证缓存/重复读/角色/经验,最后新冻结Purge AB/BA。旧失败不删除、不自动重放。 + +详细阶段交付与逐项验收见[开发路线](DEVELOPMENT-ROADMAP-OFFICIAL-FIRST.md)。后续新增测试用于发现的真实边界和完整流程;不反复重跑已冻结的无关大套测试,也不为文档改动冒称源码回归。 + +## 调查限制与纠正 + +- O18原推断误把Chat流式说成选第一个候选,原脚本/结果未修改;O18-C内存SSE独立确认“非流式选choices[0]、流式非0本地拒绝”。 +- MCP注册配置字段仍为500–45000ms;实际unrestricted执行超时已是0。这里确认的运行限制是30次操作与Schema子集,不能说full运行仍45秒停止。 +- 全量存储/响应内存是静态规模风险,没有新的压力基准或OOM实测。 +- 官方页面403/超时、调查输出截断、猜测路径不存在及第一次数据组装找错图像源码锚点均保留为调查过程,未算成产品/供应商测试失败。来源账本与计划说明具体范围。 +- 本轮修改审查/路线/当前文档入口,未修产品代码、未部署;当前缺陷与未知事项由后续独立阶段继续处理。 + +原始新证据在 `.qa/official-first-system-audit-2026-11-06/`(Git忽略);提交的是带SHA的审查数据、来源索引与[封存摘要](official-first-audit-evidence-2026-11-06.json)。历史报告和旧原件不重写。 diff --git a/docs/TIANSHU-ADOPTION-PLAN.md b/docs/TIANSHU-ADOPTION-PLAN.md index 0b7078e1..b8a311d1 100644 --- a/docs/TIANSHU-ADOPTION-PLAN.md +++ b/docs/TIANSHU-ADOPTION-PLAN.md @@ -1,5 +1,7 @@ # Tianshu 研究后的 Agent Workbench 采用计划 +当前入口更新(2026-11-06):A要求证据/B单轮观察/C工具前缀/D提醒/E经验/F角色机制已经沿各自发行证据落地,正式只读health为1.25.12;机制完成不等于普遍收益通过。G真实AD/AE已有交付,AG网络失败致Purge收益unknown;H要求的来源/失败/回顾/提交纪律继续保持。[全层审查](SYSTEM-AUDIT-OFFICIAL-FIRST-2026-11-06.md)和[官方优先路线](DEVELOPMENT-ROADMAP-OFFICIAL-FIRST.md)是最新顺序:先参数保真/残留上限与官方契约,再独立发行及新真实实验。D标题中的旧预算不恢复为默认本地限制,下方历史方案原样保留。 + Claude容量纠正:官方已公开Opus5.5同步128000/上下文1000000,2bd7b24a已实现目录缺项时采用精确官方规格,独立67/67/网页4组通过,见[官方容量验收](QA-CLAUDE-OFFICIAL-CAPACITY.md)。正式旧源码尚未部署此修复;下方未知是原调查时点,真实代理兼容/Purge收益仍单独验收。 New API中文305正文/201接口的全量学习与能力对照完成,源码新增目录事实刷新ae5123c5和实际wire截断说明341f5f69,独立候选相关126/126及build通过,见[兼容复核](QA-NEWAPI-COMPATIBILITY-REVIEW.md)。本轮未部署、重启或请求真实模型;F正式记录和Claude旧2048声明仍保持,缺失容量不能猜测。真实角色/缓存/重复读取普遍改善与Purge收益仍unknown,下一发行与新真实实验分别验收。 diff --git a/docs/official-first-audit-evidence-2026-11-06.json b/docs/official-first-audit-evidence-2026-11-06.json new file mode 100644 index 00000000..c2e63caa --- /dev/null +++ b/docs/official-first-audit-evidence-2026-11-06.json @@ -0,0 +1,50 @@ +{ + "schema": 1, + "kind": "official-first-audit-evidence-seal-summary", + "sourceHead": "4da2550becdf4750320cedd5ee30a7e38baba2f9", + "sealedAt": "2026-11-06T14:30:31.248488+00:00", + "manifestPath": ".qa/official-first-system-audit-2026-11-06/MANIFEST.json", + "manifestSha256": "a6119c607bfbbc96a876afd499298356c4a79404e47acfa47acec9bc18b33b9d", + "files": 73, + "bytes": 2365463, + "deliverableInputs": [ + { + "path": "docs/SYSTEM-AUDIT-OFFICIAL-FIRST-2026-11-06.md", + "bytes": 28557, + "sha256": "027e40be202793b18f31c53a6ba411b8436286015d0f09671b3deb55516df22b" + }, + { + "path": "docs/DEVELOPMENT-ROADMAP-OFFICIAL-FIRST.md", + "bytes": 21331, + "sha256": "ce3b67678be1575f0f28ac925d77456852ce5be3101ed2b5eb484bf42c2b97d3" + }, + { + "path": "docs/official-first-system-audit-2026-11-06.json", + "bytes": 79121, + "sha256": "46b808a5809d04d6968b81a7d2b37ec15899fdc0208af5757ba067d94c13652b" + }, + { + "path": "docs/official-first-sources-2026-11-06.json", + "bytes": 106114, + "sha256": "d230cfa36f12653757c9fb73d238dcbf0790d8587e3a66dffcb83765c5caaf59" + } + ], + "excludedFromThisCutoff": [ + "MANIFEST.json", + "verification/**", + "mutable commit/push/progress receipts outside raw evidence" + ], + "scope": "all new raw audit files at cutoff plus four immutable final report/data inputs; source-only docs task", + "recordedObservations": { + "numbered": 21, + "corrections": [ + "O18-C" + ] + }, + "verificationReceipt": "docs/official-first-audit-validation-2026-11-06.json", + "realModelRequests": 0, + "realCredentialsRead": 0, + "productionWrites": 0, + "releaseOrRestart": false, + "validationMeaning": "integrity and audit completion; not global provider acceptance or 21 passing product tests" +} diff --git a/docs/official-first-audit-validation-2026-11-06.json b/docs/official-first-audit-validation-2026-11-06.json new file mode 100644 index 00000000..35955306 --- /dev/null +++ b/docs/official-first-audit-validation-2026-11-06.json @@ -0,0 +1,69 @@ +{ + "schema": 1, + "kind": "official-first-audit-integrity-verification", + "checkedAt": "2026-11-06T14:30:53.410667+00:00", + "sourceHead": "4da2550becdf4750320cedd5ee30a7e38baba2f9", + "status": "passed", + "meaning": "audit integrity and complete matrix/roadmap mapping, not product regression or live-provider success", + "manifestPath": ".qa/official-first-system-audit-2026-11-06/MANIFEST.json", + "manifestSha256": "a6119c607bfbbc96a876afd499298356c4a79404e47acfa47acec9bc18b33b9d", + "rawFilesVerified": 73, + "rawBytesVerified": 2365463, + "immutableReportDataInputsVerified": 4, + "sourceInventoryVerified": 715, + "programSourceChanges": [], + "issuesVerified": 38, + "capabilitiesVerified": 30, + "numberedObservations": 21, + "corrections": [ + "O18-C" + ], + "stagesVerified": 9, + "allIssueDependenciesValidAndAcyclic": true, + "publicSourcesVerified": 24, + "unavailableSourceRecordsPreserved": 17, + "newAndCurrentPointerLinksVerified": 89, + "pointerDocumentsVerified": [ + "AGENTS.md", + "docs/CAPABILITIES.md", + "docs/NEXT-STEPS.md", + "docs/PRODUCT-PLAN.md", + "docs/RESEARCH-ROADMAP.md", + "docs/TIANSHU-ADOPTION-PLAN.md", + "docs/PLAN-1.25.md" + ], + "ownedUi": { + "priorBuildInputs": 7, + "compiledOutputs": 30, + "webObservations": 5, + "mobileDimensions": { + "viewport": 390, + "document": 390 + }, + "errors": [] + }, + "diffCheck": "passed", + "changedDocumentPaths": [ + "AGENTS.md", + "docs/CAPABILITIES.md", + "docs/DEVELOPMENT-ROADMAP-OFFICIAL-FIRST.md", + "docs/NEXT-STEPS.md", + "docs/OFFICIAL-FIRST-SYSTEM-AUDIT-PLAN.md", + "docs/PLAN-1.25.md", + "docs/PRODUCT-PLAN.md", + "docs/RESEARCH-ROADMAP.md", + "docs/SYSTEM-AUDIT-OFFICIAL-FIRST-2026-11-06.md", + "docs/TIANSHU-ADOPTION-PLAN.md", + "docs/official-first-audit-evidence-2026-11-06.json", + "docs/official-first-sources-2026-11-06.json", + "docs/official-first-system-audit-2026-11-06.json", + "findings.md", + "progress.md", + "task_plan.md" + ], + "secretScan": "no key material detected", + "realModelRequests": 0, + "realCredentialsRead": 0, + "productionWrites": 0, + "releaseOrRestart": false +} diff --git a/docs/official-first-sources-2026-11-06.json b/docs/official-first-sources-2026-11-06.json new file mode 100644 index 00000000..2d1c2bd2 --- /dev/null +++ b/docs/official-first-sources-2026-11-06.json @@ -0,0 +1,2186 @@ +{ + "schema": 1, + "kind": "official-first-audit-public-source-ledger", + "recordedAt": "2026-11-06T14:09:15.016711+00:00", + "sourceHead": "4da2550becdf4750320cedd5ee30a7e38baba2f9", + "officialSdkCommit": "c6fc264689d384d5caec150fafa1a9567c4a3f56", + "acquired": [ + { + "id": "deepseek-create", + "requestedUrl": "https://api-docs.deepseek.com/api/create-chat-completion", + "observedAt": "2026-11-06T13:05:31.918033+00:00", + "readLimitBytes": 3500000, + "status": 200, + "finalUrl": "https://api-docs.deepseek.com/api/create-chat-completion/", + "contentType": "text/html", + "contentLength": "129498", + "bytes": 129498, + "completeByReadLimit": true, + "sha256": "b43ca950bd04ae7f6f537b42e6cb729f543d942dcd848ec43ad4e9bfe059fe1f", + "bodyPath": ".qa/official-first-system-audit-2026-11-06/official-public/deepseek-create.body", + "attemptLedger": ".qa/official-first-system-audit-2026-11-06/official-public/attempts.json", + "authority": "vendor-owned-public-source", + "readStatus": "targeted-sections-reviewed", + "reviewPath": ".qa/official-first-system-audit-2026-11-06/official-public/deepseek-create.text", + "reviewScope": [ + "max_tokens", + "reasoning_effort", + "thinking", + "usage" + ], + "reviewAnchors": [ + { + "line": 193, + "text": "(Beta) Used for the thinking mode in the Chat Prefix Completion feature as the input for the CoT in the last assistant message. When using this feature, the prefix parameter must be set to true.", + "snippetComplete": true + }, + { + "line": 287, + "text": "thinking", + "snippetComplete": true + }, + { + "line": 291, + "text": "Controls the switch between thinking and non-thinking mode. When thinking is omitted or null, the mode follows reasoning_effort; if reasoning_effort is also omitted, thinking is enabled. When this object is provided, type is required and takes precedence over reasoning_effort for selecting the mode.", + "snippetComplete": true + }, + { + "line": 297, + "text": "If set to enabled, then use thinking mode. If set to disabled, then use non-thinking model.", + "snippetComplete": true + }, + { + "line": 299, + "text": "reasoning_effort string", + "snippetComplete": true + }, + { + "line": 303, + "text": "Controls the thinking mode toggle and the thinking effort. none disables thinking mode; low / high / max enable thinking mode. The default effort is high. For compatibility with existing software, minimal is accepted and mapped to low, and medium / xhigh are accepted and mapped to high.", + "snippetComplete": true + }, + { + "line": 305, + "text": "max_tokens integernullable", + "snippetComplete": true + }, + { + "line": 311, + "text": "The value must be between 1 and 384K (393216). When not set, the default is 8K in non-thinking mode, 64K in thinking mode (128K with reasoning_effort set to max). Please refer to the Models & Pricing page for details.", + "snippetComplete": true + }, + { + "line": 319, + "text": "Important: When using JSON Output, you must also instruct the model to produce JSON yourself via a system or user message. If the model's output cannot be recognized as JSON, the API replaces the non-JSON content with spaces (U+0020). As a result, the message content may contain only spaces (appearing empty), even when finish_reason is stop. Also note that the message content may be partially cut off if finish_reason=\"length\", which indicates the generation exceeded max_tokens or the conversatio", + "snippetComplete": false + }, + { + "line": 362, + "text": "include_usage boolean", + "snippetComplete": true + }, + { + "line": 364, + "text": "If set to true, all chunks in the stream will include a usage field, whose value is null on every chunk except the last one. If omitted or set to false, the usage field is absent from all chunks except the last one.", + "snippetComplete": true + }, + { + "line": 366, + "text": "Either way, the last chunk before the data: [DONE] message carries the token usage statistics for the entire request in its usage field. Note that no separate usage-only chunk is emitted: the statistics ride on the last content chunk, whose choices array always contains exactly one element that carries no new content and a non-null finish_reason.", + "snippetComplete": true + }, + { + "line": 376, + "text": "We generally recommend altering this or top_p but not both. Has no effect in thinking mode.", + "snippetComplete": true + }, + { + "line": 386, + "text": "deepseek-flash: In thinking mode, the default is 1.0 and values below 0.95 are treated as 0.95. In non-thinking mode, it is fixed at 1.0 and the value you pass is ignored.", + "snippetComplete": true + } + ], + "notEvidenceOf": [ + "live provider acceptance", + "all vendor features implemented", + "full website review" + ] + }, + { + "id": "deepseek-thinking", + "requestedUrl": "https://api-docs.deepseek.com/guides/thinking_mode", + "observedAt": "2026-11-06T13:05:31.918033+00:00", + "readLimitBytes": 3500000, + "status": 200, + "finalUrl": "https://api-docs.deepseek.com/guides/thinking_mode/", + "contentType": "text/html", + "contentLength": "117778", + "bytes": 117778, + "completeByReadLimit": true, + "sha256": "42151b5f86827bff663a61693080a61cdbf99074dee94af756c9fce855baa7cd", + "bodyPath": ".qa/official-first-system-audit-2026-11-06/official-public/deepseek-thinking.body", + "attemptLedger": ".qa/official-first-system-audit-2026-11-06/official-public/attempts.json", + "authority": "vendor-owned-public-source", + "readStatus": "targeted-sections-reviewed", + "reviewPath": ".qa/official-first-system-audit-2026-11-06/official-public/deepseek-thinking.text", + "reviewScope": [ + "reasoning_content", + "reasoning_effort", + "thinking mode", + "all previous" + ], + "reviewAnchors": [ + { + "line": 3, + "text": "Thinking Mode | DeepSeek API Docs", + "snippetComplete": true + }, + { + "line": 22, + "text": "Thinking Mode", + "snippetComplete": true + }, + { + "line": 34, + "text": "Thinking Mode", + "snippetComplete": true + }, + { + "line": 40, + "text": "Thinking Mode", + "snippetComplete": true + }, + { + "line": 42, + "text": "The DeepSeek model supports the thinking mode: before outputting the final answer, the model will first output a chain-of-thought reasoning to improve the accuracy of the final response.", + "snippetComplete": true + }, + { + "line": 44, + "text": "Thinking Mode Toggle and Effort Control​", + "snippetComplete": true + }, + { + "line": 47, + "text": "Thinking Mode Toggle(1){\"thinking\": {\"type\": \"enabled/disabled\"}}{\"reasoning\": {\"effort\": \"none/low/high/max\"}}(none disables thinking mode)", + "snippetComplete": true + }, + { + "line": 48, + "text": "Thinking Effort Control(2){\"reasoning_effort\": \"low/high/max\"}{\"output_config\": {\"effort\": \"low/high/max\"}}", + "snippetComplete": true + }, + { + "line": 50, + "text": "(1) Thinking mode is enabled by default, with the default effort being high", + "snippetComplete": true + }, + { + "line": 63, + "text": "response = client.chat.completions.create( model=\"deepseek-flash\", # ... reasoning_effort=\"high\", extra_body={\"thinking\": {\"type\": \"enabled\"}},)", + "snippetComplete": true + }, + { + "line": 67, + "text": "Thinking mode does not support the temperature, presence_penalty, or frequency_penalty parameters. Please note that, for compatibility with existing software, setting these parameters will not trigger an error but will also have no effect.", + "snippetComplete": true + }, + { + "line": 71, + "text": "deepseek-flash: only takes effect in thinking mode, where the effective range is 0.95–1.0; values below 0.95 are treated as 0.95. In non-thinking mode it is fixed at 1.0 and your value is ignored.", + "snippetComplete": true + }, + { + "line": 73, + "text": "deepseek-v4-pro: in thinking mode your value is ignored (fixed at 0.95 when tool calling is enabled, otherwise 1.0); in non-thinking mode it takes effect. Setting tool_choice to none disables tool calling even when tools is provided.", + "snippetComplete": true + }, + { + "line": 75, + "text": "In thinking mode, the chain-of-thought content is returned via the reasoning_content parameter, at the same level as content. In subsequent requests, whether reasoning_content should be passed back and whether it will be concatenated into the context depends on whether the request carries the tools parameter:", + "snippetComplete": true + } + ], + "notEvidenceOf": [ + "live provider acceptance", + "all vendor features implemented", + "full website review" + ] + }, + { + "id": "deepseek-cache", + "requestedUrl": "https://api-docs.deepseek.com/guides/kv_cache", + "observedAt": "2026-11-06T13:05:31.918033+00:00", + "readLimitBytes": 3500000, + "status": 200, + "finalUrl": "https://api-docs.deepseek.com/guides/kv_cache/", + "contentType": "text/html", + "contentLength": "35850", + "bytes": 35850, + "completeByReadLimit": true, + "sha256": "061c0c7055ceb96d0a522ae1829ff5f7126ff08f26af217643e371abb740a081", + "bodyPath": ".qa/official-first-system-audit-2026-11-06/official-public/deepseek-cache.body", + "attemptLedger": ".qa/official-first-system-audit-2026-11-06/official-public/attempts.json", + "authority": "vendor-owned-public-source", + "readStatus": "targeted-sections-reviewed", + "reviewPath": ".qa/official-first-system-audit-2026-11-06/official-public/deepseek-cache.text", + "reviewScope": [ + "prefix", + "best-effort", + "prompt_cache_hit_tokens", + "prompt_cache_miss_tokens" + ], + "reviewAnchors": [ + { + "line": 24, + "text": "Chat Prefix Completion (Beta)", + "snippetComplete": true + }, + { + "line": 44, + "text": "Each user request will trigger the construction of a hard disk cache. If subsequent requests have overlapping prefixes with previous requests, the overlapping part will only be fetched from the cache, which counts as a \"cache hit.\"", + "snippetComplete": true + }, + { + "line": 48, + "text": "A cache hit requires that the corresponding prefix has already been \"persisted\" (written to the disk cache). Due to the Sliding Window Attention mechanism, the storage and matching of cached prefixes differs from before. Each cached prefix is an independent, complete unit. A subsequent request can only hit the cache if it fully matches a cache prefix unit.", + "snippetComplete": true + }, + { + "line": 50, + "text": "When cache prefixes are persisted:​", + "snippetComplete": true + }, + { + "line": 52, + "text": "Persistence at request boundaries: Each request will produce two cache prefix units at the end position of the user input and the end position of the model output. A subsequent request can hit the cache if it fully matches them.", + "snippetComplete": true + }, + { + "line": 54, + "text": "Common prefix detection persistence: When the system detects a common prefix across multiple requests, it will persist that common prefix as an independent cache prefix unit. A subsequent request can hit the cache if it fully reuses that cache prefix unit.", + "snippetComplete": true + }, + { + "line": 56, + "text": "Persistence at fixed token intervals: For long inputs or long outputs, the system will carve out cache prefix units at fixed token intervals, to avoid long prefixes from being completely uncacheable due to never reaching an end position.", + "snippetComplete": true + }, + { + "line": 58, + "text": "Example 1: A user's first-round request is A + B, and the second-round request is A + B + C. The second request can fully match the cache prefix unit A + B, hitting the cache for A + B. See Example 1 below.", + "snippetComplete": true + }, + { + "line": 60, + "text": "Example 2: A user's first-round request is A + B, and the second-round request is A + C. The second request cannot hit the cache, because A + C does not fully match the first round's cache prefix unit (A + B). However, at this point the system will detect that the two requests share a common prefix A, and persist A as a cache prefix unit. When a third-round request A + D arrives, it can fully match the cache prefix unit A, hitting the cache for A. See Example 2 below.", + "snippetComplete": true + }, + { + "line": 72, + "text": "In this example, the second request can fully reuse the cache prefix unit from the first request, which will count as a \"cache hit.\"", + "snippetComplete": true + }, + { + "line": 88, + "text": "In the above example, the first two requests will not hit the cache. After the first two requests are completed, the system will identify the system message + in the user message as a cache prefix unit and persist it. In the third request, since it fully matches the previously persisted cache prefix unit, it can hit the cache.", + "snippetComplete": true + }, + { + "line": 94, + "text": "prompt_cache_hit_tokens: The number of tokens in the input of this request that resulted in a cache hit.", + "snippetComplete": true + }, + { + "line": 96, + "text": "prompt_cache_miss_tokens: The number of tokens in the input of this request that did not result in a cache hit.", + "snippetComplete": true + }, + { + "line": 100, + "text": "The hard disk cache only matches the prefix part of the user's input. The output is still generated through computation and inference, and it is influenced by parameters such as temperature, introducing randomness.", + "snippetComplete": true + } + ], + "notEvidenceOf": [ + "live provider acceptance", + "all vendor features implemented", + "full website review" + ] + }, + { + "id": "qwen-compatible", + "requestedUrl": "https://help.aliyun.com/zh/model-studio/compatibility-of-openai-with-dashscope", + "observedAt": "2026-11-06T13:07:38.225266+00:00", + "readLimitBytes": 3500000, + "status": 200, + "finalUrl": "https://help.aliyun.com/zh/model-studio/compatibility-of-openai-with-dashscope", + "contentType": "text/html;charset=UTF-8", + "contentLength": null, + "bytes": 167084, + "completeByReadLimit": true, + "sha256": "0b58c3f79d2e330af6457a5c98cd347397ab8db4cd09c7d50e0cb41d6daf6c1a", + "bodyPath": ".qa/official-first-system-audit-2026-11-06/official-public/qwen-compatible.body", + "attemptLedger": ".qa/official-first-system-audit-2026-11-06/official-public/attempts.json", + "authority": "vendor-owned-public-source", + "readStatus": "targeted-sections-reviewed", + "reviewPath": ".qa/official-first-system-audit-2026-11-06/official-public/qwen-compatible.text", + "reviewScope": [ + "enable_thinking", + "thinking_budget", + "compatible-mode", + "model", + "usage" + ], + "reviewAnchors": [ + { + "line": 3, + "text": " 如何通过OpenAI接口调用千问模型-大模型服务平台百炼(Model Studio)-阿里云帮助中心", + "snippetComplete": true + }, + { + "line": 55, + "text": "北京:https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", + "snippetComplete": true + }, + { + "line": 56, + "text": "弗吉尼亚:https://{WorkspaceId}.us-east-1.maas.aliyuncs.com/compatible-mode/v1", + "snippetComplete": true + }, + { + "line": 57, + "text": "新加坡:https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1", + "snippetComplete": true + }, + { + "line": 58, + "text": "日本(东京):https://{WorkspaceId}.ap-northeast-1.maas.aliyuncs.com/compatible-mode/v1", + "snippetComplete": true + }, + { + "line": 62, + "text": "北京:POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1/chat/completions", + "snippetComplete": true + }, + { + "line": 63, + "text": "弗吉尼亚:POST https://{WorkspaceId}.us-east-1.maas.aliyuncs.com/compatible-mode/v1/chat/completions", + "snippetComplete": true + }, + { + "line": 64, + "text": "新加坡:POST https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1/chat/completions", + "snippetComplete": true + }, + { + "line": 65, + "text": "日本(东京):POST https://{WorkspaceId}.ap-northeast-1.maas.aliyuncs.com/compatible-mode/v1/chat/completions", + "snippetComplete": true + }, + { + "line": 122, + "text": " # 新加坡和北京地域的API Key不同。获取API Key:https://help.aliyun.com/zh/model-studio/get-api-key", + "snippetComplete": true + }, + { + "line": 125, + "text": " base_url=\"https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1\",", + "snippetComplete": true + }, + { + "line": 128, + "text": " model=\"qwen3.8-max\", # 此处以qwen-plus为例,可按需更换模型名称。模型列表:https://help.aliyun.com/zh/model-studio/getting-started/models", + "snippetComplete": true + }, + { + "line": 132, + "text": " print(completion.model_dump_json())", + "snippetComplete": true + }, + { + "line": 155, + "text": " \"model\": \"qwen3.8-max\",", + "snippetComplete": true + } + ], + "notEvidenceOf": [ + "live provider acceptance", + "all vendor features implemented", + "full website review" + ] + }, + { + "id": "zai-chat", + "requestedUrl": "https://docs.z.ai/api-reference/llm/chat-completion", + "observedAt": "2026-11-06T13:07:38.865783+00:00", + "readLimitBytes": 3500000, + "status": 200, + "finalUrl": "https://docs.z.ai/api-reference/llm/chat-completion", + "contentType": "text/markdown; charset=utf-8", + "contentLength": null, + "bytes": 42735, + "completeByReadLimit": true, + "sha256": "5e608d3c82b283ed8dac0b7991187b625e378ca706a060f0eaa40ddf90d1e110", + "bodyPath": ".qa/official-first-system-audit-2026-11-06/official-public/zai-chat.body", + "attemptLedger": ".qa/official-first-system-audit-2026-11-06/official-public/attempts.json", + "authority": "vendor-owned-public-source", + "readStatus": "targeted-sections-reviewed", + "reviewPath": ".qa/official-first-system-audit-2026-11-06/official-public/zai-chat.body", + "reviewScope": [ + "thinking", + "max_tokens", + "tool_choice", + "stream", + "usage" + ], + "reviewAnchors": [ + { + "line": 7, + "text": "> Create a chat completion model that generates AI replies for given conversation messages. It supports multimodal inputs (text, images, audio, video, file), offers configurable parameters (like temperature, max tokens, tool use), and supports both streaming and non-streaming output modes.", + "snippetComplete": true + }, + { + "line": 38, + "text": " max tokens, tool use), and supports both streaming and non-streaming", + "snippetComplete": true + }, + { + "line": 63, + "text": " stream: false", + "snippetComplete": true + }, + { + "line": 64, + "text": " Stream Example:", + "snippetComplete": true + }, + { + "line": 75, + "text": " stream: true", + "snippetComplete": true + }, + { + "line": 76, + "text": " Thinking Example:", + "snippetComplete": true + }, + { + "line": 86, + "text": " thinking:", + "snippetComplete": true + }, + { + "line": 88, + "text": " stream: true", + "snippetComplete": true + }, + { + "line": 103, + "text": " stream: true", + "snippetComplete": true + }, + { + "line": 165, + "text": " tool_choice: auto", + "snippetComplete": true + }, + { + "line": 345, + "text": " stream:", + "snippetComplete": true + }, + { + "line": 354, + "text": " standard Event Stream. When the Event Stream ends, a `data: [DONE]`", + "snippetComplete": true + }, + { + "line": 356, + "text": " thinking:", + "snippetComplete": true + }, + { + "line": 357, + "text": " $ref: '#/components/schemas/ChatThinking'", + "snippetComplete": true + } + ], + "notEvidenceOf": [ + "live provider acceptance", + "all vendor features implemented", + "full website review" + ] + }, + { + "id": "moonshot-chat", + "requestedUrl": "https://platform.moonshot.cn/docs/api/chat", + "observedAt": "2026-11-06T13:07:38.880185+00:00", + "readLimitBytes": 3500000, + "status": 200, + "finalUrl": "https://platform.kimi.com/docs/api/chat", + "contentType": "text/markdown; charset=utf-8", + "contentLength": null, + "bytes": 62409, + "completeByReadLimit": true, + "sha256": "db19cb0b1047af8a9b534f0e67d0cedb529265527986ed31c873d09697558409", + "bodyPath": ".qa/official-first-system-audit-2026-11-06/official-public/moonshot-chat.body", + "attemptLedger": ".qa/official-first-system-audit-2026-11-06/official-public/attempts.json", + "authority": "vendor-owned-public-source", + "readStatus": "targeted-sections-reviewed", + "reviewPath": ".qa/official-first-system-audit-2026-11-06/official-public/moonshot-chat.body", + "reviewScope": [ + "prompt_cache_options", + "cache_write_tokens", + "cached_tokens", + "uncached_tokens" + ], + "reviewAnchors": [ + { + "line": 208, + "text": "> \"cache_write_tokens\": 256", + "snippetComplete": true + }, + { + "line": 229, + "text": "> data: {\"id\":\"chatcmpl-6adb0ef5ba679679914a2719\",\"object\":\"chat.completion.chunk\",\"created\":1792741109,\"model\":\"kimi-k3.1-pro\",\"choices\":[],\"usage\":{\"prompt_tokens\":309,\"completion_tokens\":101,\"total_tokens\":410,\"completion_tokens_details\":{\"reasoning_tokens\":71},\"prompt_tokens_details\":{\"cache_write_tokens\":256}}}", + "snippetComplete": true + }, + { + "line": 240, + "text": "> Cache Write 适用于 `kimi-k3.1-pro`、`kimi-k3.1-pro-highspeed` 和 `kimi-k3`。Kimi API 会对重复的请求前缀自动启用上下文缓存,命中部分以更低价格计费且响应更快。通过 `prompt_cache_options` 可以控制缓存写入行为:", + "snippetComplete": true + }, + { + "line": 244, + "text": "> | `prompt_cache_options` | object | 上下文缓存写入选项。不传时默认开启缓存写入(`5m` 档) |", + "snippetComplete": true + }, + { + "line": 245, + "text": "> | `prompt_cache_options.mode` | `\"implicit\"` | 缓存写入模式,仅支持 `implicit`:自动将请求前缀写入缓存 |", + "snippetComplete": true + }, + { + "line": 246, + "text": "> | `prompt_cache_options.ttl` | `\"5m\"` \\| `\"1h\"` | 缓存写入的有效期,仅支持 `5m`、`1h` 两档(默认 `5m`),两档缓存相互独立 |", + "snippetComplete": true + }, + { + "line": 252, + "text": "> \"prompt_cache_options\": { \"mode\": \"implicit\", \"ttl\": \"1h\" }", + "snippetComplete": true + }, + { + "line": 270, + "text": "> | `usage.prompt_tokens_details.cached_tokens` | 命中缓存的 Token 数量(缓存读取) |", + "snippetComplete": true + }, + { + "line": 271, + "text": "> | `usage.prompt_tokens_details.cache_write_tokens` | 本次写入缓存的 Token 数量(缓存写入) |", + "snippetComplete": true + }, + { + "line": 273, + "text": "> `cached_tokens`、`cache_write_tokens` 与未缓存部分互斥,三者之和等于 `prompt_tokens`,即未缓存部分 = `prompt_tokens` − `cached_tokens` − `cache_write_tokens`。", + "snippetComplete": true + }, + { + "line": 483, + "text": "> data: {\"id\":\"chatcmpl-6adb0efa76cd871ffe47cb4d\",\"object\":\"chat.completion.chunk\",\"created\":1792741114,\"model\":\"kimi-k3.1-pro\",\"choices\":[],\"usage\":{\"prompt_tokens\":90,\"completion_tokens\":1139,\"total_tokens\":1229,\"completion_tokens_details\":{\"reasoning_tokens\":620},\"prompt_tokens_details\":{\"cache_write_tokens\":0}}}", + "snippetComplete": true + }, + { + "line": 863, + "text": " cached_tokens:", + "snippetComplete": true + }, + { + "line": 869, + "text": " 输入 Token 的分类明细。cached_tokens、cache_write_tokens 与未缓存部分互斥,三者之和等于", + "snippetComplete": true + }, + { + "line": 872, + "text": " cached_tokens:", + "snippetComplete": true + } + ], + "notEvidenceOf": [ + "live provider acceptance", + "all vendor features implemented", + "full website review" + ] + }, + { + "id": "bedrock-converse", + "requestedUrl": "https://docs.aws.amazon.com/bedrock/latest/APIReference/API_runtime_Converse.html", + "observedAt": "2026-11-06T13:07:40.021384+00:00", + "readLimitBytes": 3500000, + "status": 200, + "finalUrl": "https://docs.aws.amazon.com/bedrock/latest/APIReference/API_runtime_Converse.html", + "contentType": "text/markdown", + "contentLength": "33273", + "bytes": 33273, + "completeByReadLimit": true, + "sha256": "85ac5b5807d42802d888781df64c6ee6ee44c5b5261ced83691b6022cb21a59a", + "bodyPath": ".qa/official-first-system-audit-2026-11-06/official-public/bedrock-converse.body", + "attemptLedger": ".qa/official-first-system-audit-2026-11-06/official-public/attempts.json", + "authority": "vendor-owned-public-source", + "readStatus": "targeted-sections-reviewed", + "reviewPath": ".qa/official-first-system-audit-2026-11-06/official-public/bedrock-converse.body", + "reviewScope": [ + "Request Syntax", + "inferenceConfig", + "additionalModelRequestFields", + "cacheReadInputTokens", + "cacheWriteInputTokens" + ], + "reviewAnchors": [ + { + "line": 12, + "text": "You can also submit a prompt from Prompt management by specifying the ARN of the prompt version and including a map of variables to values in the `promptVariables` field. You can append more messages to the prompt by using the `messages` field. If you use a prompt from Prompt management, you can't include the following fields in the request: `additionalModelRequestFields`, `inferenceConfig`, `system`, or `toolConfig`. Instead, these fields must be defined through Prompt management. For more info", + "snippetComplete": false + }, + { + "line": 25, + "text": "## Request Syntax", + "snippetComplete": true + }, + { + "line": 33, + "text": " \"additionalModelRequestFields\": {{JSON value}},", + "snippetComplete": true + }, + { + "line": 40, + "text": " \"inferenceConfig\": { ", + "snippetComplete": true + }, + { + "line": 106, + "text": " ** [additionalModelRequestFields](#API_runtime_Converse_RequestSyntax) ** ", + "snippetComplete": true + }, + { + "line": 107, + "text": "Additional inference parameters that the model supports, beyond the base set of inference parameters that `Converse` and `ConverseStream` support in the `inferenceConfig` field. For more information, see [Model parameters](https://docs.aws.amazon.com/bedrock/latest/userguide/model-parameters.html). ", + "snippetComplete": true + }, + { + "line": 126, + "text": " ** [inferenceConfig](#API_runtime_Converse_RequestSyntax) ** ", + "snippetComplete": true + }, + { + "line": 127, + "text": "Inference parameters to pass to the model. `Converse` and `ConverseStream` support a base set of inference parameters. If you need to pass additional parameters that the model supports, use the `additionalModelRequestFields` request field. ", + "snippetComplete": true + }, + { + "line": 128, + "text": "Type: [InferenceConfiguration](API_runtime_InferenceConfiguration.md) object ", + "snippetComplete": true + }, + { + "line": 431, + "text": " \"cacheReadInputTokens\": number,", + "snippetComplete": true + }, + { + "line": 432, + "text": " \"cacheWriteInputTokens\": number,", + "snippetComplete": true + }, + { + "line": 553, + "text": " \"inferenceConfig\": {", + "snippetComplete": true + }, + { + "line": 598, + "text": "In the following example, the request passess a field (`top_k`) that the `Converse` field doesn't support. You pass the additional field in the `additionalModelRequestFields` field. The example also shows how to set the paths for the additional fields sent in the response from the model. ", + "snippetComplete": true + }, + { + "line": 619, + "text": " \"inferenceConfig\": {", + "snippetComplete": true + } + ], + "notEvidenceOf": [ + "live provider acceptance", + "all vendor features implemented", + "full website review" + ] + }, + { + "id": "azure-endpoints", + "requestedUrl": "https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/switching-endpoints", + "observedAt": "2026-11-06T13:07:40.454248+00:00", + "readLimitBytes": 3500000, + "status": 200, + "finalUrl": "https://learn.microsoft.com/en-us/azure/foundry-classic/openai/how-to/switching-endpoints", + "contentType": "text/markdown; charset=utf-8", + "contentLength": "6356", + "bytes": 6356, + "completeByReadLimit": true, + "sha256": "10920c1be8f3c1ac6d8f3a49f2ad292e2831905042ba95c38e4c0974befb13be", + "bodyPath": ".qa/official-first-system-audit-2026-11-06/official-public/azure-endpoints.body", + "attemptLedger": ".qa/official-first-system-audit-2026-11-06/official-public/attempts.json", + "authority": "vendor-owned-public-source", + "readStatus": "targeted-sections-reviewed", + "reviewPath": ".qa/official-first-system-audit-2026-11-06/official-public/azure-endpoints.body", + "reviewScope": [ + "base_url", + "api_key", + "token_provider", + "deployment", + "v1/" + ], + "reviewAnchors": [ + { + "line": 65, + "text": "| ```python
import os
from openai import OpenAI

client = OpenAI(
api_key=os.getenv(\"OPENAI_API_KEY\")
)



``` | ```python
import os
from openai import OpenAI

client = OpenAI(
api_key=os.getenv(\"AZURE_OPENAI_API_KEY\"),
base_url = \"https://YOUR-RESOURCE-NAME.openai.azure.com/openai/v1/\"
)
``` |", + "snippetComplete": true + }, + { + "line": 71, + "text": "| ```python
import os
from openai import OpenAI

client = OpenAI(
api_key=os.getenv(\"OPENAI_API_KEY\")
)








``` | ```python
from azure.identity import DefaultAzureCredential, get_bearer_token_provider
from openai import OpenAI

token_provider = get_bearer_token_provider(
DefaultAzureCredential(), \"https://ai.azure.com/.default\"
)

client = OpenAI(
base_url = \"https://YOUR-RESOURCE-NAME.openai.azure.com/openai", + "snippetComplete": false + }, + { + "line": 75, + "text": "OpenAI uses the `model` keyword argument to specify what model to use. Azure OpenAI has the concept of unique model [deployments](create-resource?pivots=web-portal#deploy-a-model). When you use Azure OpenAI, `model` should refer to the underlying deployment name you chose when you deployed the model.", + "snippetComplete": true + }, + { + "line": 79, + "text": "When you access the model via the API in Azure OpenAI, you need to refer to the deployment name rather than the underlying model name in API calls, which is one of the [key differences](switching-endpoints) between OpenAI and Azure OpenAI. OpenAI only requires the model name. Azure OpenAI always requires deployment name, even when using the model parameter. In our docs, we often have examples where deployment names are represented as identical to model names to help indicate which model works wi", + "snippetComplete": false + }, + { + "line": 83, + "text": "| ```python
response = client.responses.create(
model=\"gpt-4.1-nano\", # Replace with your model deployment name
input=\"This is a test.\"
)

chat_completion = client.chat.completions.create(
model=\"gpt-4o\",
messages=\"\"
)

embedding = client.embeddings.create(
model=\"text-embedding-3-large\",
input=\"\"
)
``` | ```python
response = client.responses.create(
model=\"gpt-4.1-nano\", # Replace with your mo", + "snippetComplete": false + }, + { + "line": 91, + "text": "| ```python
inputs = [\"A\", \"B\", \"C\"]

embedding = client.embeddings.create(
input=inputs,
model=\"text-embedding-3-large\"
)


``` | ```python
inputs = [\"A\", \"B\", \"C\"] #max array size=2048

embedding = client.embeddings.create(
input=inputs,
model=\"text-embedding-3-large\" # This must match the custom deployment name you chose for your model.
# engine=\"text-embedding-ada-002\"
)

``` |", + "snippetComplete": true + } + ], + "notEvidenceOf": [ + "live provider acceptance", + "all vendor features implemented", + "full website review" + ] + }, + { + "id": "openai-sdk-commit", + "requestedUrl": "https://api.github.com/repos/openai/openai-python/commits/main", + "observedAt": "2026-11-06T13:30:56.222Z", + "credentials": false, + "totalTimeoutMs": 18000, + "status": 200, + "finalUrl": "https://api.github.com/repos/openai/openai-python/commits/main", + "contentType": "application/json; charset=utf-8", + "bytes": 95778, + "sha256": "bbdefbc797ccb444f22af75122532ebee5052c648ffb47b28e1415744eb6f49a", + "bodyPath": ".qa/official-first-system-audit-2026-11-06/official-public/openai-sdk-commit.body", + "attemptLedger": ".qa/official-first-system-audit-2026-11-06/official-public/openai-sdk-attempts.json", + "authority": "vendor-owned-public-source", + "readStatus": "source-identity-only", + "reviewPath": ".qa/official-first-system-audit-2026-11-06/official-public/openai-sdk-commit.body", + "reviewScope": [], + "reviewAnchors": [], + "notEvidenceOf": [ + "live provider acceptance", + "all vendor features implemented", + "full website review" + ] + }, + { + "id": "openai-sdk-tools", + "requestedUrl": "https://raw.githubusercontent.com/openai/openai-python/c6fc264689d384d5caec150fafa1a9567c4a3f56/src/openai/types/responses/tool_param.py", + "observedAt": "2026-11-06T13:30:57.740Z", + "credentials": false, + "totalTimeoutMs": 18000, + "status": 200, + "finalUrl": "https://raw.githubusercontent.com/openai/openai-python/c6fc264689d384d5caec150fafa1a9567c4a3f56/src/openai/types/responses/tool_param.py", + "contentType": "text/plain; charset=utf-8", + "bytes": 16632, + "sha256": "8b5ddf4bf72fa3704808626189c91969dce7d2039967e30036dc46d698455af8", + "bodyPath": ".qa/official-first-system-audit-2026-11-06/official-public/openai-sdk-tools.body", + "attemptLedger": ".qa/official-first-system-audit-2026-11-06/official-public/openai-sdk-attempts.json", + "authority": "vendor-owned-public-source", + "readStatus": "targeted-sections-reviewed", + "reviewPath": ".qa/official-first-system-audit-2026-11-06/official-public/openai-sdk-tools.body", + "reviewScope": [ + "ToolParam", + "WebSearch", + "Computer", + "Mcp", + "Shell", + "ApplyPatch", + "ImageGeneration" + ], + "reviewAnchors": [ + { + "line": 9, + "text": "from .custom_tool_param import CustomToolParam", + "snippetComplete": true + }, + { + "line": 10, + "text": "from .computer_tool_param import ComputerToolParam", + "snippetComplete": true + }, + { + "line": 11, + "text": "from .function_tool_param import FunctionToolParam", + "snippetComplete": true + }, + { + "line": 12, + "text": "from .namespace_tool_param import NamespaceToolParam", + "snippetComplete": true + }, + { + "line": 14, + "text": " Filters as _compat_WebSearchToolFilters,", + "snippetComplete": true + }, + { + "line": 15, + "text": " UserLocation as _compat_WebSearchToolUserLocation,", + "snippetComplete": true + }, + { + "line": 16, + "text": " WebSearchToolParam,", + "snippetComplete": true + }, + { + "line": 17, + "text": " WebSearchToolParam as _compat_WebSearchTool,", + "snippetComplete": true + }, + { + "line": 19, + "text": "from .apply_patch_tool_param import ApplyPatchToolParam", + "snippetComplete": true + }, + { + "line": 20, + "text": "from .file_search_tool_param import FileSearchToolParam", + "snippetComplete": true + }, + { + "line": 21, + "text": "from .tool_search_tool_param import ToolSearchToolParam", + "snippetComplete": true + }, + { + "line": 22, + "text": "from .function_shell_tool_param import FunctionShellToolParam", + "snippetComplete": true + }, + { + "line": 23, + "text": "from .web_search_preview_tool_param import WebSearchPreviewToolParam", + "snippetComplete": true + }, + { + "line": 24, + "text": "from .computer_use_preview_tool_param import ComputerUsePreviewToolParam", + "snippetComplete": true + } + ], + "notEvidenceOf": [ + "live provider acceptance", + "all vendor features implemented", + "full website review" + ] + }, + { + "id": "openai-sdk-responses", + "requestedUrl": "https://raw.githubusercontent.com/openai/openai-python/c6fc264689d384d5caec150fafa1a9567c4a3f56/src/openai/types/responses/response_create_params.py", + "observedAt": "2026-11-06T13:30:57.738Z", + "credentials": false, + "totalTimeoutMs": 18000, + "status": 200, + "finalUrl": "https://raw.githubusercontent.com/openai/openai-python/c6fc264689d384d5caec150fafa1a9567c4a3f56/src/openai/types/responses/response_create_params.py", + "contentType": "text/plain; charset=utf-8", + "bytes": 21917, + "sha256": "dbce54c51a1d1b3764616fe5d723acf4af219a3ca3f3c8199abec60b3d59b4b7", + "bodyPath": ".qa/official-first-system-audit-2026-11-06/official-public/openai-sdk-responses.body", + "attemptLedger": ".qa/official-first-system-audit-2026-11-06/official-public/openai-sdk-attempts.json", + "authority": "vendor-owned-public-source", + "readStatus": "targeted-sections-reviewed", + "reviewPath": ".qa/official-first-system-audit-2026-11-06/official-public/openai-sdk-responses.body", + "reviewScope": [ + "max_output_tokens:", + "reasoning:", + "text:", + "prompt_cache_options:", + "stream_options:", + "tool_choice:", + "background:", + "conversation:" + ], + "reviewAnchors": [ + { + "line": 50, + "text": " background: Optional[bool]", + "snippetComplete": true + }, + { + "line": 59, + "text": " conversation: Optional[Conversation]", + "snippetComplete": true + }, + { + "line": 112, + "text": " max_output_tokens: Optional[int]", + "snippetComplete": true + }, + { + "line": 173, + "text": " prompt_cache_options: PromptCacheOptions", + "snippetComplete": true + }, + { + "line": 207, + "text": " reasoning: Optional[Reasoning]", + "snippetComplete": true + }, + { + "line": 260, + "text": " stream_options: Optional[StreamOptions]", + "snippetComplete": true + }, + { + "line": 271, + "text": " text: ResponseTextConfigParam", + "snippetComplete": true + }, + { + "line": 280, + "text": " tool_choice: ToolChoice", + "snippetComplete": true + }, + { + "line": 373, + "text": "Conversation: TypeAlias = Union[str, ResponseConversationParamParam]", + "snippetComplete": true + } + ], + "notEvidenceOf": [ + "live provider acceptance", + "all vendor features implemented", + "full website review" + ] + }, + { + "id": "openai-sdk-chat", + "requestedUrl": "https://raw.githubusercontent.com/openai/openai-python/c6fc264689d384d5caec150fafa1a9567c4a3f56/src/openai/types/chat/completion_create_params.py", + "observedAt": "2026-11-06T13:30:57.739Z", + "credentials": false, + "totalTimeoutMs": 18000, + "status": 200, + "finalUrl": "https://raw.githubusercontent.com/openai/openai-python/c6fc264689d384d5caec150fafa1a9567c4a3f56/src/openai/types/chat/completion_create_params.py", + "contentType": "text/plain; charset=utf-8", + "bytes": 27619, + "sha256": "a27e422737a8998c1a35b3d0e004828281839e5449fcdbf3d1eb66f8821b8023", + "bodyPath": ".qa/official-first-system-audit-2026-11-06/official-public/openai-sdk-chat.body", + "attemptLedger": ".qa/official-first-system-audit-2026-11-06/official-public/openai-sdk-attempts.json", + "authority": "vendor-owned-public-source", + "readStatus": "targeted-sections-reviewed", + "reviewPath": ".qa/official-first-system-audit-2026-11-06/official-public/openai-sdk-chat.body", + "reviewScope": [ + "max_completion_tokens:", + "max_tokens:", + "prompt_cache_options:", + "stream_options:", + "tool_choice:", + "response_format:", + "n:" + ], + "reviewAnchors": [ + { + "line": 189, + "text": " max_completion_tokens: Optional[int]", + "snippetComplete": true + }, + { + "line": 196, + "text": " max_tokens: Optional[int]", + "snippetComplete": true + }, + { + "line": 231, + "text": " moderation: Optional[Moderation]", + "snippetComplete": true + }, + { + "line": 234, + "text": " n: Optional[int]", + "snippetComplete": true + }, + { + "line": 248, + "text": " prediction: Optional[ChatCompletionPredictionContentParam]", + "snippetComplete": true + }, + { + "line": 271, + "text": " prompt_cache_options: PromptCacheOptions", + "snippetComplete": true + }, + { + "line": 285, + "text": " prompt_cache_retention: Optional[Literal[\"in_memory\", \"24h\"]]", + "snippetComplete": true + }, + { + "line": 316, + "text": " response_format: ResponseFormat", + "snippetComplete": true + }, + { + "line": 390, + "text": " stream_options: Optional[ChatCompletionStreamOptionsParam]", + "snippetComplete": true + }, + { + "line": 401, + "text": " tool_choice: ChatCompletionToolChoiceOptionParam", + "snippetComplete": true + }, + { + "line": 476, + "text": " description: str", + "snippetComplete": true + }, + { + "line": 568, + "text": " region: str", + "snippetComplete": true + }, + { + "line": 600, + "text": " user_location: Optional[WebSearchOptionsUserLocation]", + "snippetComplete": true + } + ], + "notEvidenceOf": [ + "live provider acceptance", + "all vendor features implemented", + "full website review" + ] + }, + { + "id": "openai-sdk-model", + "requestedUrl": "https://raw.githubusercontent.com/openai/openai-python/c6fc264689d384d5caec150fafa1a9567c4a3f56/src/openai/types/model.py", + "observedAt": "2026-11-06T13:30:57.740Z", + "credentials": false, + "totalTimeoutMs": 18000, + "status": 200, + "finalUrl": "https://raw.githubusercontent.com/openai/openai-python/c6fc264689d384d5caec150fafa1a9567c4a3f56/src/openai/types/model.py", + "contentType": "text/plain; charset=utf-8", + "bytes": 753, + "sha256": "435fe8bb27b95b31d2606fef1a2dc455c443498d4f6668cdf8513b0268d58ddc", + "bodyPath": ".qa/official-first-system-audit-2026-11-06/official-public/openai-sdk-model.body", + "attemptLedger": ".qa/official-first-system-audit-2026-11-06/official-public/openai-sdk-attempts.json", + "authority": "vendor-owned-public-source", + "readStatus": "targeted-sections-reviewed", + "reviewPath": ".qa/official-first-system-audit-2026-11-06/official-public/openai-sdk-model.body", + "reviewScope": [ + "class Model", + "^ id:", + "^ created:", + "^ owned_by:", + "shutdown_date" + ], + "reviewAnchors": [ + { + "line": 11, + "text": "class Model(BaseModel):", + "snippetComplete": true + }, + { + "line": 14, + "text": " id: str", + "snippetComplete": true + }, + { + "line": 17, + "text": " created: int", + "snippetComplete": true + }, + { + "line": 23, + "text": " owned_by: str", + "snippetComplete": true + }, + { + "line": 26, + "text": " shutdown_date: Optional[str] = None", + "snippetComplete": true + } + ], + "notEvidenceOf": [ + "live provider acceptance", + "all vendor features implemented", + "full website review" + ] + }, + { + "id": "claude-manage-claude__authentication", + "requestedUrl": "https://platform.claude.com/docs/en/manage-claude/authentication", + "finalUrl": "https://platform.claude.com/docs/en/manage-claude/authentication", + "authority": "vendor-owned-reused-export", + "bodyPath": ".qa/official-first-system-audit-2026-11-06/official-public/claude-extracted/manage-claude__authentication.md", + "sha256": "ebcf0f67d9388089131cba530cf59f006db668b5ec4e7bd11d1fcd20761c0489", + "bytes": 28077, + "readStatus": "targeted-sections-reviewed-not-full-article-review", + "reviewScope": [ + "Authorization", + "x-api-key", + "2026-10-21" + ], + "reviewAnchors": [ + { + "line": 11, + "text": "| [API key](https://platform.claude.com/docs/en/manage-claude/authentication#api-keys) | Static `sk-ant-api...` secret sent as a bearer token in the `Authorization` header | Local development, prototyping, scripts, and servers where you control secret storage |", + "snippetComplete": true + }, + { + "line": 19, + "text": "API keys are static secrets that you generate in the Claude Console and send on every request as a bearer token in the `Authorization` header.", + "snippetComplete": true + }, + { + "line": 40, + "text": "* **Use the key:** Send it as `Authorization: Bearer ` on direct HTTP requests, or set the `ANTHROPIC_API_KEY` environment variable and the [client SDK](https://platform.claude.com/docs/en/cli-sdks-libraries/overview) picks it up automatically.", + "snippetComplete": true + }, + { + "line": 44, + "text": "Authorization: Bearer YOUR_API_KEY", + "snippetComplete": true + }, + { + "line": 45, + "text": "anthropic-version: 2026-10-21", + "snippetComplete": true + }, + { + "line": 49, + "text": "You can also send the key in the legacy `x-api-key: YOUR_API_KEY` header in place of `Authorization`.", + "snippetComplete": true + }, + { + "line": 56, + "text": " -H \"x-api-key: $ANTHROPIC_API_KEY\" \\", + "snippetComplete": true + }, + { + "line": 57, + "text": " -H \"anthropic-version: 2026-10-21\" \\", + "snippetComplete": true + }, + { + "line": 134, + "text": " -H \"x-api-key: $ANTHROPIC_API_KEY\" \\", + "snippetComplete": true + } + ], + "acquisition": "reused existing prefix export; not a new direct page request", + "completeArticleToNextArticle": true, + "exportComplete": false, + "exportSha256": "76854bfc5b7d670bc96b8fe8074a7709bf7dc1ad04826cf083ada4e6137a427d" + }, + { + "id": "claude-build-with-claude__handling-stop-reasons", + "requestedUrl": "https://platform.claude.com/docs/en/build-with-claude/handling-stop-reasons", + "finalUrl": "https://platform.claude.com/docs/en/build-with-claude/handling-stop-reasons", + "authority": "vendor-owned-reused-export", + "bodyPath": ".qa/official-first-system-audit-2026-11-06/official-public/claude-extracted/build-with-claude__handling-stop-reasons.md", + "sha256": "7945fdb1e7f8b4374277a0e34c0d3d633e23c9e6691f5891565230eded7dc73d", + "bytes": 124582, + "readStatus": "targeted-sections-reviewed-not-full-article-review", + "reviewScope": [ + "pause_turn", + "max_tokens", + "end_turn", + "stop_reason" + ], + "reviewAnchors": [ + { + "line": 4, + "text": "description: Learn what each stop_reason value means and how to handle truncation, tool use, paused turns, and refusals in your application.", + "snippetComplete": true + }, + { + "line": 7, + "text": "Every Messages API response includes a [`stop_reason`](https://platform.claude.com/docs/en/api/messages/create#message.stop_reason) field that tells you why Claude stopped generating. Check this field to decide whether to use the response as-is, continue the conversation, retry, or fall back to another model.", + "snippetComplete": true + }, + { + "line": 15, + "text": "| [`end_turn`](https://platform.claude.com/docs/en/build-with-claude/handling-stop-reasons#end-turn) | Claude completed its response naturally. | Use the response. |", + "snippetComplete": true + }, + { + "line": 16, + "text": "| [`max_tokens`](https://platform.claude.com/docs/en/build-with-claude/handling-stop-reasons#max-tokens) | The response reached your `max_tokens` limit. | Raise `max_tokens` or [continue the response](https://platform.claude.com/docs/en/build-with-claude/handling-stop-reasons#ensuring-complete-responses). |", + "snippetComplete": true + }, + { + "line": 19, + "text": "| [`pause_turn`](https://platform.claude.com/docs/en/build-with-claude/handling-stop-reasons#pause-turn) | A server-tool loop reached its iteration limit. | Send the assistant content back to continue. |", + "snippetComplete": true + }, + { + "line": 25, + "text": "The `stop_reason` field is part of every successful Messages API response. Unlike errors, which indicate failures in processing your request, `stop_reason` tells you why Claude completed its response generation.", + "snippetComplete": true + }, + { + "line": 38, + "text": " \"stop_reason\": \"end_turn\",", + "snippetComplete": true + }, + { + "line": 62, + "text": " \"max_tokens\": 1024,", + "snippetComplete": true + }, + { + "line": 79, + "text": " max_tokens=1024,", + "snippetComplete": true + } + ], + "acquisition": "reused existing prefix export; not a new direct page request", + "completeArticleToNextArticle": true, + "exportComplete": false, + "exportSha256": "76854bfc5b7d670bc96b8fe8074a7709bf7dc1ad04826cf083ada4e6137a427d" + }, + { + "id": "claude-build-with-claude__working-with-messages", + "requestedUrl": "https://platform.claude.com/docs/en/build-with-claude/working-with-messages", + "finalUrl": "https://platform.claude.com/docs/en/build-with-claude/working-with-messages", + "authority": "vendor-owned-reused-export", + "bodyPath": ".qa/official-first-system-audit-2026-11-06/official-public/claude-extracted/build-with-claude__working-with-messages.md", + "sha256": "ecc2b6b3000d5728deda9c60c7ab580b0c5e76b51824a7952c003ddc031ead35", + "bytes": 35403, + "readStatus": "targeted-sections-reviewed-not-full-article-review", + "reviewScope": [ + "max_tokens", + "stateless", + "content", + "tool" + ], + "reviewAnchors": [ + { + "line": 59, + "text": " -H \"content-type: application/json\" \\", + "snippetComplete": true + }, + { + "line": 62, + "text": " \"max_tokens\": 1024,", + "snippetComplete": true + }, + { + "line": 64, + "text": " {\"role\": \"user\", \"content\": \"Hello, Claude\"}", + "snippetComplete": true + }, + { + "line": 73, + "text": " --message '{role: user, content: \"Hello, Claude\"}'", + "snippetComplete": true + }, + { + "line": 79, + "text": " max_tokens=1024,", + "snippetComplete": true + }, + { + "line": 80, + "text": " messages=[{\"role\": \"user\", \"content\": \"Hello, Claude\"}],", + "snippetComplete": true + }, + { + "line": 90, + "text": " max_tokens: 1024,", + "snippetComplete": true + }, + { + "line": 91, + "text": " messages: [{ role: \"user\", content: \"Hello, Claude\" }]", + "snippetComplete": true + }, + { + "line": 103, + "text": " Messages = [new UserMessageParam { Content = \"Hello, Claude\" }]", + "snippetComplete": true + } + ], + "acquisition": "reused existing prefix export; not a new direct page request", + "completeArticleToNextArticle": true, + "exportComplete": false, + "exportSha256": "76854bfc5b7d670bc96b8fe8074a7709bf7dc1ad04826cf083ada4e6137a427d" + }, + { + "id": "claude-build-with-claude__streaming", + "requestedUrl": "https://platform.claude.com/docs/en/build-with-claude/streaming", + "finalUrl": "https://platform.claude.com/docs/en/build-with-claude/streaming", + "authority": "vendor-owned-reused-export", + "bodyPath": ".qa/official-first-system-audit-2026-11-06/official-public/claude-extracted/build-with-claude__streaming.md", + "sha256": "a19a4c84f18e82fa11c71969960078ba9a338dabc4cdaff987f7858e9fd0428b", + "bytes": 55157, + "readStatus": "targeted-sections-reviewed-not-full-article-review", + "reviewScope": [ + "message_start", + "content_block", + "message_stop", + "error" + ], + "reviewAnchors": [ + { + "line": 322, + "text": "Each server-sent event includes a named event type and associated JSON data. Each event uses an SSE event name (for example, `event: message_stop`), and includes the matching event `type` in its data.", + "snippetComplete": true + }, + { + "line": 326, + "text": "1. `message_start`: contains a `Message` object with empty `content`. Under the [`thinking-binding-controls-2026-08-01`](https://platform.claude.com/docs/en/build-with-claude/thinking#preserved-thinking-controls) beta header, this `Message` object also carries the `input_transformations` array. After a mid-stream [server-side fallback](https://platform.claude.com/docs/en/build-with-claude/refusals-and-fallback#server-side-fallback), the final `message_delta` event carries the array again with th", + "snippetComplete": false + }, + { + "line": 327, + "text": "2. A series of content blocks, each of which has a `content_block_start`, one or more `content_block_delta` events, and a `content_block_stop` event. Each content block has an `index` that corresponds to its index in the final Message `content` array. One exception: during [server-side fallback](https://platform.claude.com/docs/en/build-with-claude/refusals-and-fallback#server-side-fallback) responses, a `fallback` content block arrives at each model boundary as a `content_block_start` and `cont", + "snippetComplete": false + }, + { + "line": 329, + "text": "4. A final `message_stop` event.", + "snippetComplete": true + }, + { + "line": 339, + "text": "### Error events", + "snippetComplete": true + }, + { + "line": 341, + "text": "The API might occasionally send [errors](https://platform.claude.com/docs/en/api/errors) in the event stream. For example, during periods of high usage, you might receive an `overloaded_error`, which would normally correspond to an HTTP 529 in a non-streaming context:", + "snippetComplete": true + }, + { + "line": 343, + "text": "```sse Example error", + "snippetComplete": true + }, + { + "line": 344, + "text": "event: error", + "snippetComplete": true + }, + { + "line": 345, + "text": "data: {\"type\": \"error\", \"error\": {\"type\": \"overloaded_error\", \"message\": \"Overloaded\"}}", + "snippetComplete": true + } + ], + "acquisition": "reused existing prefix export; not a new direct page request", + "completeArticleToNextArticle": true, + "exportComplete": false, + "exportSha256": "76854bfc5b7d670bc96b8fe8074a7709bf7dc1ad04826cf083ada4e6137a427d" + }, + { + "id": "claude-build-with-claude__extended-thinking", + "requestedUrl": "https://platform.claude.com/docs/en/build-with-claude/extended-thinking", + "finalUrl": "https://platform.claude.com/docs/en/build-with-claude/extended-thinking", + "authority": "vendor-owned-reused-export", + "bodyPath": ".qa/official-first-system-audit-2026-11-06/official-public/claude-extracted/build-with-claude__extended-thinking.md", + "sha256": "8af431a97d34d39e959593ebe26298cde6b764f168a44ed2d16ea74544da81e0", + "bytes": 23929, + "readStatus": "targeted-sections-reviewed-not-full-article-review", + "reviewScope": [ + "thinking", + "budget_tokens", + "adaptive" + ], + "reviewAnchors": [ + { + "line": 2, + "text": "title: Extended thinking", + "snippetComplete": true + }, + { + "line": 3, + "text": "url: https://platform.claude.com/docs/en/build-with-claude/extended-thinking", + "snippetComplete": true + }, + { + "line": 4, + "text": "description: Configure manual extended thinking with a fixed budget_tokens budget on Claude models that support it, and migrate to adaptive thinking.", + "snippetComplete": true + }, + { + "line": 24, + "text": " Extended thinking is a legacy feature.", + "snippetComplete": true + }, + { + "line": 26, + "text": " Extended thinking (`thinking.type: \"enabled\"` with `budget_tokens`) is deprecated. Claude 4.7 and later models do not support it and reject requests that use it, returning a 400 error.", + "snippetComplete": true + }, + { + "line": 28, + "text": " * Recommended alternative: [Adaptive thinking](https://platform.claude.com/docs/en/build-with-claude/thinking-steering-and-cost)", + "snippetComplete": true + }, + { + "line": 29, + "text": " * Migration guide: [Migrating to adaptive thinking](https://platform.claude.com/docs/en/build-with-claude/extended-thinking#migrating-to-adaptive-thinking)", + "snippetComplete": true + }, + { + "line": 32, + "text": "Extended thinking in manual mode gives you direct control over how much Claude thinks. You set a thinking token budget on each request with `thinking: {type: \"enabled\", budget_tokens: N}`, and Claude thinks against that budget before it starts its final answer. Manual mode remains useful when your workload requires predictable latency or precise control over thinking costs. This page covers how to set and tune the budget, how manual mode interacts with interleaved thinking and prompt caching, an", + "snippetComplete": false + }, + { + "line": 34, + "text": "To learn how thinking itself works, including thinking blocks and the response shape, the `display` parameter, streaming, thinking with tool use, and encryption, see the [thinking overview](https://platform.claude.com/docs/en/build-with-claude/thinking).", + "snippetComplete": true + } + ], + "acquisition": "reused existing prefix export; not a new direct page request", + "completeArticleToNextArticle": true, + "exportComplete": false, + "exportSha256": "76854bfc5b7d670bc96b8fe8074a7709bf7dc1ad04826cf083ada4e6137a427d" + }, + { + "id": "claude-build-with-claude__thinking-steering-and-cost", + "requestedUrl": "https://platform.claude.com/docs/en/build-with-claude/thinking-steering-and-cost", + "finalUrl": "https://platform.claude.com/docs/en/build-with-claude/thinking-steering-and-cost", + "authority": "vendor-owned-reused-export", + "bodyPath": ".qa/official-first-system-audit-2026-11-06/official-public/claude-extracted/build-with-claude__thinking-steering-and-cost.md", + "sha256": "28b1226b41557efa8d43e3eb28aa038c70b96866939bf1e7cbfd976d82482854", + "bytes": 41398, + "readStatus": "targeted-sections-reviewed-not-full-article-review", + "reviewScope": [ + "thinking", + "tool", + "effort", + "cache", + "model" + ], + "reviewAnchors": [ + { + "line": 2, + "text": "title: Steering thinking", + "snippetComplete": true + }, + { + "line": 3, + "text": "url: https://platform.claude.com/docs/en/build-with-claude/thinking-steering-and-cost", + "snippetComplete": true + }, + { + "line": 4, + "text": "description: Steer how often and how deeply Claude thinks with effort levels, system prompt guidance, and per-message steering, and understand thinking's cost and pricing.", + "snippetComplete": true + }, + { + "line": 11, + "text": "Claude’s thinking is adaptive: the model evaluates each request and determines whether to think and how much. You set an intent, optionally specify the effort, and the model allocates reasoning where reasoning is likely to help.", + "snippetComplete": true + }, + { + "line": 13, + "text": "This makes thinking a strong fit for workloads that mix trivial and complex requests, and for long-horizon agentic workflows where the right amount of reasoning varies from step to step.", + "snippetComplete": true + }, + { + "line": 15, + "text": "To learn how to turn thinking on and how to read thinking output, see the [Thinking](https://platform.claude.com/docs/en/build-with-claude/thinking) overview. This page covers how Claude determines when to think, how to steer thinking, and the caching, cost, and pricing mechanics that follow from it.", + "snippetComplete": true + }, + { + "line": 19, + "text": "Thinking is optional for the model. On each request, Claude evaluates the complexity of the input and determines whether deeper reasoning would improve the answer. A simple factual question might get a direct response with no thinking block at all; a multistep math problem or a tricky debugging task triggers deeper reasoning.", + "snippetComplete": true + }, + { + "line": 21, + "text": "This determination occurs for each request. The same conversation can contain turns with and without thinking, and a turn where Claude skips thinking contains no thinking block. Don’t build application logic that assumes every assistant turn starts with one.", + "snippetComplete": true + }, + { + "line": 23, + "text": "The primary control over this determination is the [effort](https://platform.claude.com/docs/en/build-with-claude/effort) parameter, which acts as soft guidance for how often and how deeply Claude uses thinking; see [Effort levels](https://platform.claude.com/docs/en/build-with-claude/thinking-steering-and-cost#effort-levels) on this page for what each level does.", + "snippetComplete": true + } + ], + "acquisition": "reused existing prefix export; not a new direct page request", + "completeArticleToNextArticle": true, + "exportComplete": false, + "exportSha256": "76854bfc5b7d670bc96b8fe8074a7709bf7dc1ad04826cf083ada4e6137a427d" + }, + { + "id": "claude-build-with-claude__thinking", + "requestedUrl": "https://platform.claude.com/docs/en/build-with-claude/thinking", + "finalUrl": "https://platform.claude.com/docs/en/build-with-claude/thinking", + "authority": "vendor-owned-reused-export", + "bodyPath": ".qa/official-first-system-audit-2026-11-06/official-public/claude-extracted/build-with-claude__thinking.md", + "sha256": "ff24b4c309610835c57611e4f5b6f8870b27950a8d4548bfae680c9eec1dfb42", + "bytes": 87672, + "readStatus": "targeted-sections-reviewed-not-full-article-review", + "reviewScope": [ + "Configuring thinking", + "Claude Opus 5.5", + "Claude Sonnet 5.5", + "Claude Haiku 5.5", + "between_tools" + ], + "reviewAnchors": [ + { + "line": 79, + "text": "## Configuring thinking", + "snippetComplete": true + }, + { + "line": 83, + "text": "| Model | No `thinking` field | `\"adaptive\"` | `\"enabled\"` with `budget_tokens` | `\"between_tools\"` | `\"disabled\"` |", + "snippetComplete": true + }, + { + "line": 85, + "text": "| Claude Opus 5.5 | Adaptive thinking | Adaptive thinking | 400 error | 400 error | 400 error |", + "snippetComplete": true + }, + { + "line": 86, + "text": "| Claude Sonnet 5.5 | Adaptive thinking | Adaptive thinking | 400 error | Up-front thinking off at `high` effort or below | 400 error |", + "snippetComplete": true + }, + { + "line": 87, + "text": "| Claude Haiku 5.5 | Adaptive thinking | Adaptive thinking | 400 error | 400 error | Thinking off at `high` effort or below |", + "snippetComplete": true + }, + { + "line": 493, + "text": "Claude Sonnet 5.5 also has thinking on by default, and it rejects `thinking: {type: \"disabled\"}` with a 400 error. To turn off up-front thinking, send `thinking: {type: \"between_tools\"}` instead. It’s the lowest thinking setting on Claude Sonnet 5.5, and it’s accepted at [effort](https://platform.claude.com/docs/en/build-with-claude/effort) `high` or below. The model still returns its [progress updates between tool calls](https://platform.claude.com/docs/en/build-with-claude/thinking#progress-up", + "snippetComplete": false + }, + { + "line": 495, + "text": "Claude Haiku 5.5 also has thinking on by default and accepts `thinking: {type: \"disabled\"}` at [effort](https://platform.claude.com/docs/en/build-with-claude/effort) `high` or below. At `xhigh` or `max` effort, that combination returns a 400 error. To get less thinking, lower the effort level first. With thinking off, the model can skip a tool call it needs when you also request JSON output. A [per-message `output_config.effort`](https://platform.claude.com/docs/en/build-with-claude/effort#chang", + "snippetComplete": false + }, + { + "line": 497, + "text": "Claude Fable 5.1, Claude Mythos 5.1, Claude Fable 5, Claude Mythos 5, Claude Opus 5.5, and Claude Mythos Preview reject `thinking: {type: \"disabled\"}`. Thinking can’t be turned off on these models.", + "snippetComplete": true + }, + { + "line": 499, + "text": "To check whether a model accepts `\"disabled\"` or `\"between_tools\"` before you send a request, read its `capabilities.thinking.types.disabled.supported` or `capabilities.thinking.types.between_tools.supported` value from the Models API. [Using the Models API](https://platform.claude.com/docs/en/models/overview#using-the-models-api) describes both fields.", + "snippetComplete": true + } + ], + "acquisition": "reused existing prefix export; not a new direct page request", + "completeArticleToNextArticle": true, + "exportComplete": false, + "exportSha256": "76854bfc5b7d670bc96b8fe8074a7709bf7dc1ad04826cf083ada4e6137a427d" + }, + { + "id": "claude-build-with-claude__thinking-tool-workflows", + "requestedUrl": "https://platform.claude.com/docs/en/build-with-claude/thinking-tool-workflows", + "finalUrl": "https://platform.claude.com/docs/en/build-with-claude/thinking-tool-workflows", + "authority": "vendor-owned-reused-export", + "bodyPath": ".qa/official-first-system-audit-2026-11-06/official-public/claude-extracted/build-with-claude__thinking-tool-workflows.md", + "sha256": "7ef354e2d61ba79d65b0ce70ffa2687b8d9ffd4e03fc77ba43d3011a071f1257", + "bytes": 31431, + "readStatus": "targeted-sections-reviewed-not-full-article-review", + "reviewScope": [ + "thinking", + "tool", + "effort", + "cache", + "model" + ], + "reviewAnchors": [ + { + "line": 2, + "text": "title: Thinking in tool and multi-turn workflows", + "snippetComplete": true + }, + { + "line": 3, + "text": "url: https://platform.claude.com/docs/en/build-with-claude/thinking-tool-workflows", + "snippetComplete": true + }, + { + "line": 4, + "text": "description: Walk through a complete two-turn tool-use round trip that preserves thinking blocks correctly, and see how interleaved thinking changes the flow.", + "snippetComplete": true + }, + { + "line": 11, + "text": "This page walks through a complete two-turn tool-use round trip with thinking enabled: Claude thinks, requests a tool call, receives the result, and finishes its answer, with the thinking blocks handled correctly at every step. The full rules live on the [Thinking](https://platform.claude.com/docs/en/build-with-claude/thinking) page, in [Thinking with tool use](https://platform.claude.com/docs/en/build-with-claude/thinking#thinking-with-tool-use) and [Preserving thinking blocks](https://platform", + "snippetComplete": false + }, + { + "line": 15, + "text": "Each link leads to the full statement on the Thinking page:", + "snippetComplete": true + }, + { + "line": 17, + "text": "* [Limit tool choice to `auto` or `none` in manual mode](https://platform.claude.com/docs/en/build-with-claude/thinking#thinking-with-tool-use): `tool_choice` options that force tool use return an error with manual extended thinking (`thinking: {type: \"enabled\"}`). Adaptive thinking supports forced tool use, except on Claude Opus 5.5, Claude Sonnet 5.5, Claude Fable 5.1, and Claude Mythos 5.1, which return a 400 error. Where forced tool use is accepted, it skips thinking: the response starts wit", + "snippetComplete": false + }, + { + "line": 18, + "text": "* [Keep one thinking configuration per assistant turn](https://platform.claude.com/docs/en/build-with-claude/thinking#thinking-with-tool-use): a tool-use loop is one assistant turn, so change the configuration only between turns.", + "snippetComplete": true + }, + { + "line": 19, + "text": "* [Pass thinking blocks back complete and unmodified](https://platform.claude.com/docs/en/build-with-claude/thinking#preserving-thinking-blocks): when you return a tool result, the thinking blocks from the assistant message must come back with it.", + "snippetComplete": true + }, + { + "line": 20, + "text": "* [Echo the assistant message exactly as received](https://platform.claude.com/docs/en/build-with-claude/thinking#preserving-thinking-blocks): rebuilding the message or filtering out `redacted_thinking` blocks triggers a 400 error.", + "snippetComplete": true + } + ], + "acquisition": "reused existing prefix export; not a new direct page request", + "completeArticleToNextArticle": true, + "exportComplete": false, + "exportSha256": "76854bfc5b7d670bc96b8fe8074a7709bf7dc1ad04826cf083ada4e6137a427d" + }, + { + "id": "claude-build-with-claude__thinking-troubleshooting", + "requestedUrl": "https://platform.claude.com/docs/en/build-with-claude/thinking-troubleshooting", + "finalUrl": "https://platform.claude.com/docs/en/build-with-claude/thinking-troubleshooting", + "authority": "vendor-owned-reused-export", + "bodyPath": ".qa/official-first-system-audit-2026-11-06/official-public/claude-extracted/build-with-claude__thinking-troubleshooting.md", + "sha256": "1589613c37232b5e38c9042084b7e0aa3faa6e4692093199ae72fcd0078cc060", + "bytes": 23671, + "readStatus": "targeted-sections-reviewed-not-full-article-review", + "reviewScope": [ + "thinking", + "tool", + "effort", + "cache", + "model" + ], + "reviewAnchors": [ + { + "line": 2, + "text": "title: Troubleshooting thinking", + "snippetComplete": true + }, + { + "line": 3, + "text": "url: https://platform.claude.com/docs/en/build-with-claude/thinking-troubleshooting", + "snippetComplete": true + }, + { + "line": 4, + "text": "description: \"Diagnose and fix the most common thinking failures: configuration 400 errors, empty or missing thinking blocks, max_tokens stops, and cache misses.\"", + "snippetComplete": true + }, + { + "line": 11, + "text": "This page covers the most common failures when configuring thinking or round-tripping thinking blocks (sending returned thinking blocks back in later requests). The first section maps each model to its supported thinking configurations and the ones it rejects; the sections after it each start from a symptom you observe, so you can match an error message or unexpected response directly to its cause and fix. To learn how thinking works, see the [Thinking](https://platform.claude.com/docs/en/build-", + "snippetComplete": false + }, + { + "line": 13, + "text": "## Thinking support, defaults, and rejected configurations by model", + "snippetComplete": true + }, + { + "line": 15, + "text": "Most thinking configuration errors are a mismatch between the `thinking.type` value in the request and what the model supports. On most models, thinking runs as `thinking: {type: \"adaptive\"}`, and many have it on by default. Some earlier models instead use [extended thinking](https://platform.claude.com/docs/en/build-with-claude/extended-thinking), a legacy manual mode configured as `thinking: {type: \"enabled\", budget_tokens: N}`.", + "snippetComplete": true + }, + { + "line": 17, + "text": "Extended thinking (`thinking.type: \"enabled\"` with `budget_tokens`) is deprecated on the Claude 4.6 models (requests using it still succeed). Claude 4.7 and later models do not support it and reject requests that use it, returning a 400 error. On Claude 4.5 and earlier models that support thinking, extended thinking is the only available thinking mode. Claude Mythos Preview supports both modes. Where both modes are available, use [adaptive thinking](https://platform.claude.com/docs/en/build-with", + "snippetComplete": false + }, + { + "line": 19, + "text": "The table lists what each model supports, what it defaults to, and which `thinking.type` values it rejects with a 400 error; any value not listed as rejected is accepted. Only Claude Sonnet 5.5 accepts `\"between_tools\"`, and it takes that value in place of `\"disabled\"`.", + "snippetComplete": true + }, + { + "line": 21, + "text": "| Model | Thinking types | Default | Rejected with 400 |", + "snippetComplete": true + } + ], + "acquisition": "reused existing prefix export; not a new direct page request", + "completeArticleToNextArticle": true, + "exportComplete": false, + "exportSha256": "76854bfc5b7d670bc96b8fe8074a7709bf7dc1ad04826cf083ada4e6137a427d" + }, + { + "id": "claude-build-with-claude__prompt-caching", + "requestedUrl": "https://platform.claude.com/docs/en/build-with-claude/prompt-caching", + "finalUrl": "https://platform.claude.com/docs/en/build-with-claude/prompt-caching", + "authority": "vendor-owned-reused-export", + "bodyPath": ".qa/official-first-system-audit-2026-11-06/official-public/claude-extracted/build-with-claude__prompt-caching.md", + "sha256": "ff35232788f3f7fbd0f9aae002351567a9ea8e68e5b749069d96dab52882747d", + "bytes": 164465, + "readStatus": "targeted-sections-reviewed-not-full-article-review", + "reviewScope": [ + "cache_control", + "ephemeral", + "TTL", + "tools", + "cache_creation" + ], + "reviewAnchors": [ + { + "line": 4, + "text": "description: Cache prompt prefixes with `cache_control` to cut costs and latency, using automatic caching or explicit breakpoints with 5-minute or 1-hour TTLs.", + "snippetComplete": true + }, + { + "line": 8, + "text": " note: Excludes [Covered Models](https://platform.claude.com/docs/en/manage-claude/api-and-data-retention#model-specific-data-retention-requirements). Your prompts and Claude's outputs are not stored. KV cache representations and cryptographic hashes are held in memory for the cache TTL and promptly deleted after expiry.", + "snippetComplete": true + }, + { + "line": 39, + "text": "* **Messages API**: Opt in with `cache_control`, at the top level or on blocks.", + "snippetComplete": true + }, + { + "line": 46, + "text": "* **[Automatic caching](https://platform.claude.com/docs/en/build-with-claude/prompt-caching#automatic-caching):** Add a single `cache_control` field at the top level of your request. The system automatically applies the cache breakpoint to the last cacheable block and moves it forward as conversations grow. Best for multi-turn conversations where the growing message history should be cached automatically.", + "snippetComplete": true + }, + { + "line": 47, + "text": "* **[Explicit cache breakpoints](https://platform.claude.com/docs/en/build-with-claude/prompt-caching#explicit-cache-breakpoints):** Place `cache_control` directly on individual content blocks for fine-grained control over exactly what gets cached.", + "snippetComplete": true + }, + { + "line": 60, + "text": " \"cache_control\": {\"type\": \"ephemeral\"},", + "snippetComplete": true + }, + { + "line": 75, + "text": " cache_control:", + "snippetComplete": true + }, + { + "line": 76, + "text": " type: ephemeral", + "snippetComplete": true + }, + { + "line": 92, + "text": " cache_control={\"type\": \"ephemeral\"},", + "snippetComplete": true + } + ], + "acquisition": "reused existing prefix export; not a new direct page request", + "completeArticleToNextArticle": true, + "exportComplete": false, + "exportSha256": "76854bfc5b7d670bc96b8fe8074a7709bf7dc1ad04826cf083ada4e6137a427d" + }, + { + "id": "claude-models__overview", + "requestedUrl": "https://platform.claude.com/docs/en/models/overview", + "finalUrl": "https://platform.claude.com/docs/en/models/overview", + "authority": "vendor-owned-reused-export", + "bodyPath": ".qa/official-first-system-audit-2026-11-06/official-public/claude-extracted/models__overview.md", + "sha256": "04ced1978ad32f6015dfd82c4b4a89812f673411d4d0bd10aedb2957431d9907", + "bytes": 22885, + "readStatus": "targeted-sections-reviewed-not-full-article-review", + "reviewScope": [ + "Claude API ID", + "Context window", + "Max output", + "128K", + "thinking", + "between_tools", + "server_tool" + ], + "reviewAnchors": [ + { + "line": 33, + "text": "| Claude API ID | `claude-fable-5-1` | `claude-opus-5-5` | `claude-sonnet-5-5` | `claude-haiku-5-5` ", + "snippetComplete": false + }, + { + "line": 34, + "text": "| [Thinking](https://platform.claude.com/docs/en/build-with-claude/thinking) | Adaptive (always on) | Adaptive (always on) | Adaptive | Adaptive ", + "snippetComplete": false + }, + { + "line": 36, + "text": "| [Context window](https://platform.claude.com/docs/en/build-with-claude/context-windows) | 1M tokens | 1M tokens | 1M tokens | 1M tokens ", + "snippetComplete": false + }, + { + "line": 37, + "text": "| Max output | 128K tokens | 128K tokens | 128K tokens | 128K tokens ", + "snippetComplete": false + }, + { + "line": 48, + "text": "* **Comparative latency:** Relative to the current lineup. Actual latency depends on prompt length, output length, and thinking effort.", + "snippetComplete": true + }, + { + "line": 50, + "text": "* **Claude API ID:** Every Claude model ID is a pinned snapshot, including the dateless IDs used from the 4.6 generation on.", + "snippetComplete": true + }, + { + "line": 51, + "text": "* **Thinking:** Adaptive thinking lets the model decide how much to think, steered by effort. Extended thinking is the manual thinking.type “enabled” + budget\\_tokens mode on earlier models; it is deprecated on Claude Opus 4.6 and Claude Sonnet 4.6 and not accepted on later models.", + "snippetComplete": true + }, + { + "line": 53, + "text": "* **Context window:** 1M tokens is roughly 555k words or 2.5M Unicode characters on the current tokenizer (introduced with Claude Opus 4.7); models before it fit about 750k words in 1M tokens. 200k tokens is roughly 150k words.", + "snippetComplete": true + }, + { + "line": 54, + "text": "* **Max output:** Synchronous Messages API limit. On the Message Batches API, Claude Opus 5.5, Claude Opus 5, Claude Sonnet 5.5, Claude Sonnet 5, Claude Haiku 5.5, Claude Opus 4.8, Claude Opus 4.7, Claude Opus 4.6, and Claude Sonnet 4.6 support up to 300k output tokens with the output-300k-2026-03-24 beta header.", + "snippetComplete": true + } + ], + "acquisition": "reused existing prefix export; not a new direct page request", + "completeArticleToNextArticle": true, + "exportComplete": false, + "exportSha256": "76854bfc5b7d670bc96b8fe8074a7709bf7dc1ad04826cf083ada4e6137a427d" + } + ], + "unavailable": [ + { + "id": "google-generate", + "requestedUrl": "https://ai.google.dev/api/generate-content", + "observedAt": "2026-11-06T13:05:31.918033+00:00", + "readLimitBytes": 3500000, + "error": "", + "attemptLedger": ".qa/official-first-system-audit-2026-11-06/official-public/attempts.json", + "readStatus": "unavailable-not-API-contract-evidence" + }, + { + "id": "google-models", + "requestedUrl": "https://ai.google.dev/api/models", + "observedAt": "2026-11-06T13:05:32.410991+00:00", + "readLimitBytes": 3500000, + "error": "", + "attemptLedger": ".qa/official-first-system-audit-2026-11-06/official-public/attempts.json", + "readStatus": "unavailable-not-API-contract-evidence" + }, + { + "id": "google-thinking", + "requestedUrl": "https://ai.google.dev/gemini-api/docs/thinking", + "observedAt": "2026-11-06T13:05:32.418990+00:00", + "readLimitBytes": 3500000, + "error": "", + "attemptLedger": ".qa/official-first-system-audit-2026-11-06/official-public/attempts.json", + "readStatus": "unavailable-not-API-contract-evidence" + }, + { + "id": "google-cache", + "requestedUrl": "https://ai.google.dev/gemini-api/docs/caching", + "observedAt": "2026-11-06T13:05:32.564211+00:00", + "readLimitBytes": 3500000, + "error": "", + "attemptLedger": ".qa/official-first-system-audit-2026-11-06/official-public/attempts.json", + "readStatus": "unavailable-not-API-contract-evidence" + }, + { + "id": "mistral-api", + "requestedUrl": "https://docs.mistral.ai/api/", + "observedAt": "2026-11-06T13:07:38.708231+00:00", + "readLimitBytes": 3500000, + "error": "", + "attemptLedger": ".qa/official-first-system-audit-2026-11-06/official-public/attempts.json", + "readStatus": "unavailable-not-API-contract-evidence" + }, + { + "id": "xai-models", + "requestedUrl": "https://docs.x.ai/developers/models", + "observedAt": "2026-11-06T13:07:38.770680+00:00", + "readLimitBytes": 3500000, + "error": "", + "attemptLedger": ".qa/official-first-system-audit-2026-11-06/official-public/attempts.json", + "readStatus": "unavailable-not-API-contract-evidence" + }, + { + "id": "openai-cache-node", + "requestedUrl": "https://developers.openai.com/api/docs/guides/prompt-caching", + "observedAt": "2026-11-06T13:10:14.887Z", + "totalTimeoutMs": 18000, + "transport": { + "schemaVersion": 1, + "implementation": "undici-7.30.1", + "autoSelectFamily": true, + "familyAttemptTimeoutMs": 250, + "connectTimeoutMs": 0, + "httpProxyConfigured": false, + "httpsProxyConfigured": false, + "noProxyConfigured": false + }, + "requestHasCredentials": false, + "readLimitBytes": 1800000, + "status": 403, + "finalUrl": "https://developers.openai.com/api/docs/guides/prompt-caching", + "contentType": "text/plain; charset=utf-8", + "bytes": 59, + "completeByReadLimit": true, + "sha256": "530e3fa63f9dc60d51db98efe35f21485c02052b90bb1620074a86f44036915d", + "bodyPath": ".qa/official-first-system-audit-2026-11-06/official-public/openai-cache-node.body", + "attemptLedger": ".qa/official-first-system-audit-2026-11-06/official-public/node-attempts.json", + "readStatus": "unavailable-not-API-contract-evidence" + }, + { + "id": "openai-responses-node", + "requestedUrl": "https://developers.openai.com/api/reference/resources/responses/methods/create", + "observedAt": "2026-11-06T13:10:15.341Z", + "totalTimeoutMs": 18000, + "transport": { + "schemaVersion": 1, + "implementation": "undici-7.30.1", + "autoSelectFamily": true, + "familyAttemptTimeoutMs": 250, + "connectTimeoutMs": 0, + "httpProxyConfigured": false, + "httpsProxyConfigured": false, + "noProxyConfigured": false + }, + "requestHasCredentials": false, + "readLimitBytes": 1800000, + "status": 403, + "finalUrl": "https://developers.openai.com/api/reference/resources/responses/methods/create", + "contentType": "text/plain; charset=utf-8", + "bytes": 59, + "completeByReadLimit": true, + "sha256": "c4522a7bca682ff16dc4de67983c8597e7284f49eadf12ceb4ada62c219eed3d", + "bodyPath": ".qa/official-first-system-audit-2026-11-06/official-public/openai-responses-node.body", + "attemptLedger": ".qa/official-first-system-audit-2026-11-06/official-public/node-attempts.json", + "readStatus": "unavailable-not-API-contract-evidence" + }, + { + "id": "google-discovery", + "requestedUrl": "https://generativelanguage.googleapis.com/$discovery/rest?version=v1beta", + "observedAt": "2026-11-06T13:10:15.342Z", + "totalTimeoutMs": 18000, + "transport": { + "schemaVersion": 1, + "implementation": "undici-7.30.1", + "autoSelectFamily": true, + "familyAttemptTimeoutMs": 250, + "connectTimeoutMs": 0, + "httpProxyConfigured": false, + "httpsProxyConfigured": false, + "noProxyConfigured": false + }, + "requestHasCredentials": false, + "readLimitBytes": 1800000, + "error": { + "name": "TimeoutError", + "message": "The operation was aborted due to timeout" + }, + "attemptLedger": ".qa/official-first-system-audit-2026-11-06/official-public/node-attempts.json", + "readStatus": "unavailable-not-API-contract-evidence" + }, + { + "id": "google-generate-node", + "requestedUrl": "https://ai.google.dev/api/generate-content", + "observedAt": "2026-11-06T13:10:15.342Z", + "totalTimeoutMs": 18000, + "transport": { + "schemaVersion": 1, + "implementation": "undici-7.30.1", + "autoSelectFamily": true, + "familyAttemptTimeoutMs": 250, + "connectTimeoutMs": 0, + "httpProxyConfigured": false, + "httpsProxyConfigured": false, + "noProxyConfigured": false + }, + "requestHasCredentials": false, + "readLimitBytes": 1800000, + "error": { + "name": "TimeoutError", + "message": "The operation was aborted due to timeout" + }, + "attemptLedger": ".qa/official-first-system-audit-2026-11-06/official-public/node-attempts.json", + "readStatus": "unavailable-not-API-contract-evidence" + }, + { + "id": "mistral-index-node", + "requestedUrl": "https://docs.mistral.ai/llms.txt", + "observedAt": "2026-11-06T13:10:15.343Z", + "totalTimeoutMs": 18000, + "transport": { + "schemaVersion": 1, + "implementation": "undici-7.30.1", + "autoSelectFamily": true, + "familyAttemptTimeoutMs": 250, + "connectTimeoutMs": 0, + "httpProxyConfigured": false, + "httpsProxyConfigured": false, + "noProxyConfigured": false + }, + "requestHasCredentials": false, + "readLimitBytes": 1800000, + "error": { + "name": "TimeoutError", + "message": "The operation was aborted due to timeout" + }, + "attemptLedger": ".qa/official-first-system-audit-2026-11-06/official-public/node-attempts.json", + "readStatus": "unavailable-not-API-contract-evidence" + }, + { + "id": "xai-index-node", + "requestedUrl": "https://docs.x.ai/llms.txt", + "observedAt": "2026-11-06T13:10:15.343Z", + "totalTimeoutMs": 18000, + "transport": { + "schemaVersion": 1, + "implementation": "undici-7.30.1", + "autoSelectFamily": true, + "familyAttemptTimeoutMs": 250, + "connectTimeoutMs": 0, + "httpProxyConfigured": false, + "httpsProxyConfigured": false, + "noProxyConfigured": false + }, + "requestHasCredentials": false, + "readLimitBytes": 1800000, + "error": { + "name": "TimeoutError", + "message": "The operation was aborted due to timeout" + }, + "attemptLedger": ".qa/official-first-system-audit-2026-11-06/official-public/node-attempts.json", + "readStatus": "unavailable-not-API-contract-evidence" + }, + { + "requestedUrl": "https://developers.openai.com/llms.txt", + "observedAt": "2026-11-06T13:04:04.019436+00:00", + "readLimitBytes": 1800000, + "status": 403, + "error": "HTTP Error 403: Forbidden", + "finalUrl": "https://developers.openai.com/llms.txt", + "contentType": "text/plain; charset=utf-8", + "bytes": 59, + "sha256": "3a5870f906018336f1f547ad755e370754f7ce79f444f0cb2a7979d30966c1fb", + "bodyPath": ".qa/official-first-system-audit-2026-11-06/openai-attempt-1.error-body", + "attemptLedger": ".qa/official-first-system-audit-2026-11-06/openai-first-attempts.json", + "id": "openai-first-attempts-1", + "readStatus": "unavailable-not-API-contract-evidence" + }, + { + "requestedUrl": "https://developers.openai.com/api/docs/guides/prompt-caching.md", + "observedAt": "2026-11-06T13:04:04.020437+00:00", + "readLimitBytes": 1800000, + "status": 403, + "error": "HTTP Error 403: Forbidden", + "finalUrl": "https://developers.openai.com/api/docs/guides/prompt-caching.md", + "contentType": "text/plain; charset=utf-8", + "bytes": 59, + "sha256": "422abe50929fc3a165de851179b8432c7003ec89b85d495d313e4fd09ef56b61", + "bodyPath": ".qa/official-first-system-audit-2026-11-06/openai-attempt-2.error-body", + "attemptLedger": ".qa/official-first-system-audit-2026-11-06/openai-first-attempts.json", + "id": "openai-first-attempts-2", + "readStatus": "unavailable-not-API-contract-evidence" + }, + { + "requestedUrl": "https://developers.openai.com/api/docs/guides/reasoning.md", + "observedAt": "2026-11-06T13:04:04.020437+00:00", + "readLimitBytes": 1800000, + "status": 403, + "error": "HTTP Error 403: Forbidden", + "finalUrl": "https://developers.openai.com/api/docs/guides/reasoning.md", + "contentType": "text/plain; charset=utf-8", + "bytes": 59, + "sha256": "4e0849b85e841fcbed1f848b58db441822f65dbf3a3660c036bab6981b63a845", + "bodyPath": ".qa/official-first-system-audit-2026-11-06/openai-attempt-3.error-body", + "attemptLedger": ".qa/official-first-system-audit-2026-11-06/openai-first-attempts.json", + "id": "openai-first-attempts-3", + "readStatus": "unavailable-not-API-contract-evidence" + }, + { + "requestedUrl": "https://platform.openai.com/docs/api-reference/responses/create", + "observedAt": "2026-11-06T13:04:04.020437+00:00", + "readLimitBytes": 1800000, + "status": 403, + "error": "HTTP Error 403: Forbidden", + "finalUrl": "https://platform.openai.com/docs/api-reference/responses/create", + "contentType": "text/html; charset=UTF-8", + "bytes": 5660, + "sha256": "02fb68a392f7539657f390047853c6dd1c9071a66ee568354245d4ea21cc2018", + "bodyPath": ".qa/official-first-system-audit-2026-11-06/openai-attempt-4.error-body", + "attemptLedger": ".qa/official-first-system-audit-2026-11-06/openai-first-attempts.json", + "id": "openai-first-attempts-4", + "readStatus": "unavailable-not-API-contract-evidence" + }, + { + "requestedUrl": "https://developers.openai.com/api/docs/guides/text.md", + "observedAt": "2026-11-06T13:04:04.497667+00:00", + "readLimitBytes": 1800000, + "status": 403, + "error": "HTTP Error 403: Forbidden", + "finalUrl": "https://developers.openai.com/api/docs/guides/text.md", + "contentType": "text/plain; charset=utf-8", + "bytes": 59, + "sha256": "af001f5caf9d2fff94894166c9ec84a2849e4cb8dab47b77781d7e9bb5768f1c", + "bodyPath": ".qa/official-first-system-audit-2026-11-06/openai-attempt-5.error-body", + "attemptLedger": ".qa/official-first-system-audit-2026-11-06/openai-first-attempts.json", + "id": "openai-first-attempts-5", + "readStatus": "unavailable-not-API-contract-evidence" + } + ], + "reused": [ + { + "path": "docs/claude-official-capacity-2026-11-06.json", + "sha256": "44c980b97c3058658cb461031977c92fe53a015cb78f3b2aae584e5d50bad353", + "scope": "four exact current Claude IDs; synchronous 128000 output / 1000000 context; prior source verification, not live capacity validation" + }, + { + "path": "docs/newapi-documentation-review-2026-11-06.json", + "sha256": "b3d3c36a494563ed7ceb993dae7a1f9b346b543877843108e035c0a26b5d1601", + "scope": "305 article bodies / 201 embedded schemas already reviewed; reused without new full crawl", + "sourceManifestSha256": "4cb416511da1aa6f14171024729a4c8f891f589250ed4c676da5f26cd85b56e6", + "repositoryCommit": "d254f1842dc3d08d60f2269e2c9d4c9876a3dfbc", + "authority": "gateway-and-compatibility-reference-not-provider-authority" + } + ], + "limits": [ + "targeted contract review, not all official websites or every global provider", + "SDK schema is an official contract source, not acquired current web guides or live provider acceptance", + "403/timeout pages are acquisition failures, not API model failures", + "official capacity does not authenticate a proxy backend or establish its effective limit" + ] +} diff --git a/docs/official-first-system-audit-2026-11-06.json b/docs/official-first-system-audit-2026-11-06.json new file mode 100644 index 00000000..89b78dcb --- /dev/null +++ b/docs/official-first-system-audit-2026-11-06.json @@ -0,0 +1,1982 @@ +{ + "schema": 1, + "kind": "official-first-complete-system-key-path-audit", + "recordedAt": "2026-11-06T14:09:15.016711+00:00", + "sourceHead": "4da2550becdf4750320cedd5ee30a7e38baba2f9", + "branch": "feat/task-agreement-1.25", + "scope": "inventory and all major-layer key-path audit, not manual line-by-line review of every file", + "production": { + "freshHealthEvidence": ".qa/official-first-system-audit-2026-11-06/formal-health-readonly.json", + "freshHealthSha256": "456fdd09327c95232c00cc054824419d06517dcf3b24312e8a7fbc5b4e6440b0", + "observedHealth": [ + { + "route": "/api/health", + "status": 200, + "observedAt": "2026-11-06T13:08:18.279628+00:00", + "body": { + "app": "agent-workbench", + "version": "1.25.12", + "pid": 71628, + "ready": true, + "uptimeMs": 43891093 + }, + "sha256": "d7df430a07e4daf2aeea5f26d5ea77b8554f45849d90cf835df2dac2de656901" + } + ], + "priorDeployedSource": "6867cd5a3094c133b09a7c50384ae844ef34c038", + "retentionScope": "prior F deployment evidence reused; health is not fresh data/credential retention verification" + }, + "pendingSourceFixes": [ + { + "commit": "ae5123c5", + "change": "catalog existing-model metadata refresh", + "deployed": false + }, + { + "commit": "341f5f69", + "change": "diagnostic follows actual output token wire fields", + "deployed": false + }, + { + "commit": "2bd7b24a", + "change": "exact official Claude capacity fallback", + "deployed": false + } + ], + "thisAudit": { + "productSourceChanges": 0, + "realModelRequests": 0, + "realCredentialsRead": 0, + "productionWrites": 0, + "releaseOrRestart": false, + "publicDocumentationRequests": true, + "ownedLocalUiFixture": true, + "offlineNumberedObservations": 21, + "separateCorrections": [ + "O18-C" + ], + "sourceInventoryVerified": 715, + "inventory": { + "src": { + "files": 272, + "lines": 31738 + }, + "tests": { + "files": 171, + "lines": 20981 + }, + "scripts": { + "files": 272, + "lines": 24917 + } + } + }, + "validationBoundary": { + "offlineProbes": "21 observations and one correction; success means reproduction completed, not product correctness", + "ui": [ + "owned1280 settings and persistence; owned390 actual controls/full form", + "seven prior build inputs and thirty compiled outputs verified unchanged", + "not current production Electron acceptance" + ], + "priorTargetedChecks": [ + "126/126 New API repairs", + "67/67 Claude capacity repair" + ], + "priorFailures": "original regression/environment/network failures remain; not rerun or relabelled as pass" + }, + "sourcesPath": "docs/official-first-sources-2026-11-06.json", + "sourcesSha256": "d230cfa36f12653757c9fb73d238dcbf0790d8587e3a66dffcb83765c5caaf59", + "evidenceSeal": { + "manifestPath": ".qa/official-first-system-audit-2026-11-06/MANIFEST.json", + "summaryPath": "docs/official-first-audit-evidence-2026-11-06.json", + "scope": "immutable raw audit files plus four final report/data inputs; mutable commit/push receipts kept separately" + }, + "observations": [ + { + "id": "O01", + "title": "Responses effort control deletes sibling summary", + "category": "defect", + "expected": { + "reasoning": { + "effort": "high", + "summary": "auto" + } + }, + "observed": { + "reasoning": { + "effort": "high" + }, + "text": { + "verbosity": "medium" + } + }, + "evidencePath": ".qa/official-first-system-audit-2026-11-06/offline-contract-observations.json", + "evidenceSha256": "f2a8c574aabcda2a6c0e7d2290a462dd6fca8673b10ecab56d0a32dbf808b4c0" + }, + { + "id": "O02", + "title": "Messages effort/mode controls delete format/display", + "category": "defect", + "expected": { + "preserve": [ + "output_config.format", + "thinking.display" + ] + }, + "observed": { + "effortChange": { + "output_config": { + "effort": "max" + }, + "thinking": { + "type": "adaptive", + "display": "summarized" + } + }, + "modeChange": { + "output_config": { + "effort": "high", + "format": { + "type": "json_schema", + "schema": { + "type": "object" + } + } + }, + "thinking": { + "type": "adaptive" + } + } + }, + "evidencePath": ".qa/official-first-system-audit-2026-11-06/offline-contract-observations.json", + "evidenceSha256": "f2a8c574aabcda2a6c0e7d2290a462dd6fca8673b10ecab56d0a32dbf808b4c0" + }, + { + "id": "O03", + "title": "Metadata-absent Opus 5.5 still offers rejected manual/disabled modes", + "category": "defect", + "expected": { + "officialModes": [ + "adaptive" + ], + "manualEnabled": "official 400", + "disabled": "official 400" + }, + "observed": { + "offered": [ + "disabled", + "enabled", + "adaptive" + ], + "generated": { + "thinking": { + "type": "enabled", + "budget_tokens": 1024 + } + } + }, + "evidencePath": ".qa/official-first-system-audit-2026-11-06/offline-contract-observations.json", + "evidenceSha256": "f2a8c574aabcda2a6c0e7d2290a462dd6fca8673b10ecab56d0a32dbf808b4c0" + }, + { + "id": "O04", + "title": "Catalog loses between_tools and empty supported-mode list falls back to all modes", + "category": "defect", + "expected": { + "sonnetModes": [ + "adaptive", + "between_tools" + ], + "explicitNegativeModes": [] + }, + "observed": { + "sonnetCatalog": { + "id": "claude-sonnet-5-5", + "name": "claude-sonnet-5-5", + "owner": "", + "lifecycle": null, + "contextLimit": 1000000, + "maxOutputTokens": 128000, + "capabilityDeclarations": { + "text": "unknown", + "tools": "unknown", + "streaming": "unknown", + "vision": "unknown", + "audio": "unknown" + }, + "features": { + "thinking": "supported", + "effort": "unknown", + "pdf_input": "unknown", + "structured_outputs": "unknown", + "server_tools": "unknown", + "code_execution": "unknown", + "batch": "unknown", + "citations": "unknown", + "context_management": "unknown" + }, + "efforts": [], + "modes": [ + "adaptive" + ], + "parameters": [], + "metadataSource": "upstream-model-list" + }, + "negativeProfile": { + "id": "anthropic", + "label": "Claude Messages 兼容", + "source": "按品牌、型号与协议匹配,可手动更改", + "efforts": [ + "low", + "medium", + "high", + "xhigh", + "max" + ], + "modes": [ + "disabled", + "enabled", + "adaptive" + ], + "thinking": true + } + }, + "evidencePath": ".qa/official-first-system-audit-2026-11-06/offline-contract-observations.json", + "evidenceSha256": "f2a8c574aabcda2a6c0e7d2290a462dd6fca8673b10ecab56d0a32dbf808b4c0" + }, + { + "id": "O05", + "title": "Feature table says structured parameters rejected while open wire forwards them", + "category": "display_drift", + "expected": "Describe forwarding separately from structured-output validation support", + "observed": { + "row": { + "name": "结构化输出约束", + "upstream": "待确认", + "local": "尚未接入,参数保存时明确拒绝" + }, + "forwarded": { + "type": "json_object" + } + }, + "evidencePath": ".qa/official-first-system-audit-2026-11-06/offline-contract-observations.json", + "evidenceSha256": "f2a8c574aabcda2a6c0e7d2290a462dd6fca8673b10ecab56d0a32dbf808b4c0" + }, + { + "id": "O06", + "title": "Messages block/tool cache controls are dropped; top-level control survives", + "category": "missing_adapter_capability", + "expected": "Retain official block/tool fields via an explicit native content/tool representation", + "observed": { + "system": [ + { + "type": "text", + "text": "stable audit prefix" + } + ], + "tools": [ + { + "name": "audit_tool", + "description": "No execution", + "input_schema": { + "type": "object" + } + } + ], + "topLevel": { + "type": "ephemeral" + } + }, + "evidencePath": ".qa/official-first-system-audit-2026-11-06/offline-contract-observations.json", + "evidenceSha256": "f2a8c574aabcda2a6c0e7d2290a462dd6fca8673b10ecab56d0a32dbf808b4c0" + }, + { + "id": "O07", + "title": "Kimi cache-write usage preserved raw but absent from normalized usage", + "category": "missing_adapter_capability", + "expected": "Normalized cache read/write/uncached fields with provider provenance", + "observed": { + "prompt_tokens": 1000, + "completion_tokens": 100, + "total_tokens": 1100, + "prompt_tokens_details": { + "cached_tokens": 600 + }, + "raw": { + "prompt_tokens": 1000, + "completion_tokens": 100, + "total_tokens": 1100, + "prompt_tokens_details": { + "cached_tokens": 600, + "cache_write_tokens": 300, + "uncached_tokens": 100 + } + }, + "protocol": "chat-completions" + }, + "evidencePath": ".qa/official-first-system-audit-2026-11-06/offline-contract-observations.json", + "evidenceSha256": "f2a8c574aabcda2a6c0e7d2290a462dd6fca8673b10ecab56d0a32dbf808b4c0" + }, + { + "id": "O08", + "title": "Automatic pool selects declared unsupported member in unrestricted mode", + "category": "selection_defect", + "expected": "Automatic ranking prefers capable member while explicit manual selection remains unrestricted", + "observed": { + "first": "incapable", + "requirements": [ + "tools" + ], + "declarations": [ + { + "id": "incapable", + "tools": "unsupported" + }, + { + "id": "capable", + "tools": "supported" + } + ] + }, + "evidencePath": ".qa/official-first-system-audit-2026-11-06/offline-contract-observations.json", + "evidenceSha256": "f2a8c574aabcda2a6c0e7d2290a462dd6fca8673b10ecab56d0a32dbf808b4c0" + }, + { + "id": "O09", + "title": "Affinity is Run-scoped, not session/task-scoped", + "category": "missing_routing_capability", + "expected": "Explicit conversational affinity strategy for cross-Run continuations", + "observed": { + "runA": "incapable", + "runAAgain": "incapable", + "nextRun": "capable", + "affinityKey": "runId" + }, + "evidencePath": ".qa/official-first-system-audit-2026-11-06/offline-contract-observations.json", + "evidenceSha256": "f2a8c574aabcda2a6c0e7d2290a462dd6fca8673b10ecab56d0a32dbf808b4c0" + }, + { + "id": "O10", + "title": "Unlimited attempt persistence conflicts with advertised retention and lacks full ledger view", + "category": "display_and_storage_gap", + "expected": { + "actualRetention": "unlimited", + "fullLedger": "paginated" + }, + "observed": { + "persisted": 1001, + "view": 100, + "advertisedRetention": 1000 + }, + "evidencePath": ".qa/official-first-system-audit-2026-11-06/offline-contract-observations.json", + "evidenceSha256": "f2a8c574aabcda2a6c0e7d2290a462dd6fca8673b10ecab56d0a32dbf808b4c0" + }, + { + "id": "O11", + "title": "Native server-tool blocks and pause continuation not implemented", + "category": "missing_adapter_capability", + "expected": "Typed native output and continuation handling, distinct from truncated/incomplete tools", + "observed": [ + { + "id": "server-tool", + "accepted": false, + "diagnostic": { + "kind": "protocol", + "source": "model_connection", + "origin": "local", + "stage": "model_request", + "title": "模型响应未通过完整性核对", + "message": "Messages 包含未接入或不完整的内容块:server_tool_use。", + "retryable": false, + "occurredAt": "2026-11-06T13:21:14.809Z" + }, + "message": "Messages 包含未接入或不完整的内容块:server_tool_use。" + }, + { + "id": "pause-turn", + "accepted": false, + "diagnostic": { + "kind": "protocol", + "source": "model_connection", + "origin": "local", + "stage": "model_request", + "title": "模型响应未通过完整性核对", + "message": "Messages 未完整结束或停止原因为上下文上限、暂停或未知;原响应未用于执行工具。", + "retryable": false, + "occurredAt": "2026-11-06T13:21:14.810Z" + }, + "message": "Messages 未完整结束或停止原因为上下文上限、暂停或未知;原响应未用于执行工具。" + } + ], + "evidencePath": ".qa/official-first-system-audit-2026-11-06/offline-contract-observations.json", + "evidenceSha256": "f2a8c574aabcda2a6c0e7d2290a462dd6fca8673b10ecab56d0a32dbf808b4c0" + }, + { + "id": "O12", + "title": "Gemini-style entries parse but native nextPageToken does not", + "category": "missing_native_protocol", + "expected": "Native list/endpoint/pagination/generate/stream/auth contract", + "observed": { + "entries": [ + { + "id": "models/audit-gemini", + "name": "models/audit-gemini", + "owner": "", + "lifecycle": null, + "contextLimit": 1000000, + "maxOutputTokens": 65536, + "capabilityDeclarations": { + "text": "unknown", + "tools": "unknown", + "streaming": "unknown", + "vision": "unknown", + "audio": "unknown" + }, + "features": { + "thinking": "unknown", + "effort": "unknown", + "pdf_input": "unknown", + "structured_outputs": "unknown", + "server_tools": "unknown", + "code_execution": "unknown", + "batch": "unknown", + "citations": "unknown", + "context_management": "unknown" + }, + "efforts": [], + "modes": [], + "parameters": [], + "metadataSource": "upstream-model-list" + } + ], + "hasMore": false, + "cursor": "" + }, + "evidencePath": ".qa/official-first-system-audit-2026-11-06/offline-contract-observations.json", + "evidenceSha256": "f2a8c574aabcda2a6c0e7d2290a462dd6fca8673b10ecab56d0a32dbf808b4c0" + }, + { + "id": "O13", + "category": "missing_mcp_schema_support", + "title": "Ordinary union and 65-property schemas rejected regardless of open mode", + "observed": [ + { + "id": "anyOf", + "accepted": false, + "message": "工具schema超出当前支持的有界JSON Schema范围。" + }, + { + "id": "65-properties", + "accepted": false, + "message": "工具schema属性范围无效。" + } + ], + "expected": "Support vendor/standard schemas without arbitrary work-size caps; unsupported schema must remain explicit until implemented.", + "evidencePath": ".qa/official-first-system-audit-2026-11-06/offline-harness-observations.json", + "evidenceSha256": "16ecdcefb2ea46b5f810e247a473c0a17a73a56cbaedd2b939ebff7c6c76e8e6" + }, + { + "id": "O14", + "category": "remaining_hard_cap", + "title": "Open-mode MCP task still blocked after 30 operations", + "observed": "当前任务已达到30次MCP操作上限。", + "expected": "0/unlimited in unrestricted mode; keep each invocation identity and unknown outcome handling.", + "evidencePath": ".qa/official-first-system-audit-2026-11-06/offline-harness-observations.json", + "evidenceSha256": "16ecdcefb2ea46b5f810e247a473c0a17a73a56cbaedd2b939ebff7c6c76e8e6" + }, + { + "id": "O15", + "category": "remaining_hard_cap", + "title": "Open-mode checkpoint still blocks after 2000 tool references", + "observed": "当前任务超过恢复包2000项工具引用的上限,请缩小任务范围。", + "expected": "Paged/streamed verified checkpoints with unlimited default, not a request to shrink the task.", + "evidencePath": ".qa/official-first-system-audit-2026-11-06/offline-harness-observations.json", + "evidenceSha256": "16ecdcefb2ea46b5f810e247a473c0a17a73a56cbaedd2b939ebff7c6c76e8e6" + }, + { + "id": "O16", + "category": "instruction_drift", + "title": "MCP model-facing instruction always asks confirmation although full policy automatically approves", + "observed": { + "description": "请求用户确认后调用本次目录的已启用MCP工具。先核对固定程序/实际握手/schema,不自动重试;完整结果按工件读取。", + "fullModePolicy": { + "approve": true, + "mode": "full", + "source": "full-access-policy" + } + }, + "expected": "Model-facing description derived from active approval mode; do not imply an extra manual confirmation.", + "evidencePath": ".qa/official-first-system-audit-2026-11-06/offline-harness-observations.json", + "evidenceSha256": "16ecdcefb2ea46b5f810e247a473c0a17a73a56cbaedd2b939ebff7c6c76e8e6" + }, + { + "id": "O17", + "category": "silent_parameter_rewrite", + "title": "Chat streaming replaces explicit stream_options", + "expected": { + "include_usage": false, + "include_obfuscation": false + }, + "observed": { + "include_usage": true + }, + "evidencePath": ".qa/official-first-system-audit-2026-11-06/offline-forwarding-observations.json", + "evidenceSha256": "ef657bbe3d07c6c5a85227da29ccde24208f1fcd7e68fcf95f2706aa871c94a9" + }, + { + "id": "O18", + "category": "incomplete_response_selection", + "title": "Multiple choices remain raw; harness consumes only first choice", + "observed": { + "acceptedRawChoices": 2, + "canonicalChoice": 0 + }, + "expected": "Explicit candidate-selection contract and complete raw retention; never apply multiple alternatives blindly.", + "sourceInference": "Provider and Chat stream choose index 0; this probe establishes acceptance, not a live stream/UI run.", + "evidencePath": ".qa/official-first-system-audit-2026-11-06/offline-forwarding-observations.json", + "evidenceSha256": "ef657bbe3d07c6c5a85227da29ccde24208f1fcd7e68fcf95f2706aa871c94a9" + }, + { + "id": "O19", + "category": "cache_observation_gap", + "title": "Prompt prefix observation excludes native parameters that may affect supplier cache", + "observed": { + "schemaVersion": 2, + "method": "ordered-wire-prompt-fields-utf8", + "upstreamCache": "unknown", + "variants": [ + { + "connectionId": null, + "protocol": "chat-completions", + "model": "audit-only-model", + "toolsSha256": "4f53cda18c2baa0c0354bb5f9a3ecbe5ed12ab4d8e11ba873c2f11161202b945", + "toolBytes": 2, + "toolNames": [], + "promptBytes": 100, + "previousPromptBytes": 100, + "commonPrefixBytes": 100, + "reason": "same_prompt", + "blocks": [ + { + "path": "model", + "role": null, + "sha256": "bc9b64afd40848b6216a18919080b26718cc5875166afe314b32309c95e83e69", + "bytes": 18 + }, + { + "path": "tools", + "role": null, + "sha256": "4f53cda18c2baa0c0354bb5f9a3ecbe5ed12ab4d8e11ba873c2f11161202b945", + "bytes": 2 + }, + { + "path": "system", + "role": null, + "sha256": "74234e98afe7498fb5daf1f36ac2d78acc339464f950703b8c019892f982b90b", + "bytes": 4 + }, + { + "path": "messages[0]", + "role": "user", + "sha256": "684dfefce8cea4319a87556ed5340b6d2b9f530c31cc5637ea07de5524cc62bd", + "bytes": 33 + } + ], + "firstChange": null, + "tools": [] + } + ] + }, + "expected": "Keep ordered prompt observation distinct from actual supplier cache; add a vendor-specific cache-relevant parameter fingerprint.", + "evidencePath": ".qa/official-first-system-audit-2026-11-06/offline-forwarding-observations.json", + "evidenceSha256": "ef657bbe3d07c6c5a85227da29ccde24208f1fcd7e68fcf95f2706aa871c94a9" + }, + { + "id": "O18-C", + "corrects": "O18.sourceInference", + "kind": "correction", + "originalEvidence": "offline-forwarding-observations.json", + "originalPreserved": true, + "finding": "Nonstreaming accepts multiple raw choices and Provider consumes choices[0]; Chat streaming rejects every choice.index other than 0 locally.", + "observed": { + "streamedSecondChoiceRejected": true, + "diagnostic": "当前只接入单个Chat候选流。", + "streamTerminal": false, + "capturedBytes": 247 + }, + "limitation": "Synthetic in-memory SSE; no live supplier, HTTP request, UI assertion, or alternative execution.", + "evidencePath": ".qa/official-first-system-audit-2026-11-06/offline-correction-observations.json", + "evidenceSha256": "8fd84fb828a72e1ca145d33b908e2e090467dce744dabf2a21d6a14f8a2669a7" + }, + { + "id": "O20", + "category": "policy_display_drift", + "title": "Identity policy says require-match while unrestricted execution records and continues", + "observed": { + "configuredPolicy": "require-match", + "executionMode": "unrestricted", + "reportedIdentity": "mismatch", + "requestStatus": "completed", + "consumptionOutcome": "completed", + "mockFetchCalls": 1 + }, + "sourceInterpretation": "ModelIdentityDetails displays pause for require-match regardless of executionMode; production UI was not changed or exercised by this probe.", + "expected": "Expose requested and effective policy accurately, and preserve the unrestricted default; do not reintroduce compulsory identity interception.", + "evidencePath": ".qa/official-first-system-audit-2026-11-06/offline-correction-observations.json", + "evidenceSha256": "8fd84fb828a72e1ca145d33b908e2e090467dce744dabf2a21d6a14f8a2669a7" + }, + { + "id": "O21", + "category": "fixed-load-count-cap", + "title": "Saved-credential metadata with 102 entries is marked unavailable even in unrestricted mode", + "observed": [ + { + "entries": 101, + "blocked": false, + "loadedEntries": 101, + "originalMetaEntries": 101 + }, + { + "entries": 102, + "blocked": true, + "loadedEntries": 0, + "originalMetaEntries": 102 + } + ], + "limitation": "Only synthetic metadata count gate; no encryption/decryption, valid saved payload restore, real Key or real vault read; original meta is not deleted.", + "expected": "Saved-credential metadata should load by supported version and record validity, not a fixed connection-count cap.", + "evidencePath": ".qa/official-first-system-audit-2026-11-06/offline-credential-scope-observations.json", + "evidenceSha256": "fb9cba4aaadf9ab1bb4cc4f143f9b9ecf32cff0636a2c9aa78e786f92fd0cec1" + } + ], + "capabilities": [ + { + "id": "C01", + "name": "模型发现、多连接、多模型池", + "implementation": "implemented-partial", + "validation": "local-and-recorded-live", + "deployment": "1.25.12; refresh repair source-only", + "issues": [ + "I04", + "I13", + "I15", + "I29" + ] + }, + { + "id": "C02", + "name": "原生参数与推理选择", + "implementation": "implemented-with-defects", + "validation": "offline-and-owned-web-observed", + "deployment": "1.25.12; current controls still defective", + "issues": [ + "I01", + "I02", + "I03", + "I08" + ] + }, + { + "id": "C03", + "name": "官方容量、默认本地透传", + "implementation": "implemented-partial", + "validation": "67 prior targeted plus current audit", + "deployment": "passthrough deployed; official fallback source-only", + "issues": [ + "I29" + ] + }, + { + "id": "C04", + "name": "官方认证/版本/云平台", + "implementation": "implemented-bearer-and-x-api-key-only", + "validation": "local-contract-not-global-live", + "deployment": "1.25.12 for current subset", + "issues": [ + "I14", + "I16", + "I17" + ] + }, + { + "id": "C05", + "name": "JSON/SSE、原错与unknown用量", + "implementation": "implemented-with-coverage-gaps", + "validation": "historical-live-and-current-offline", + "deployment": "1.25.12; wire diagnostic repair source-only", + "issues": [ + "I02", + "I10", + "I18", + "I21", + "I28", + "I29" + ] + }, + { + "id": "C06", + "name": "ask/auto/full持久权限与开放执行", + "implementation": "implemented-with-residual-caps", + "validation": "historical-runtime-and-current-offline", + "deployment": "1.25.12", + "issues": [ + "I05", + "I06", + "I07", + "I08", + "I33" + ] + }, + { + "id": "C07", + "name": "型号声明/响应身份记录", + "implementation": "implemented-not-weight-authentication", + "validation": "historical-live-and-current-mock", + "deployment": "1.25.12", + "issues": [ + "I08" + ] + }, + { + "id": "C08", + "name": "自动适配与会话连续性", + "implementation": "implemented-run-routing-only", + "validation": "offline-observed", + "deployment": "1.25.12", + "issues": [ + "I04", + "I11", + "I15" + ] + }, + { + "id": "C09", + "name": "原生多模态/文件", + "implementation": "basic-png-image-only", + "validation": "prior-owned-vision-ui", + "deployment": "1.25.12 for basic image", + "issues": [ + "I20" + ] + }, + { + "id": "C10", + "name": "原生服务端工具/续接/候选/后台", + "implementation": "not-implemented-completely", + "validation": "rejection-observed-and-source", + "deployment": "not deployed as workflow", + "issues": [ + "I18", + "I19", + "I21", + "I37" + ] + }, + { + "id": "C11", + "name": "文件读写/改动集/自动应用/Git回滚", + "implementation": "implemented", + "validation": "historical-Node-React-Python-delivery-and-local-QA", + "deployment": "1.25.12", + "issues": [] + }, + { + "id": "C12", + "name": "环境诊断/依赖/Node/Python检查", + "implementation": "implemented", + "validation": "AH/local-and-historical-AD-AE", + "deployment": "1.25.12 including AH/AI", + "issues": [] + }, + { + "id": "C13", + "name": "命令执行/完整日志/停止/Git证据", + "implementation": "implemented-pipes", + "validation": "AI/local-and-historical-live", + "deployment": "1.25.12", + "issues": [ + "I22" + ] + }, + { + "id": "C14", + "name": "要求证据/单轮观察/终态报告", + "implementation": "implemented", + "validation": "historical-A-B-AI-QA", + "deployment": "1.25.12", + "issues": [ + "I24", + "I35" + ] + }, + { + "id": "C15", + "name": "工具目录/原请求前缀/重复读取观察", + "implementation": "implemented-observation-not-benefit", + "validation": "historical-C-D-and-current-offline", + "deployment": "1.25.12", + "issues": [ + "I12", + "I35" + ] + }, + { + "id": "C16", + "name": "缓存控制/归一统计/会话缓存", + "implementation": "implemented-partial", + "validation": "one-live-cache-observation-and-offline", + "deployment": "1.25.12 subset", + "issues": [ + "I09", + "I10", + "I11", + "I12", + "I35" + ] + }, + { + "id": "C17", + "name": "可关闭提醒与实际回读", + "implementation": "implemented-default-off", + "validation": "D-owned-HTTP-web-desktop-release", + "deployment": "1.25.10 onward", + "issues": [ + "I35" + ] + }, + { + "id": "C18", + "name": "来源经验草稿/采用撤销/完整修订", + "implementation": "implemented-semantic-benefit-unknown", + "validation": "E-owned-and-release-QA", + "deployment": "1.25.11 onward", + "issues": [ + "I35" + ] + }, + { + "id": "C19", + "name": "Worker角色/模型绑定/真实质量回收", + "implementation": "implemented-partial", + "validation": "F-owned-and-release-QA", + "deployment": "1.25.12", + "issues": [ + "I24", + "I25", + "I35" + ] + }, + { + "id": "C20", + "name": "Skills/MCP生态", + "implementation": "implemented-local-subset", + "validation": "historical-QA-and-current-offline", + "deployment": "1.25.12", + "issues": [ + "I05", + "I06", + "I08", + "I25" + ] + }, + { + "id": "C21", + "name": "内置浏览器/页面预览/DOM动作", + "implementation": "implemented-fixed-action-set", + "validation": "historical-web-desktop-and-current-owned-form", + "deployment": "1.25.12", + "issues": [ + "I23" + ] + }, + { + "id": "C22", + "name": "Windows桌面桥/本机加密凭据", + "implementation": "implemented-not-general-OS-control", + "validation": "historical-DPAPI-and-install-QA-not-real-key-read", + "deployment": "1.25.12", + "issues": [ + "I23", + "I38" + ] + }, + { + "id": "C23", + "name": "DeepSeek Harness/外部完整引擎", + "implementation": "DeepSeek-bridge-implemented-others-missing", + "validation": "fixed-SDK-owned-and-historical-live", + "deployment": "1.25.12 bridge; full Codex/Claude Code not implemented", + "issues": [] + }, + { + "id": "C24", + "name": "原版Purge加载与对照", + "implementation": "mechanism-implemented-benefit-unknown", + "validation": "historical-AG-four-groups-not-comparable", + "deployment": "1.25.12 default-off option", + "issues": [ + "I36" + ] + }, + { + "id": "C25", + "name": "REA来源分析与证据工件", + "implementation": "implemented-static-job-adapter", + "validation": "historical-source-and-owned-QA", + "deployment": "1.25.12", + "issues": [] + }, + { + "id": "C26", + "name": "Companion手机查看/批准/停止", + "implementation": "implemented-real-current-device-unverified", + "validation": "historical-owned-contracts", + "deployment": "1.25.12", + "issues": [ + "I32", + "I33" + ] + }, + { + "id": "C27", + "name": "Mnemo及跨产品协作", + "implementation": "read-only-contract-feedback-only", + "validation": "synthetic-original-package-review", + "deployment": "native export/inbound delegation missing", + "issues": [ + "I31" + ] + }, + { + "id": "C28", + "name": "SQLite事件/工件/恢复检查点", + "implementation": "implemented-with-caps-and-scale-risk", + "validation": "historical-recovery-and-current-offline/source", + "deployment": "1.25.12", + "issues": [ + "I07", + "I13", + "I26" + ] + }, + { + "id": "C29", + "name": "源码打包/独立程序/安装回退", + "implementation": "implemented-source-data-separated", + "validation": "historical-F-release-and-retention", + "deployment": "1.25.12; three later source repairs pending", + "issues": [ + "I27", + "I28", + "I29" + ] + }, + { + "id": "C30", + "name": "Orven三配色/任务与模型中心界面", + "implementation": "implemented-with-copy-and-controls-drift", + "validation": "historical-release plus owned1280/390", + "deployment": "1.25.12 for brand; no new production UI audit", + "issues": [ + "I03", + "I08", + "I30" + ] + } + ], + "issues": [ + { + "id": "I01", + "title": "推理快捷编辑丢失嵌套原生字段", + "priority": "P0", + "statusAtAudit": "confirmed-defect", + "impact": "修改 effort 或 thinking 会删除 summary、output_config.format、thinking.display,用户请求未如实保留。", + "nextChange": "只修改目标键;删除目标值也不删除同对象其他字段;明确默认重置的范围。", + "acceptance": "三协议往返、JSON/快捷控件/保存/重新打开,非目标字段逐项保持。", + "observations": [ + "O01", + "O02" + ], + "references": [ + { + "path": "src/shared/contracts/model-parameter-profiles.mjs", + "line": 54, + "sha256": "1264a90ba612b721ea639206cd6d8f595309c7bda38f80c45bb13c78af4a2d1f", + "match": " else next[effortField] = effortField === 'reasoning_effort' ? effort : { effort };" + }, + { + "path": "src/shared/contracts/model-parameter-profiles.mjs", + "line": 57, + "sha256": "1264a90ba612b721ea639206cd6d8f595309c7bda38f80c45bb13c78af4a2d1f", + "match": " else if (field === 'budget') next.thinking = { type: 'enabled', budget_tokens: value };" + } + ], + "dependsOn": [], + "fixedInThisAudit": false + }, + { + "id": "I02", + "title": "Chat 流式覆盖用户 stream_options", + "priority": "P0", + "statusAtAudit": "confirmed-defect", + "impact": "include_usage=false 与其他上游 stream_options 被整对象替换。", + "nextChange": "透传显式原生设置;默认建议与用户值分开,不隐式覆盖。", + "acceptance": "实际本机 HTTP wire 与用户对象一致,缺失字段才按明确策略补充。", + "observations": [ + "O17" + ], + "references": [ + { + "path": "src/shared/contracts/model-request.mjs", + "line": 22, + "sha256": "5606dc2c245e3461c35613773c0d1b47718acd28a6fe4abd07406bbe80af2e11", + "match": " ...(stream ? { stream_options: { include_usage: true } } : {}) };" + } + ], + "dependsOn": [], + "fixedInThisAudit": false + }, + { + "id": "I03", + "title": "Claude 型号推理能力与官方未对齐", + "priority": "P0", + "statusAtAudit": "confirmed-defect", + "impact": "Opus5.5 仍能生成官方拒绝的 enabled/disabled;between_tools 未解析;全否定目录又回退全部选项。", + "nextChange": "以精确型号、协议、官方版本解析模式/effort/条件,区分未知与明确不支持。", + "acceptance": "官方新旧型号矩阵、目录全否定和缺失情况通过;UI/保存/wire 一致;兼容显式覆盖仍可用。", + "observations": [ + "O03", + "O04" + ], + "references": [ + { + "path": "src/shared/contracts/model-catalog.mjs", + "line": 38, + "sha256": "abb7b41982d072ecdb8c8d7c2feb67228eeb03c751c3f10b8557ac12b25022e1", + "match": " const modes = ['disabled', 'enabled', 'adaptive'].filter(value => supported(caps.thinking?.types?.[value]) === 'supported');" + }, + { + "path": "src/shared/contracts/model-parameter-profiles.mjs", + "line": 27, + "sha256": "1264a90ba612b721ea639206cd6d8f595309c7bda38f80c45bb13c78af4a2d1f", + "match": " modes: metadata?.features?.thinking === 'unsupported' ? [] : metadata?.modes?.length ? metadata.modes : ['disabled', 'enabled', ...(protocol === 'anthropic-messages' ? ['adaptive'] : [])]," + } + ], + "dependsOn": [], + "fixedInThisAudit": false + }, + { + "id": "I04", + "title": "自动模型池可选明确不支持工具的成员", + "priority": "P0", + "statusAtAudit": "confirmed-selection-defect", + "impact": "需要工程工具时,自动模式仍可把请求发给明确不支持工具的模型。", + "nextChange": "自动候选按能力与任务排序,显式手选/强制尝试与自动适配分别记录,不恢复全局拒发。", + "acceptance": "自动选择工具支持成员;未知有明确策略;用户手选仍可透传;记录选择依据。", + "observations": [ + "O08" + ], + "references": [ + { + "path": "src/core/models/model-hub.mjs", + "line": 321, + "sha256": "630c276a445641e3abaf920957e01d6ffd1b50f8264c0227ee55d29073d9dbb9", + "match": " const eligible = routing.members.filter(value => unrestricted(this.store) || capabilityRejections(value, requirements, value.credentialRevision).length === 0);" + } + ], + "dependsOn": [], + "fixedInThisAudit": false + }, + { + "id": "I05", + "title": "开放模式仍有每任务 30 次 MCP 操作上限", + "priority": "P0", + "statusAtAudit": "confirmed-execution-cap", + "impact": "长工程在已有开放权限下仍被固定次数挡住。", + "nextChange": "移除开放模式次数门槛,完整账本用游标/分页;保留每次真实身份、结果和停止。", + "acceptance": "同任务第31次及更长序列可执行且无旧记录丢失;停止/重启 unknown 保留。", + "observations": [ + "O14" + ], + "references": [ + { + "path": "src/core/tools/mcp-registry.mjs", + "line": 121, + "sha256": "6e91279f20aa08850478f29155702f6f6b41d576ea7d58204d84a66d7b4ec360", + "match": " if (this.jobs(session).filter(job => job.taskId === session.taskId).length >= 30) throw mcpError('当前任务已达到30次MCP操作上限。');" + } + ], + "dependsOn": [], + "fixedInThisAudit": false + }, + { + "id": "I06", + "title": "MCP JSON Schema 子集拒绝常见合法工具", + "priority": "P0", + "statusAtAudit": "confirmed-adapter-gap", + "impact": "anyOf 和65属性等普通 schema 无法接入,与开放工具能力目标不符。", + "nextChange": "支持官方/服务器声明的 JSON Schema 或按协议原样转交;不用固定属性数假冒模型能力。", + "acceptance": "union/ref/嵌套/65以上属性工具真实握手后可选择;错误与原 schema 可核对。", + "observations": [ + "O13" + ], + "references": [ + { + "path": "src/core/tools/mcp-contract.mjs", + "line": 24, + "sha256": "9f534148fcbf1e6c016b9b6da70d9af2046ffa910dc4710b8a085c90c8c99dab", + "match": " if (!plain(value) || ++nodes > 256 || depth > 8 || Object.keys(value).some(key => !keywords.has(key))) throw mcpError('工具schema超出当前支持的有界JSON Schema范围。', 400);" + } + ], + "dependsOn": [], + "fixedInThisAudit": false + }, + { + "id": "I07", + "title": "检查点仍受 30/2000/64MiB 固定上限", + "priority": "P0", + "statusAtAudit": "confirmed-execution-cap", + "impact": "工具引用2000项、检查点30个或结果64MiB可阻断开放工程恢复。", + "nextChange": "采用分段索引、流式校验与分页恢复,去掉开放模式固定额度。", + "acceptance": "跨各旧边界建立和恢复检查点,旧工具/结果原 SHA 与归属完整。", + "observations": [ + "O15" + ], + "references": [ + { + "path": "src/core/tasks/task-checkpoint-manager.mjs", + "line": 15, + "sha256": "2f50f9de2e6b8c90376028ed2814eb795633e1ea917ab944a977299220ad820f", + "match": "export const TASK_CHECKPOINT_LIMITS = Object.freeze({ checkpoints: 30, tools: 2000, resultBytes: 64 * 1024 * 1024 });" + } + ], + "dependsOn": [], + "fixedInThisAudit": false + }, + { + "id": "I08", + "title": "能力/权限说明与实际行为不一致", + "priority": "P1", + "statusAtAudit": "confirmed-display-drift", + "impact": "结构化参数被说成拒绝、MCP仍提示确认、require-match显示暂停但开放模式继续。", + "nextChange": "展示声明能力、本地实现、实际权限/身份策略与可选覆盖;同步模型提示。", + "acceptance": "ask/auto/full 与完整参数/身份原件在网页和桌面一致;不恢复强制拦截。", + "observations": [ + "O05", + "O16", + "O20" + ], + "references": [ + { + "path": "src/shared/contracts/model-parameter-profiles.mjs", + "line": 81, + "sha256": "1264a90ba612b721ea639206cd6d8f595309c7bda38f80c45bb13c78af4a2d1f", + "match": " { name: '结构化输出约束', upstream: states[connection.upstreamModel?.features?.structured_outputs ?? 'unknown'], local: '尚未接入,参数保存时明确拒绝' }," + }, + { + "path": "src/core/tools/mcp-contract.mjs", + "line": 74, + "sha256": "9f534148fcbf1e6c016b9b6da70d9af2046ffa910dc4710b8a085c90c8c99dab", + "match": "export const mcpToolDefinitions = [{ type: 'function', function: { name: 'call_registered_mcp', description: '请求用户确认后调用本次目录的已启用MCP工具。先核对固定程序/实际握手/schema,不自动重试;完整结果按工件读取。'," + }, + { + "path": "src/client/ModelIdentityDetails.tsx", + "line": 13, + "sha256": "3e8bd9f735e21cc62abf217fc5f184c04014ea246d45115c42a95616248ac40b", + "match": "
处理方式
{identity?.policy === 'require-match' ? '不一致或无法核对时暂停' : '记录原结果和型号差异'}
" + }, + { + "path": "src/server/provider.mjs", + "line": 222, + "sha256": "9653f4318df2ed3b0570d58fda1d8cda6346dd2b6a0b016a4405360458bbc015", + "match": " if (config.executionMode === 'unrestricted' || !identity || identity.policy !== 'require-match' || modelIdentityAccepted(identity)) return;" + } + ], + "dependsOn": [], + "fixedInThisAudit": false + }, + { + "id": "I09", + "title": "Messages 块/工具缓存注解丢失", + "priority": "P1", + "statusAtAudit": "confirmed-adapter-gap", + "impact": "顶层 cache_control 能透传,但系统/内容块和工具上的缓存边界被转换丢掉。", + "nextChange": "增加原生块/工具元数据保留及供应商缓存控制,按官方层级/TTL处理。", + "acceptance": "工具/系统/用户块 cache_control 经过请求与历史往返不变,缓存读写由上游 usage 证明。", + "observations": [ + "O06" + ], + "references": [ + { + "path": "src/shared/contracts/anthropic-messages.mjs", + "line": 48, + "sha256": "ed0341d03ccffe73b1fb1fb9630ddac691a5a6ab36318a4fcc63ae689398001b", + "match": " const definitions = (tools ?? []).map(tool => {" + } + ], + "dependsOn": [], + "fixedInThisAudit": false + }, + { + "id": "I10", + "title": "Kimi cache-write 未归一展示", + "priority": "P1", + "statusAtAudit": "confirmed-usage-gap", + "impact": "原 usage 还在,但归一统计缺 cache_write_tokens/uncached,不能正确比较缓存成本。", + "nextChange": "提供带供应商语义/来源的读、写、未缓存、推理 token 统计,不补造费用。", + "acceptance": "三个互斥输入类别与总量相符;流式最终用量/unknown/raw保持。", + "observations": [ + "O07" + ], + "references": [ + { + "path": "src/core/models/protocols.mjs", + "line": 19, + "sha256": "0437d536c49d05651928a705bb6d92d5c76c6e1ce9f1a39121a6273cac873b27", + "match": " const cached = raw.cache_read_input_tokens ?? 0, created = raw.cache_creation_input_tokens ?? 0;" + } + ], + "dependsOn": [], + "fixedInThisAudit": false + }, + { + "id": "I11", + "title": "亲和仅绑定 Run,缺会话缓存与状态连续性", + "priority": "P1", + "statusAtAudit": "confirmed-routing-gap", + "impact": "续接新Run可换连接/型号;跨型号历史与DeepSeek reasoning_content不能视为可无损互换。", + "nextChange": "会话/Task级可配置亲和和供应商状态绑定,用户换模显式记录影响。", + "acceptance": "同会话续接、停止恢复、失效换线、手选换模分别核对模型/历史/缓存与状态。", + "observations": [ + "O09" + ], + "references": [ + { + "path": "src/core/models/model-hub.mjs", + "line": 326, + "sha256": "630c276a445641e3abaf920957e01d6ffd1b50f8264c0227ee55d29073d9dbb9", + "match": " const remembered = routing.sticky && this.sticky.get(runId);" + } + ], + "dependsOn": [], + "fixedInThisAudit": false + }, + { + "id": "I12", + "title": "前缀观察没有供应商缓存相关参数指纹", + "priority": "P1", + "statusAtAudit": "confirmed-observation-gap", + "impact": "改effort仍显示same_prompt;这个观察不能解释供应商实际缓存命中。", + "nextChange": "保留原有序提示观察,另记录缓存相关原生参数/型号/版本/工具/TTL指纹。", + "acceptance": "参数变化单列原因;不把字节共前缀或相同指纹当缓存命中。", + "observations": [ + "O19" + ], + "references": [ + { + "path": "src/core/context/prefix-audit.mjs", + "line": 30, + "sha256": "69f9407893c67bee6c5682666122cf6105ca347a8413e5b1122283481df0d6e2", + "match": " return { schemaVersion, method: 'ordered-wire-prompt-fields-utf8', upstreamCache: 'unknown', variants: variants(body).map(current => {" + } + ], + "dependsOn": [], + "fixedInThisAudit": false + }, + { + "id": "I13", + "title": "请求账本显示100/宣称1000,无法完整分页", + "priority": "P1", + "statusAtAudit": "confirmed-observability-gap", + "impact": "开放模式已保存1001以上尝试,但视图仅100且宣称保留1000,历史难核查。", + "nextChange": "完整游标检索/导出和准确保存说明;逐请求存储替代整数组回写。", + "acceptance": "所有历史尝试可分页/导出且归属、usage、unknown、raw SHA不变。", + "observations": [ + "O10" + ], + "references": [ + { + "path": "src/core/models/model-hub.mjs", + "line": 148, + "sha256": "630c276a445641e3abaf920957e01d6ffd1b50f8264c0227ee55d29073d9dbb9", + "match": " storage: { credentials: this.credentialVault ? 'memory-or-windows-user-dpapi' : 'memory', restart: '临时凭据在宿主退出后需要重新填写;已明确本机加密保存的连接按用户与原配置加载。', attemptsRetained: 1000 } }; }" + } + ], + "dependsOn": [], + "fixedInThisAudit": false + }, + { + "id": "I14", + "title": "官方认证/版本/端点策略未形成完整契约", + "priority": "P1", + "statusAtAudit": "partial-implementation", + "impact": "仅Bearer/x-api-key;anthropic-version写死2023-06-01,缺版本/beta/额外头策略。旧版本失效未被证明。", + "nextChange": "供应商、传输协议、官方API版本和代理覆盖分别配置并记录实际wire。", + "acceptance": "官方示例契约与版本/beta场景可复现;自定义端点原值与原错误保持。", + "observations": [], + "references": [ + { + "path": "src/server/provider.mjs", + "line": 186, + "sha256": "9653f4318df2ed3b0570d58fda1d8cda6346dd2b6a0b016a4405360458bbc015", + "match": " headers['anthropic-version'] = '2023-06-01';" + } + ], + "dependsOn": [], + "fixedInThisAudit": false + }, + { + "id": "I15", + "title": "模型发现不等于完整官方能力注册表", + "priority": "P1", + "statusAtAudit": "partial-implementation", + "impact": "目录容量/功能字段不完整;OpenAI Models类型无容量,Gemini nextPageToken未识别。", + "nextChange": "自动发现+官方版本化能力表+上游声明+显式覆盖,逐字段记录来源和冲突。", + "acceptance": "多页/多型号/失效/无元数据/代理差异不覆盖用户字段;不从型号子串猜能力。", + "observations": [ + "O12" + ], + "references": [ + { + "path": "src/shared/contracts/model-catalog.mjs", + "line": 19, + "sha256": "abb7b41982d072ecdb8c8d7c2feb67228eeb03c751c3f10b8557ac12b25022e1", + "match": "export function catalogModel(item, config = {}) {" + }, + { + "path": "src/shared/contracts/model-capabilities.mjs", + "line": 1, + "sha256": "a71775c388d1afc5dfc9057528502882361694c6d9fb541e6f55257d0ca5686a", + "match": "export const MODEL_PROTOCOLS = ['chat-completions', 'responses', 'anthropic-messages'];" + } + ], + "dependsOn": [], + "fixedInThisAudit": false + }, + { + "id": "I16", + "title": "缺 Gemini 原生协议", + "priority": "P1", + "statusAtAudit": "not-implemented", + "impact": "能解析类似目录不等于支持generateContent/streamGenerateContent和原生内容/思考状态。", + "nextChange": "另建官方Gemini适配器,取得可读官方schema后固定契约来源。", + "acceptance": "发现、分页、认证、流式、工具、多模态、状态、usage、原错误均有隔离与真实证据。", + "observations": [], + "references": [ + { + "path": "src/shared/contracts/model-capabilities.mjs", + "line": 1, + "sha256": "a71775c388d1afc5dfc9057528502882361694c6d9fb541e6f55257d0ca5686a", + "match": "export const MODEL_PROTOCOLS = ['chat-completions', 'responses', 'anthropic-messages'];" + } + ], + "dependsOn": [ + "I14", + "I15" + ], + "fixedInThisAudit": false + }, + { + "id": "I17", + "title": "缺云平台原生认证与路由", + "priority": "P1", + "statusAtAudit": "not-implemented", + "impact": "不能把Bedrock/Vertex/Foundry各种认证或部署ID当普通Bearer/v1。部分Azure兼容端点可能可用但未验。", + "nextChange": "按官方云平台分别实现身份提供器/区域/部署/签名/续期/错误,不复制存储真实Key到研究材料。", + "acceptance": "每平台固定官方契约与独立真实样本;代理转换路线单独标识。", + "observations": [], + "references": [ + { + "path": "src/server/provider.mjs", + "line": 189, + "sha256": "9653f4318df2ed3b0570d58fda1d8cda6346dd2b6a0b016a4405360458bbc015", + "match": " } else if (key) headers.Authorization = 'Bearer ' + key;" + } + ], + "dependsOn": [ + "I14", + "I15" + ], + "fixedInThisAudit": false + }, + { + "id": "I18", + "title": "服务端工具与官方停止/续接语义缺项", + "priority": "P1", + "statusAtAudit": "not-implemented-and-rejected", + "impact": "Claude server_tool_use/pause_turn和Responses多种服务端输出被本地协议拒绝。", + "nextChange": "原生服务端工具、结果块、签名、停止原因和继续请求独立适配,保留原响应。", + "acceptance": "官方块/SSE事件/暂停续接/usage/错误完整,不把未知块或工具结果当本地执行成功。", + "observations": [ + "O11" + ], + "references": [ + { + "path": "src/core/models/protocols.mjs", + "line": 120, + "sha256": "0437d536c49d05651928a705bb6d92d5c76c6e1ce9f1a39121a6273cac873b27", + "match": " } else if (item.type !== 'reasoning') throw protocolFailure('Responses包含当前尚未接入的服务端工具或输出类型:' + String(item.type ?? '未标注').slice(0, 100) + '。');" + }, + { + "path": "src/core/models/messages-stream.mjs", + "line": 11, + "sha256": "0631d37f3c7c4b498e7878673826d484f6ff0d8bf972d3fc266683feaa1e20b8", + "match": " consume(event) {" + } + ], + "dependsOn": [ + "I14", + "I15" + ], + "fixedInThisAudit": false + }, + { + "id": "I19", + "title": "原生工具选择与用户服务端工具不可配置", + "priority": "P1", + "statusAtAudit": "not-implemented", + "impact": "tools/tool_choice为托管字段,普通原生JSON无法开启官方web/computer等能力或选择官方tool_choice。", + "nextChange": "区分工作台工具与供应商原生工具,提供明确选择和实际调用记录。", + "acceptance": "用户选择在wire准确出现;本地/远端结果归属分开;禁止盲目并行应用多个候选。", + "observations": [], + "references": [ + { + "path": "src/shared/contracts/native-parameters.mjs", + "line": 10, + "sha256": "f065908247826464282ba50b34938608120c8dd740b16a74592ab342cdeb8e6b", + "match": " const managed = ['model', 'messages', 'input', 'system', 'tools', 'tool_choice', 'stream', 'apiKey', 'api_key', 'baseUrl', 'authorization', 'headers'];" + } + ], + "dependsOn": [ + "I18" + ], + "fixedInThisAudit": false + }, + { + "id": "I20", + "title": "PDF/音频/视频/文件等原生内容缺项", + "priority": "P1", + "statusAtAudit": "not-implemented-beyond-basic-image", + "impact": "当前截图/PNG基本视觉不是全球模型全多模态;JPEG被转PNG,缺官方原生文件生命周期。", + "nextChange": "按供应商内容格式实现附件/原文件/媒体/引用/返回工件,不把PNG通路当所有视觉支持。", + "acceptance": "原格式/用途/上传删除/尺寸/流式输出/错误来源与真实模型验证逐项完整。", + "observations": [], + "references": [ + { + "path": "src/core/context/input-attachments.mjs", + "line": 3, + "sha256": "64b84482aa71b03185813cddb1ad5c46516054e14da1d28326a0d88ee46f66d4", + "match": "import { PNG } from 'pngjs';" + } + ], + "dependsOn": [ + "I14", + "I15" + ], + "fixedInThisAudit": false + }, + { + "id": "I21", + "title": "Chat 多候选非流式/流式处理不一致", + "priority": "P1", + "statusAtAudit": "confirmed-contract-gap", + "impact": "非流式留raw多choices但只消费第一个;流式非0候选本地拒绝。O18首推断已单独纠正。", + "nextChange": "定义完整保留与候选选择工作流;工程执行一次只用明确选中的完整候选。", + "acceptance": "n>1两模式原结果保持、用户/自动选择一致;不盲目执行各候选的工具。", + "observations": [ + "O18", + "O18-C" + ], + "references": [ + { + "path": "src/core/models/model-stream.mjs", + "line": 86, + "sha256": "e354ae9fd96481c31b6f16cca89f4f7f7436449cedbdb2b879422e3c3d57d0cb", + "match": " if (choice.index !== 0) throw protocolFailure('当前只接入单个Chat候选流。');" + }, + { + "path": "src/server/provider.mjs", + "line": 386, + "sha256": "9653f4318df2ed3b0570d58fda1d8cda6346dd2b6a0b016a4405360458bbc015", + "match": " const received = result.choices?.[0]?.message;" + } + ], + "dependsOn": [], + "fixedInThisAudit": false + }, + { + "id": "I22", + "title": "终端缺交互PTY/输入会话", + "priority": "P1", + "statusAtAudit": "not-implemented", + "impact": "已有所属命令、完整日志/停止/Git证据,但pipes一次性输入不能替代交互终端。", + "nextChange": "增加会话PTY、stdin、resize、进程恢复与工具接口,沿用户权限直接执行。", + "acceptance": "长命令、交互输入、停止、重新附着、退出码与Git证据真实可查。", + "observations": [], + "references": [ + { + "path": "src/core/owned-command.mjs", + "line": 13, + "sha256": "40113f38ef1fa629540d31753828b92792a133f7aab4b0c9253be12d568704e3", + "match": " detached: process.platform !== 'win32', stdio: [input === null ? 'ignore' : 'pipe', 'pipe', 'pipe'] });" + } + ], + "dependsOn": [], + "fixedInThisAudit": false + }, + { + "id": "I23", + "title": "缺通用电脑操作能力", + "priority": "P2", + "statusAtAudit": "not-implemented", + "impact": "已有文件对话框/浏览器动作,不是任意Windows应用/全桌面computer use。", + "nextChange": "支持截图、坐标/语义动作、键盘/窗口/应用操作,与选定权限和任务归属连接。", + "acceptance": "真实多应用闭环及取消/恢复可观察;不能用本地浏览器DOM测试冒充OS控制。", + "observations": [], + "references": [ + { + "path": "src/core/desktop/desktop-bridge.mjs", + "line": 9, + "sha256": "bcfc675d5448e996f8e8156146ccdbf2c122febfe8a3e338e9887afd8c826541", + "match": "export class DesktopBridge {" + } + ], + "dependsOn": [], + "fixedInThisAudit": false + }, + { + "id": "I24", + "title": "Worker要求语义仍未评估", + "priority": "P1", + "statusAtAudit": "partially-implemented-benefit-unknown", + "impact": "角色/模型/成果与检查身份已绑定,但每条requirements仍为not_evaluated,不能认证角色质量。", + "nextChange": "逐要求关联实际差异/测试/观察,语义评估与确定性结果分列。", + "acceptance": "未测/失败/证据不足不通过;真实同题角色对照才证明收益。", + "observations": [], + "references": [ + { + "path": "src/core/tasks/worker-evidence.mjs", + "line": 49, + "sha256": "8c077ad19a0f2391c83971ae76ac5079d110ce90fed49440fbf4139636d2a014", + "match": " requirements: [worker.goal, ...worker.assignment.requirements].map(criterion => ({ criterion, status: 'not_evaluated', source: 'sealed_assignment' }))," + } + ], + "dependsOn": [], + "fixedInThisAudit": false + }, + { + "id": "I25", + "title": "Worker没有skills/MCP能力装配", + "priority": "P1", + "statusAtAudit": "confirmed-assembly-gap", + "impact": "父代理目录不能自动带给子任务,某些专业任务实际没有必要工具。", + "nextChange": "角色/子任务按需求装配继承资源、身份与作用范围,不默认要求额外审查轮。", + "acceptance": "子任务实际工具表/来源/SHA与请求相符,未装配原因显示,full流程无额外确认。", + "observations": [], + "references": [ + { + "path": "src/server/runtime.mjs", + "line": 461, + "sha256": "e1b349263eac824ccbda5dda713cdce7e3567d7dbc0a3613883c5e53390030c2", + "match": " const skillCatalog = run.worker ? null : this.skillRegistry().catalog(session.projectId);" + } + ], + "dependsOn": [], + "fixedInThisAudit": false + }, + { + "id": "I26", + "title": "长历史读写与响应内存规模风险", + "priority": "P2", + "statusAtAudit": "static-risk-not-benchmarked", + "impact": "启动全量加载,会话/尝试整份回写,长raw响应内存累积;未测到确定故障。", + "nextChange": "先测量,再增量事件/分页/流式工件,避免用截断历史或固定额度解决规模问题。", + "acceptance": "可复现增长曲线、RSS/时延/恢复和原字节核对;报告机器与数据规模。", + "observations": [], + "references": [ + { + "path": "src/core/storage/state-database.mjs", + "line": 55, + "sha256": "2a38f7564df44c4ee5183d850064be1d46fba7e072e2b87f73afb8e1a85a1517", + "match": " writeSession(session) {" + }, + { + "path": "src/core/models/model-hub.mjs", + "line": 40, + "sha256": "630c276a445641e3abaf920957e01d6ffd1b50f8264c0227ee55d29073d9dbb9", + "match": " if (recovered) this.saveAttempts();" + }, + { + "path": "src/server/provider.mjs", + "line": 123, + "sha256": "9653f4318df2ed3b0570d58fda1d8cda6346dd2b6a0b016a4405360458bbc015", + "match": " const parts = []; let length = 0;" + } + ], + "dependsOn": [], + "fixedInThisAudit": false + }, + { + "id": "I27", + "title": "缺仓库持续集成工作流", + "priority": "P1", + "statusAtAudit": "not-found-in-tracked-files", + "impact": "本地QA丰富但版本/适配器/包校验不能稳定由远端流水线复现。", + "nextChange": "在GitAtlas支持环境建立可重复的源码/契约/构建/安装验证任务;真实供应商另受控运行。", + "acceptance": "干净环境按锁定依赖完整执行并保留首失败/版本/source SHA;不将需浏览器/Python环境失败改绿。", + "observations": [], + "references": [], + "dependsOn": [], + "fixedInThisAudit": false + }, + { + "id": "I28", + "title": "供应商契约测试覆盖与官方更新闭环缺口", + "priority": "P1", + "statusAtAudit": "verification-gap", + "impact": "当前测试未覆盖这次发现的嵌套丢失/新推理模式/覆盖等官方契约。", + "nextChange": "官方示例/schema驱动契约夹具,版本差异/弃用更新进入能力注册表与回归。", + "acceptance": "固定来源/原字节/协议/schema/wire/真实批次相互关联;本机夹具不当真实服务验收。", + "observations": [], + "references": [ + { + "path": "docs/QA-NEWAPI-COMPATIBILITY-REVIEW.md", + "sha256": "db859bfb8ce8751dda8766037b8bbf153fa7fceb6beb581bc5899ab0f8f4326d", + "evidenceKind": "reused-historical-QA-not-rerun" + }, + { + "path": "docs/QA-CLAUDE-OFFICIAL-CAPACITY.md", + "sha256": "7d0c2036a5f1a7578ccc820e3f2c733a9b602e825fe661800c8b2692a72a0230", + "evidenceKind": "reused-historical-QA-not-rerun" + } + ], + "dependsOn": [ + "I01", + "I02", + "I03", + "I14", + "I15" + ], + "fixedInThisAudit": false + }, + { + "id": "I29", + "title": "三项已推送修复尚未正式发行", + "priority": "P0", + "statusAtAudit": "source-fixed-not-deployed", + "impact": "正式1.25.12仍是F旧源码;目录刷新/wire截断说明/官方Claude容量没有进入正式程序。", + "nextChange": "完成首批确认修复后冻结独立发行,避免逐小修反复打包;安装再做数据保持核对。", + "acceptance": "候选源码/build/wire/UI/安装身份一一对应,再部署与正式回看;源码包不包含用户项目。", + "observations": [], + "references": [ + { + "path": "docs/QA-ROLE-ROUTING-F.md", + "sha256": "a7fd3e08e44d92a4b0f71474639c63ed9578b0f9c57364e06f20ba0b43c50b05", + "evidenceKind": "reused-historical-QA-not-rerun" + }, + { + "path": "docs/QA-NEWAPI-COMPATIBILITY-REVIEW.md", + "sha256": "db859bfb8ce8751dda8766037b8bbf153fa7fceb6beb581bc5899ab0f8f4326d", + "evidenceKind": "reused-historical-QA-not-rerun" + }, + { + "path": "docs/QA-CLAUDE-OFFICIAL-CAPACITY.md", + "sha256": "7d0c2036a5f1a7578ccc820e3f2c733a9b602e825fe661800c8b2692a72a0230", + "evidenceKind": "reused-historical-QA-not-rerun" + } + ], + "dependsOn": [ + "I01", + "I02", + "I03", + "I05", + "I06", + "I07", + "I08" + ], + "fixedInThisAudit": false + }, + { + "id": "I30", + "title": "当前规划/能力文档入口停留D/E旧时点", + "priority": "P1", + "statusAtAudit": "confirmed-documentation-drift", + "impact": "F已交付却仍有当前页说未实现,读者难判断源码和部署状态。", + "nextChange": "本轮统一当前指针,保留日期明确的历史,不把规划等于完成。", + "acceptance": "AGENTS/CAPABILITIES/NEXT-STEPS/PRODUCT-PLAN/RESEARCH-ROADMAP当前入口一致可点击。", + "observations": [], + "references": [], + "dependsOn": [], + "fixedInThisAudit": true, + "statusAfterAudit": "current-entry-pointers-corrected-documents-only" + }, + { + "id": "I31", + "title": "跨产品原生导出/入站委托服务缺项", + "priority": "P2", + "statusAtAudit": "not-implemented", + "impact": "Mnemo目前只有来源包只读核对与反馈;没有运行中的导出/委托/撤销闭环。", + "nextChange": "按实际双方契约实现原字节导出、单Run/多Run归属与委托状态,不自动继承用户授权。", + "acceptance": "双方合成原件和独立真实服务验证;失败、容量缺项、撤销和旧原件完整。", + "observations": [], + "references": [ + { + "path": "docs/THREE-SYSTEM-INTERFACE-FEEDBACK-2026-11-06.md", + "sha256": "ebfd0fdf5242bdcfb07282001107f41c74c984b0cef2117b1a2bb0fd79bcf1fc", + "evidenceKind": "reused-historical-QA-not-rerun" + } + ], + "dependsOn": [], + "fixedInThisAudit": false + }, + { + "id": "I32", + "title": "新物理手机完整流程未验收", + "priority": "P2", + "statusAtAudit": "unverified-real-device", + "impact": "已有Companion实现与本地契约验收不能代替当前真实手机端到端。", + "nextChange": "独立配对/查看/选择权限/批准/停止/断线/撤销真实设备样本。", + "acceptance": "保存设备/宿主/浏览器版本和实际动作,无任务或凭据跨错归属。", + "observations": [], + "references": [ + { + "path": "src/core/companion/device-registry.mjs", + "line": 12, + "sha256": "2fe69e47a7acf303dfcf33697b667bae9ea53c7cd17e6830f9c607a7f0222537", + "match": "export class DeviceRegistry {" + } + ], + "dependsOn": [], + "fixedInThisAudit": false + }, + { + "id": "I33", + "title": "Companion操作账本达到2000后拒绝", + "priority": "P1", + "statusAtAudit": "source-confirmed-execution-cap-not-live-triggered", + "impact": "累计操作达到2000后手机操作409,并非开放模式动态账本。", + "nextChange": "分页持久化幂等操作,移除固定总数拒绝和load格式上界,不丢旧记录。", + "acceptance": "跨2000边界继续操作,幂等/unknown/断线恢复与原行保持。", + "observations": [], + "references": [ + { + "path": "src/core/companion/device-registry.mjs", + "line": 124, + "sha256": "2fe69e47a7acf303dfcf33697b667bae9ea53c7cd17e6830f9c607a7f0222537", + "match": " if (Object.keys(state.operations).length >= 2000) throw companionError('手机操作账本达到上限,请在电脑处理。', 409);" + } + ], + "dependsOn": [], + "fixedInThisAudit": false + }, + { + "id": "I34", + "title": "全球官方与代理真实兼容没有完整证明", + "priority": "P1", + "statusAtAudit": "unverified-beyond-recorded-samples", + "impact": "AD/AE有DeepSeek实际成功,也有Claude400/目录断连;不能据此宣称全球全部可用。", + "nextChange": "按正式新版本另冻官方优先的供应商样本与代理样本;旧样本不重写/续跑。", + "acceptance": "每连接发现/参数/流式/工具/媒体/usage/错误/真实工程独立分项,未知保留。", + "observations": [], + "references": [ + { + "path": "docs/QA-LIVE-VALIDATION-AD.md", + "sha256": "bbaf225418099a9fcda31f33de8911ab727f0a359d40290758bbf2534ad4b29e", + "evidenceKind": "reused-historical-QA-not-rerun" + }, + { + "path": "docs/QA-LIVE-ENGINEERING-AE.md", + "sha256": "cbf0372c0dcebce0d9b6ffe6c5536c504993cbb191cc6cd0058172e765fc99d1", + "evidenceKind": "reused-historical-QA-not-rerun" + } + ], + "dependsOn": [ + "I28", + "I29" + ], + "fixedInThisAudit": false + }, + { + "id": "I35", + "title": "重复读取/缓存/经验/角色普遍收益未证明", + "priority": "P1", + "statusAtAudit": "unverified-benefit", + "impact": "D/E/F机制已部署,单例缓存与工具事实不证明广泛因果收益。", + "nextChange": "分别定义同题基线/干预、稳定线路/模型/权限、真实raw用量与质量指标。", + "acceptance": "成功率、重复读取、cache read/write、tokens/时延和失败原因同时报告;不得以提醒触发当业务完成。", + "observations": [], + "references": [ + { + "path": "docs/QA-REMINDER-READBACK-D.md", + "sha256": "07282464b38040154c754a44bf12597cea74419fba257a69c341f09186c2efbf", + "evidenceKind": "reused-historical-QA-not-rerun" + }, + { + "path": "docs/QA-EXPERIENCE-DRAFTS-E.md", + "sha256": "157bf3a83fcb3ec78d4f6ccbb55e4f2cc6d1ac8407fb504e06dedd4b6009d530", + "evidenceKind": "reused-historical-QA-not-rerun" + }, + { + "path": "docs/QA-ROLE-ROUTING-F.md", + "sha256": "a7fd3e08e44d92a4b0f71474639c63ed9578b0f9c57364e06f20ba0b43c50b05", + "evidenceKind": "reused-historical-QA-not-rerun" + } + ], + "dependsOn": [ + "I09", + "I10", + "I11", + "I12", + "I24", + "I25", + "I34" + ], + "fixedInThisAudit": false + }, + { + "id": "I36", + "title": "Purge AB/BA收益仍unknown", + "priority": "P1", + "statusAtAudit": "unverified-benefit-network-confounded", + "impact": "AG四组只有ba-off自主完成,其余三个ETIMEDOUT/unknown,不能做收益比较。", + "nextChange": "供应商兼容与基线真实交付稳定后,新冻结原版Purge AB/BA,身份和输入严格固定。", + "acceptance": "插件确实加载/边界触发/对照可比/工程质量及reported和unknown单列;旧AG失败原样保留。", + "observations": [], + "references": [ + { + "path": "docs/QA-LIVE-PURGE-AG.md", + "sha256": "9bddff27433f9948dffa337c67814287a2b1282602c62e8846daea18bf4457c4", + "evidenceKind": "reused-historical-QA-not-rerun" + } + ], + "dependsOn": [ + "I34", + "I35" + ], + "fixedInThisAudit": false + }, + { + "id": "I37", + "title": "批处理/后台会话/文件与其他模型资源API缺项", + "priority": "P2", + "statusAtAudit": "not-implemented-as-complete-workflow", + "impact": "原生字段能部分转交不等于支持异步任务轮询/取消、批次/文件/向量/音视频图像资源全生命周期。", + "nextChange": "按工程优先级通过供应商适配器逐类扩展,显示实际已实现能力;无需复制网关账户/计费后台。", + "acceptance": "资源创建/使用/检索/取消/清理、返回状态/工件与费用来源可核对,异步响应不误当工具完成。", + "observations": [], + "references": [ + { + "path": "src/shared/contracts/model-capabilities.mjs", + "line": 1, + "sha256": "a71775c388d1afc5dfc9057528502882361694c6d9fb541e6f55257d0ca5686a", + "match": "export const MODEL_PROTOCOLS = ['chat-completions', 'responses', 'anthropic-messages'];" + }, + { + "path": "src/core/models/protocols.mjs", + "line": 12, + "sha256": "0437d536c49d05651928a705bb6d92d5c76c6e1ce9f1a39121a6273cac873b27", + "match": "export { protocolRoute } from '../../shared/contracts/model-endpoint.mjs';" + } + ], + "dependsOn": [ + "I14", + "I15", + "I18", + "I20" + ], + "fixedInThisAudit": false + }, + { + "id": "I38", + "title": "加密凭据元数据超过101项被整体拒加载", + "priority": "P1", + "statusAtAudit": "confirmed-load-count-cap", + "impact": "102项保存登记会标记unavailable,开放连接数与本机保存加载范围不一致;原metadata不删除。", + "nextChange": "按版本/逐记录有效性和分页加载凭据登记,移除固定连接总数格式门槛。", + "acceptance": "102以上自有合成/加密测试登记保存重启恢复;无旧条目损失、无真实Key访问,来源绑定仍可核对。", + "observations": [ + "O21" + ], + "references": [ + { + "path": "src/core/models/credential-vault.mjs", + "line": 20, + "sha256": "35420ba01e5ce49c8d5af6d63a51128701de80a7958d38132816c304f5895b63", + "match": " this.blocked = !state || state.schema !== 1 || !Number.isSafeInteger(state.revision) || !state.entries || typeof state.entries !== 'object' || Array.isArray(state.entries) || Object.keys(state.entries).length > 101;" + } + ], + "dependsOn": [], + "fixedInThisAudit": false + } + ], + "issueCounts": { + "total": 38, + "priority": { + "P0": 8, + "P1": 25, + "P2": 5 + }, + "documentsCorrectedHere": [ + "I30" + ], + "allOtherItemsRemainOpenOrUnverified": true + }, + "historicalOutcomes": [ + { + "sample": "W", + "qa": { + "path": "docs/QA-LIVE-DELIVERY-W.md", + "sha256": "447b6425f947b99b7e50f6b99d69c3d114a8c9fd697d05fbe93b1a1f92e4e69c", + "evidenceKind": "reused-historical-QA-not-rerun" + }, + "requests": 6, + "reportedTokens": 71044, + "outcome": "matched identity; reads only; no_progress; no proposal or delivery; outlier-v2 exclusion boundary not reached" + }, + { + "sample": "AD", + "qa": { + "path": "docs/QA-LIVE-VALIDATION-AD.md", + "sha256": "bbaf225418099a9fcda31f33de8911ab727f0a359d40290758bbf2534ad4b29e", + "evidenceKind": "reused-historical-QA-not-rerun" + }, + "requests": 21, + "reportedTokens": 199250, + "unknownRequests": 2, + "outcome": "10 prechecks 8 pass; two Claude missing max_tokens400; DeepSeek Node11 requests autonomous9tests/build/Git passed" + }, + { + "sample": "AE", + "qa": { + "path": "docs/QA-LIVE-ENGINEERING-AE.md", + "sha256": "cbf0372c0dcebce0d9b6ffe6c5536c504993cbb191cc6cd0058172e765fc99d1", + "evidenceKind": "reused-historical-QA-not-rerun" + }, + "requests": 20, + "reportedTokens": 367365, + "unknownRequests": 0, + "outcome": "React and Python delivered autonomously; original pytest/terminal-report gaps preserved and later separately repaired" + }, + { + "sample": "AG", + "qa": { + "path": "docs/QA-LIVE-PURGE-AG.md", + "sha256": "9bddff27433f9948dffa337c67814287a2b1282602c62e8846daea18bf4457c4", + "evidenceKind": "reused-historical-QA-not-rerun" + }, + "requests": 11, + "reportedTokens": 109888, + "unknownRequests": 3, + "outcome": "ba-off delivered; ab-off/ab-on/ba-on connect ETIMEDOUT; original Purge loaded in on groups; benefit unknown" + } + ], + "preservedConstraints": [ + "unrestricted/full/passthrough defaults and selected user policy", + "complete historical raw/evidence", + "unknown usage not imputed", + "original network failures", + "repeated-read observations", + "Purge benefit unknown", + "source package excludes user project data" + ] +} diff --git a/findings.md b/findings.md index 3d308ebe..df10ae1e 100644 --- a/findings.md +++ b/findings.md @@ -1,5 +1,7 @@ # 汇总发现 +最新审查入口(2026-11-06):docs/SYSTEM-AUDIT-OFFICIAL-FIRST-2026-11-06.md、docs/DEVELOPMENT-ROADMAP-OFFICIAL-FIRST.md;30区域/38事项、21编号观察和O18-C纠正,官方直连/兼容/实际供应商与工程证据分开。下方此前结果保留原时点;本轮产品源码与正式进程未改。 + Claude容量调查纠正:官方公开规格和代理实际能力是两个事实。“目录缺容量”不能推出“Claude最大输出未知”。当前官方四精确ID同步128K/上下文1M,Opus5.5的128000还有官方流式示例依据;300K是批量beta。新增静态官方规格后备只按精确ID,目录与显式原生优先,不裁上下文、不给Chat/Responses追加额度、不改原连接2048字段。UI最大值与实际请求值/来源分别显示。独立67/67及网页4组通过,正式未部署;来源直读的地域HTML、403与旧索引差异保留,不伪装成功来源。 New API兼容修复回顾:已有目录apply不能直接跳过,也不能为刷新事实增加与DPAPI凭据绑定的连接revision;仅替换upstreamModel对象并事务持久化,旧Run保持原快照。截断诊断必须依据实际序列化wire,而非笼统声称未附加输出额度;Responses的HTTP200/incomplete/max_output_tokens需独立分类。两处源码ae5123c5/341f5f69、独立候选126/126与完整build通过;首Responses夹具将payload.status误用作HTTP状态的失败保留,纠正后13/13。305正文/201定义覆盖及全部SHA/blob再次通过,缺项/版本矛盾/正式Claude旧2048与Purge unknown见docs/QA-NEWAPI-COMPATIBILITY-REVIEW.md;本轮无正式模型/Key/数据操作或部署。 @@ -1357,3 +1359,19 @@ F发行复核:大文件原240/300秒仅超时,原33文件/超过2MB断言独 F正式回顾:异步模型卡片必须等待真实字段再断言count,首失败保留而不重复安装。实际Claude连接没有同步最大容量,其旧声明2048会成为Messages必填输出值;修复缺失字段并不等同移除该协议输出参数,也不等同知道上游上限。须公开这个限制并独立调查容量来源,不能猜模型能力、改旧配置或声称真实线路已修。正式两端和最终原数据已再次exact,150份原证据封存,后续只提交记录/普通增量推送。 New API学习:305正文/201嵌入定义已分别审读;来源清单与逐篇批次独立保存。三原生协议/推理/工具/缓存语义不能混用,网关额度/计费不是工作台默认限制;插件指南与rc.40/43/47、旧账户接口与rc.34存在版本差异。生成API schema还含简化枚举、Responses用量示例与未解析管理引用。结论/差距见docs/NEWAPI-DOCUMENTATION-LEARNING.md。确认目录apply已有项continue未刷新、透传截断说明无视Messages/显式输出字段,独立修复后本机验证,不改正式配置。 + +## 官方优先全系统审查起点(2026-11-06) + +基线4da2550b;工作树干净。官方容量、目录刷新和wire截断说明三源码修复已推送,但仍未部署。New API仅提供网关与兼容参考;官方原生协议/真实线路/自主交付/Purge分别验收。首次OpenAI三页面403,后续取官方可读入口,不能把拒绝页当接口证据。大历史文档需分段,避免原件目录的重复源码污染当前源码清单。 + +当前源码清单272文件/31738行(含原版Purge锁清单),tests171文件、scripts272文件;清单是范围核对,不冒称每行人工审过。已沿原生请求/目录/能力/推理/流式/网络路径读取:仅三协议;认证仅Bearer及Messages x-api-key,API version写死2023-06-01;Gemini原生分页/生成、其他官方认证与完整内容块缺项。Claude官方原Models overview已明确后续型号不接受手动enabled预算,且有between_tools;当前表单仍按通用模式生成,须隔离复核。推理快捷控件整对象替换可能删除同对象其它原生字段。modelFeatureRows仍称结构化参数保存拒绝,与开放原生JSON路径需对照。文档CAPABILITIES/NEXT-STEPS头部仍E/D时点而实际F已交付,需统一当前入口。 + +只读正式health复核:2026-11-06T13:08:18Z,1.25.12/PID71628/ready=true;不据此宣称全部项目/凭据/工件再次exact。公共DeepSeek/Qwen/Z.ai/Kimi/Bedrock/Azure正文已取得,Google/Mistral/xAI初次公共页超时;不混入旧模型网络异常或改变其结论。 + +本轮离线观察 O01–O20 已完成,另以 O18-C 纠正原多候选流推断并保留原件。确认原生嵌套字段丢失、Chat stream_options 覆盖、官方推理模式未完整同步、块/工具缓存控制丢失、Kimi 写缓存未归一、自动池可选不支持工具的成员、Run 级亲和不能代表会话亲和、账本显示与完整读取缺口,以及开放模式仍生效的 MCP/检查点次数限制。O20 内存 Provider 观察证明 require-match 在 unrestricted 下继续记录 mismatch,而现有型号详情仍显示暂停;修复应如实展示实际策略,不恢复强制拦截。20 个观察是调查结果,不是 20 项证明产品正确的测试。 + +实际独立网页核对沿未改的官方容量候选:1280 下参数整对象丢失与非法通用模式可复现;390 下进一步检查真实控件/完整表单,文档宽 390,0 页面错误/真实模型请求。首次 390 截图只覆盖表头,不代替后续完整观察;不冒称正式 Electron 验收。官方 OpenAI 网页 403 后改取 openai/openai-python 固定 c6fc264689d384d5caec150fafa1a9567c4a3f56 的类型,类型可证明契约字段,不能代替未取得的型号容量/缓存 guide 或真实接入。 + +现有官方导出明确 Claude Opus 5.5 的 thinking 不能使用旧 enabled/disabled/between_tools;模型目录还含 disabled/between_tools/server_tool 子项而当前解析不全。DeepSeek 工具历史要求保留所有 reasoning_content;供应商缓存是 best effort,不能保证不掉缓存。各官方容量/推理/缓存/停止原因需带精确 ID、版本、协议、来源和观测时间,兼容网关信息不能升级成官方承诺。文档头部 D/E 已落后于正式 F,应增加当前入口并保留历史段落。 + +收尾新增O21:CredentialVault.reload对合成101条登记正常、102条整体blocked,原metadata不删除;不调用构造器/加解密/真实存储,0真实Key/请求/进程/正式写入。作为I38固定加载总数门槛加入首批修复,最终38事项/21编号观察+O18-C。源715输入SHA一致,24可取官方来源记录与17来源读取失败单列;30能力区域和M0–M8全部映射完成。七当前文档入口、AGENTS及规划指针已统一;历史报告不回写。 diff --git a/progress.md b/progress.md index 4ef4de22..7eb1be74 100644 --- a/progress.md +++ b/progress.md @@ -1,5 +1,7 @@ # 统一执行进度 +当前接续(2026-11-06):官方优先全层审查/路线已形成,38事项全部分配阶段;本轮正在完成证据封存、文档核对与限定提交推送。未改产品源码/正式数据、真实Key、模型请求或部署。下方每轮结果按原时点保留。 + Claude四阶段完成并推送:源码/来源2bd7b24a、纠正QA b59f6196与只读协调1b2ea607已普通推送,远端逐SHA一致,回执docs/gitatlas-claude-capacity-push.json随收尾提交保存。原源码输入摘要再次一致;本轮67/67和网页4组已完成,未部署或请求真实模型。Mnemo两文档SHA只读记录,不改变门槛,不代验176/33或发信。 Claude纠正:官方目标Models overview已取得,claude-opus-5-5同步128000/上下文1000000有明确来源;旧代理目录缺项不能当官方未知。源码2bd7b24a、官方来源/构建输入绑定、独立67/67及网页4组/两视口实看通过;原0/5、403/地域提示/索引差异与无效读取保持。原生参数、目录优先、本地透传不变,未发行/正式部署/模型请求。当前整理记录并按持续授权推送;同期协调消息仅只读Mnemo新增文档,不扩大入站或运行许可。 @@ -2025,3 +2027,11 @@ F正式交付收口:唯一安装239133ms保留PID71628/Provider/原instance; F源码与交付记录828208a7已增量推送,GitAtlas同分支逐SHA一致。继续只读容量调查:匿名模型详情401、公开文档未取得具体型号容量;无凭据请求或真实模型。发现已有连接目录apply跳过元数据更新,及Messages截断沿旧“未附加输出额度”通用说明,均作为下一独立切片保留,未倒改F已冻结源码或已封存150份原件。最后提交回执/接续并再推送核对。 New API研究阶段一234b76cc已提交;阶段二逐篇305正文/201接口全部审读,来源SHA与接口Git blob再校验通过。阅读覆盖JSON含批次/实际笔记,图片不冒称逐像素验收。输出截断/猜测只读路径不存在和一次未匹配计划patch保持为调查失误,无正式写入。准备提交阶段二,随后独立修确认缺陷;0真实模型请求,不读Key,不操作正式进程。 + +官方优先审查开始:已恢复4da2550b干净工作树、执行规范和历史QA;planning-with-files catchup无额外输出,正确OpenAI Docs技能已读。创建独立审查计划并追加历史规划文件;首次三个OpenAI官方页面HTTP403保持为来源失败,改用可读官方入口。当前仅文档/只读调查,无真实模型、Key、正式数据或部署改变。 + +官方优先审查覆盖完成:清点当前源码/测试/脚本并沿全层关键入口读取;保存 19 项离线观察、1280 与 390 独立实际页面观察,沿既有 New API 305/201 与 Claude 原件复用,不重复全量采集。新增内存核对 O18-C/O20 退出 0,原 O18 推断保留且单独纠正;共 20 个编号观察,不等于产品全绿。OpenAI 官方网页仍不可取得,固定官方 SDK 五公共文件成功;其它官方正文与超时分别保存。MCP 注册超时/实际开放超时区别已纠正。 + +正在整理官方来源账本、分层能力/问题矩阵及可执行路线。正式只读 health 仍只证明观察时点 1.25.12/PID71628 ready;本轮未读取 Key、请求真实模型、修改产品源码/正式数据、发行或重启。接续中的输出截断/猜测路径失败记录于审查计划,均非产品失败。 + +审查组装首轮因不存在的protocols图像锚点失败,原assemble-round-1.log保留,改用真实input-attachments后第二轮通过;收尾O21合成元数据发现凭据登记101上界,不读取Key,第三轮38事项/30区域/22观察记录(21编号+纠正)/715输入SHA通过。已生成全层报告、官方来源账本/机器矩阵和M0–M8路线,38项验收全部映射;七当前文档/AGENTS/规划入口已统一。正封存.raw及四最终交付输入,验证后限定本地提交和既有GitAtlas分支普通增量推送。 diff --git a/task_plan.md b/task_plan.md index d4ead24c..85fbbda5 100644 --- a/task_plan.md +++ b/task_plan.md @@ -1,5 +1,7 @@ # Agent Workbench:核心代理能力研究与产品计划 +当前接续(2026-11-06):官方优先全层审查/路线已形成,30区域/38事项、21观察+O18-C、715输入SHA核对;见docs/SYSTEM-AUDIT-OFFICIAL-FIRST-2026-11-06.md和docs/DEVELOPMENT-ROADMAP-OFFICIAL-FIRST.md。阶段1–4完成,阶段5正在封存/文档核对及提交推送;正式1.25.12,三修复未部署,本轮0真实模型/Key/正式写入。下一M1参数保真、官方模式、残留上限与说明,不将审查当修复/发行完成。下方旧接续按历史时点保留。 + ## 当前:Claude官方容量纠正(2026-11-06) 官方已明确Opus5.5同步128K/上下文1M,先前只查代理目录而列未知是调查缺口。源码2bd7b24a已增加精确型号官方后备来源及一致UI说明;独立相关67/67、build-2与网页4组/两视口通过,原输入/显式参数/上游原错保持。四阶段完成,授权增量推送至1b2ea607、远端SHA一致,回执随收尾提交保存,尚未发行/部署。见docs/CLAUDE-OFFICIAL-CAPACITY-PLAN.md和QA-CLAUDE-OFFICIAL-CAPACITY.md;旧未知文字按历史时点保留。 @@ -1082,3 +1084,7 @@ F收尾:57d075e修正独立编译QA路径,6d09ac8封存差异;202610302320 用户要求完整学习 https://docs.newapi.pro/zh/docs。按docs/NEWAPI-DOCUMENTATION-REVIEW-PLAN.md建立逐篇清单与阅读笔记,再对照协议/模型同步/推理/缓存/错误/用量实现修复;先确认中文文章总量与公开Markdown实际覆盖,不以下载量冒充阅读量。当前正式仍1.25.12,研究不操作真实模型/Key/用户工程,不委派。 来源冻结234b76cc后,305正文/201接口已逐项读完并重核SHA/Git blob,API与Plugins覆盖完整;学习笔记和逐篇批次已保存。下一只修确认的已有连接目录刷新和实际wire截断说明,保持用户配置和原证据;Gemini原生/媒体不是仅靠目录条目就已经实现,作为差距记录。 + +## 当前:官方优先全系统审查(2026-11-06) + +见 `docs/OFFICIAL-FIRST-SYSTEM-AUDIT-PLAN.md`、全层审查与官方优先路线。基线4da2550b/干净且跟踪分支一致;近期三项源码修复已推送未部署。阶段1–4已完成715输入/全层关键路径/官方与兼容来源/21观察+O18-C/实际自有1280和390表单核对,30区域38事项已全部分配M0–M8。阶段5进行封存、文档/JSON/链接/diff核对与定向提交推送。不请求真实模型、不读Key、不重启/续跑正式工程、不委派,不改开放策略或历史未知结论。公共403/超时、输出截断/猜测路径和首次组装图像锚点错误作为调查过程保持。