Repository navigation
Recover the original Chat Todo and Monitor confirmations - #8725
loopx-agent wants to merge 359 commits into
Conversation
loopx-agent
left a comment
There was a problem hiding this comment.
Reviewer: model_agent · gpt-6.1-sol · OpenAI · runtime_reported · reasoning_effort=max
Request changes conclusion (author-owned PR; GitHub blocks formal self-review)
评审提交 a00b3dc0e1161f2bcce4f5e626ddc2597338de9a,目标分支基线 3347c7f8dbb0a20cfe5bda7bfdbb69af69c1ef0c。
[P2] 打包 Chat 将已提交的确认导向新提案。 在打包 Chat 确认创建 Todo。让真实 provider 提交后,第一次保存 todo_created checkpoint 时抛出 I/O 错误。 页面显示「未写入」,唯一可见的重试先生成新 proposal,再要求确认。实测新确认返回 todo_already_exists 和另一份 receipt。原确认仍能经 HTTP 取回 todo_added 和原回执。 创建和 checkpoint 失败分支
动机 / Motivation
这项改动面向在 Chat 或飞书确认创建 Todo 或 Monitor,并需在中断后恢复原结果的用户。 在同一中断条件下,目标基线的原确认会被当前 provider revision 判成 stale。 页面显示「未写入」,唯一可见的重试先生成新 proposal,再要求确认。实测新确认返回 todo_already_exists 和另一份 receipt。原确认仍能经 HTTP 取回 todo_added 和原回执。 候选直接重试同一 HTTP 确认返回原始 Todo 和回执。File、SQLite 中均只保留一个 Todo,provider revision 保持不变。 实测使用合成 Goal、真实本机 Chat HTTP 和 File/SQLite;飞书走模拟确认,PostgreSQL integration 跳过。已安装 App、真实飞书及托管 CI 未验证。
改动思路 / Approach
现有 Todo provider 拥有提交和回执,ChatActionStore 持有原确认,Dashboard 负责展示该确认的真实状态和恢复动作。 先保存原确认的 provider revision、源注册表摘要和 operation ID,SQLite 与 File 的写路径仍用现有 CAS。已有回执时可以重新读取原操作。从当前 Goal 的持久提案和类型化原操作状态投影未知/待恢复结果,呈现同一 proposal 的重试。在已准备或已提交时保留其身份,并约束 regenerate,避免让现成恢复路径被新确认替代。 保存原确认
具体改动 / Changes
关键代码讲解
新增 v3 传入源摘要和原 provider revision。TypeScript 在新写入前检查来源及 revision,在已提交时先读原回执。HTTP 的通用错误响应随后仍把打包页面交给 apply_failed,可见按钮触发 regenerate。实测第二次确认只找到了已经创建的任务,同时给出另一个 proposal 和 receipt ID。回执、来源和 CAS 通用失败回传 界面重试
基线 docs/project-agent-todo-contract.md 的 原操作回执 和 新请求来源约束 已验证。Chat 原确认恢复 仍有上述反例。现有前端交付要求从普通入口走到错误恢复和读回。基线写入合同 前端验收路径
对主干的风险 / Risks
独立同条件下,基线回原提案得到 stale,候选 HTTP 能恢复原回执。页面第二次确认未重复写 Todo,仍让用户多确认一次且拿到不同的回执。当前 main 的合成检查还发现 local_authority_runtime.ts 冲突,需要基线集成后再验证。后续 #8727 只代表一个单独候选。
语义与 CI 对齐
28 项 Python、5 项 TypeScript 与 17 项 standard premerge 通过,另有一项 PostgreSQL 集成测试跳过。 模拟飞书确认和原生 HTTP 复读通过。当前实际展示缺口来自精确头的打包页面。
我的整体评价 / Verdict
请求修改。从当前 Goal 的持久提案和类型化原操作状态投影未知/待恢复结果,呈现同一 proposal 的重试。在已准备或已提交时保留其身份,并约束 regenerate,避免让现成恢复路径被新确认替代。 用同一个 File/SQLite 提交后故障,在打包 Chat 确认、刷新和重试原操作。断言原 proposal、Todo 和 receipt ID 一致、权威修订不变,且未提交的过期预览仍可重建。
1200 × 854 桌面和 390 × 854 窄屏均显示原任务和「未写入」,主按钮要求重新确认。窄屏按钮在首屏,两次明确确认才能得到另一个提案的结果。 实测使用合成 Goal、真实本机 Chat HTTP 和 File/SQLite;飞书走模拟确认,PostgreSQL integration 跳过。已安装 App、真实飞书及托管 CI 未验证。
English verdict: REQUEST_CHANGES at a00b3dc. An original HTTP retry recovers the committed receipt, but packaged Chat falsely reports no write and regenerates a different confirmation instead of exposing that retry.
| return self._create_confirmation_conflict(proposal, error) | ||
| todo_id = _opaque(canonical_receipt["todo_id"], field="todo_id") | ||
| self.store.save_checkpoint( | ||
| proposal = self.store.save_checkpoint( |
There was a problem hiding this comment.
[P2] 打包 Chat 将已提交的确认导向新提案。 在打包 Chat 确认创建 Todo。让真实 provider 提交后,第一次保存 todo_created checkpoint 时抛出 I/O 错误。 页面显示「未写入」,唯一可见的重试先生成新 proposal,再要求确认。实测新确认返回 todo_already_exists 和另一份 receipt。原确认仍能经 HTTP 取回 todo_added 和原回执。 从当前 Goal 的持久提案和类型化原操作状态投影未知/待恢复结果,呈现同一 proposal 的重试。在已准备或已提交时保留其身份,并约束 regenerate,避免让现成恢复路径被新确认替代。 用同一个 File/SQLite 提交后故障,在打包 Chat 确认、刷新和重试原操作。断言原 proposal、Todo 和 receipt ID 一致、权威修订不变,且未提交的过期预览仍可重建。
|
This pull request has merge conflicts with Choose the remote for the base repository, not an out-of-date fork. git fetch upstream
git rebase upstream/main
# Resolve each conflict, git add the resolved files, then git rebase --continue.
git push --force-with-lease origin HEADFor a same-repository clone whose Keep the DCO |
loopx-agent
left a comment
There was a problem hiding this comment.
Reviewer: model_agent · gpt-6.1-sol · OpenAI · runtime_reported · reasoning_effort=max
Request changes conclusion (author-owned PR; GitHub blocks formal self-review)
评审提交 05007a8e6a62b3e37041151d4e1722a1ad606e5d。GitHub 目标基线为 58cc5817e35ea9a0dc4697f7791279c6d80946bc。
[P2] 创建确认在保存计划前失败后失去恢复入口。 在 Chat 确认创建 Todo,首次保存 todo_create_prepared 时发生一次 I/O 错误,尚未调用权威 Todo writer。 两种存储都没有写入 Todo。HTTP 对原确认的四个处置入口均返回 409,而原 apply 在故障消失后能创建一次。页面的 review plan 没有重试动作,刷新后待确认列表也排除了原确认。 仅在有准备记录时才提供重试
动机 / Motivation
在 Chat 中确认创建 Todo 或 Monitor 的使用者。 在 Goal 中确认创建 Todo 或 Monitor,短暂写入故障结束后应能找回原确认和结果。 已写入 Todo 或 Monitor 但丢失确认回执时,File 和 SQLite 均保留原确认,重试取回同一个 Todo 与回执,权威修订不再变化。 已验证本地打包 Chat、合成 Goal 和真实 File/SQLite。新反例来自实际 HTTP 与当前前端 selector,已安装 App、真机飞书和 PostgreSQL 端到端尚未核验。
改动思路 / Approach
Chat 保存确认计划,Canonical Todo owner 检查来源与 revision 后写一次。已提交时从原 operation 返回回执,页面需要让这个恢复路径可见。 让已确认但未准备成功的创建有可见的原操作重试。由现有 Chat action owner 统一标记可安全重试的状态,再让待确认列表和详情使用同一标记。继续保护任何可能已经写入的确认。 原确认交给 Canonical Todo owner
具体改动 / Changes
关键代码讲解
- 先保存原 operation ID、registry 摘要和 store identity,再进入 Todo/Monitor 的真实 writer。两个存储在提交后丢回复都可找回同一回执。取消、推迟、拒绝和重新生成均被拦截,原操作续跑只读回已提交的结果。
- 首次保存准备记录失败时,实际响应仍进入 canonical_action_failed。后端保护原确认,前端等待并不存在的准备记录。待确认列表只纳入可重试的失败项,所以刷新后找不到原卡。待确认列表的筛选
- 差异含 33 个文件,+1,735/−292。大部分验证来自跨存储写入、回执和打包场景。独立确认逻辑放在现有 Chat Todo owner,后续分支继续复用它。
按现有 Todo 合同和前端交付要求核对。original-operation-replay、source-and-no-write 已验证,recoverable-confirmation-before-plan 未满足。原确认与来源约束 用户入口的恢复要求
对主干的风险 / Risks
在 GitHub 目标基线制造相同的写入前 I/O 中断,两个 provider 都保持原修订,HTTP 允许重新生成可确认的 preview。 影响范围是 Chat 已确认创建后的首个准备写失败。用户仍能看到当时错误,但刷新后会失去原确认的恢复入口。 在 HTTP 保存失败对象后,Python 保护通用失败状态;TypeScript 仅把已保存计划或已完成检查点投影为可恢复。 使用两个隔离存储,在首次保存确认计划和权威写入后分别注入一次故障,刷新确认抽屉,验证原动作可重试、来源变化仍被拒绝,原 Todo/Monitor 和回执在重复提交后保持一致。
语义与 CI 对齐
精确头的 110 项 Python、25 项 TypeScript(1 项跳过)、19 项预合并检查通过。后续 12 项打包 HTTP 测试通过。File 和 SQLite 的写前反例均可复现。 托管 CI 未查询。已合成检查与最新 main 的变化,代码结论绑定上述提交。
我的整体评价 / Verdict
请求修改。让已确认但未准备成功的创建有可见的原操作重试。由现有 Chat action owner 统一标记可安全重试的状态,再让待确认列表和详情使用同一标记。继续保护任何可能已经写入的确认。 打包 Chat 的 1200×854 桌面和 390×854 窄屏上,已保存确认的任务标题和「重试原操作」都在首屏。窄屏保留了原结果和返回入口。提前中断时,实际失败响应使当前页面把确认排除在列表外。 已验证本地打包 Chat、合成 Goal 和真实 File/SQLite。新反例来自实际 HTTP 与当前前端 selector,已安装 App、真机飞书和 PostgreSQL 端到端尚未核验。
English verdict: REQUEST_CHANGES. A transient failure before writing the create plan makes the original confirmation disappear after reload, even though retrying its original HTTP action succeeds. Keep that action visible while continuing to fence committed work.
| } | ||
| // A confirmation with a saved create plan may already have written. Only a | ||
| // complete plan for this exact proposal can offer native recovery on reload. | ||
| if (todoCreation && ["failed", "applying"].includes(String(proposal.status)) |
There was a problem hiding this comment.
[P2] 创建确认在保存计划前失败后失去恢复入口。 保存 todo_create_prepared 前的一次 I/O 故障在 File 和 SQLite 均复现。实际 HTTP 仍可重试原确认,但这个条件要求已有准备记录,导致页面隐藏恢复入口。请让两处共享原确认的可安全恢复判断,并保留已提交路径的四个拒绝。
|
This pull request has merge conflicts with Choose the remote for the base repository, not an out-of-date fork. git fetch upstream
git rebase upstream/main
# Resolve each conflict, git add the resolved files, then git rebase --continue.
git push --force-with-lease origin HEADFor a same-repository clone whose Keep the DCO |
loopx-agent
left a comment
There was a problem hiding this comment.
Reviewer: model_agent · gpt-6.1-sol · OpenAI · runtime_reported · reasoning_effort=max
Request changes conclusion (author-owned PR; GitHub blocks formal self-review)
评审提交 7bdda5bfea4846cca5e7bb3b3caa5f8c6d75ee8a,目标基线 eaff94c7449c2fa3ce49edf3adc7e6258ccd8b0c,上一评审头 05007a8e6a62b3e37041151d4e1722a1ad606e5d。
[P2] 创建确认在保存计划前失败后失去恢复入口。 在 Chat 确认创建 Todo,首次保存 todo_create_prepared 时发生一次 I/O 错误,尚未调用权威 Todo writer。 File 和 SQLite 首次都没有写入 Todo。HTTP 取消、推迟、拒绝及重新生成均返回 409;原 apply 能创建一次。实际打包页面刷新后丢失原确认和重试入口。 重试要求已保存的计划
动机 / Motivation
在 Chat 中确认创建 Todo 或 Monitor 的使用者。 在 Goal Chat 中确认创建后,即使写入前中断也能找回原操作。 提交后的回执丢失在 File 和 SQLite 都可恢复原 Todo 或 Monitor。重试保留原任务、确认回执和 provider revision。 验证使用本地打包 Chat、合成 Goal、实际 File/SQLite;已安装 App、真实飞书和 PostgreSQL 端到端待核对。
改动思路 / Approach
Chat 保存确认计划,Canonical Todo owner 校验来源与 revision 后写一次。页面用现有 review plan 投影原操作的恢复或结果。 复用现有 Chat action owner 对原确认的安全重试判断,让待确认列表和详情都能提供它。可能已经写入的确认仍走原 operation。 原确认和计划的 owner
具体改动 / Changes
关键代码讲解
- 原 proposal 冻结 operation ID、来源摘要与 store identity。Todo 和 Monitor 在权威写入后丢确认回执时,从原 ID 恢复,不增加一次 provider revision。
- 写前失败仍会产生
retry_safe=true的canonical_action_failed。后端保护原确认,前端等待尚未保存的计划,待确认列表依赖retryOriginal收入失败项。待确认列表 - 全 PR 有 33 文件,+1742/-297。上次评审之后,这 33 个路径中 27 个字节相同。其余涉及导航、确认详情、服务及 plan 编译。最新提交还移除了已核验结果里的闲置确认控件,写前恢复条件仍可复现。
original-operation-replay 和 source-and-no-write 已核对,recoverable-confirmation-before-plan 尚未满足。原确认和来源 前端任务恢复
对主干的风险 / Risks
在当前 GitHub 目标基线模拟写入前的相同故障,两种存储都保持原修订,重新生成可得 201 和待确认预览。 最新增量移除了已核验创建结果的一个失效按钮,同时叠加现有 Goal/会话导航更新。仍按旧条件决定写前恢复,并通过两个 provider 和实际页面重测。 使用原确认分别注入首次准备写入失败和提交后回执丢失,在真实 File/SQLite、打包桌面和窄屏刷新,断言原确认可恢复并只写一份。
语义与 CI 对齐
当前 110 项 Python 和 32 项 TypeScript 通过,另 1 项 PostgreSQL 跳过。标准 premerge 共执行 19 项,CLI 输出预算 smoke 超时,quota-plan smoke 因导入失败退出;补齐本地测试导入路径后,后者在相同基线和候选上单独运行均通过。 托管 CI 未查询。打包本地页面和直接 HTTP 给出一致的恢复缺口。
我的整体评价 / Verdict
请求修改。复用现有 Chat action owner 对原确认的安全重试判断,让待确认列表和详情都能提供它。可能已经写入的确认仍走原 operation。 1200×854 桌面和 390×854 窄屏完成加载后,正常待确认项仍显示;两处都找不到首次准备失败的原确认和重试按钮。 验证使用本地打包 Chat、合成 Goal、实际 File/SQLite;已安装 App、真实飞书和 PostgreSQL 端到端待核对。 在故障结束后,用户应能从原 Goal 对话找到并继续已有确认。
English verdict: REQUEST_CHANGES at 7bdda5b. The exact packaged Chat still hides an original confirmation after a transient prepare failure. Both providers can recover the original action via HTTP; expose that recovery in the original Chat view.
| && createPlan.preview_id === identity.sourceFingerprint | ||
| && textValue(createPlan.source_identity) && textValue(createPlan.source_authority) | ||
| && textValue(createPlan.expected_provider_revision) && textValue(createPlan.expected_registry_sha256) | ||
| && textValue(createPlan.expected_store_identity)) { |
There was a problem hiding this comment.
[P2] 创建确认在保存计划前失败后失去恢复入口。 File/SQLite 的真实原 apply 可以安全恢复,桌面和窄屏的已加载列表却都缺原确认。让既有 action owner 的可安全恢复状态同时驱动列表和重试。
loopx-agent
left a comment
There was a problem hiding this comment.
Reviewer: model_agent · gpt-6.1-sol · OpenAI · runtime_reported · reasoning_effort=max
Request changes conclusion (author-owned PR; GitHub blocks formal self-review)
评审提交 54c8d23b5c3d22d28bd02cb4a746bc87ba331c0d;对照目标基线 b46a76d2d64b3c14877bdf288953c56ae1c7793b。
[P2] 首次保存准备记录前中断后,原确认无法恢复。 确认请求在保存首份创建计划和写入任务之前退出;前端刷新后只读回 applying,且没有 checkpoint 或失败结果。 桌面和窄屏的原卡都还在,抽屉只显示禁用的「正在应用」,没有重试。File/SQLite 两种存储都没有首写;同一 apply 能成功完成一次,第二次返回原回执。 对应代码
[P2] 首次计划丢失后,Monitor 重试使用未审阅的新修订。 Monitor 确认遇到首份计划保存错误,未写入;有人对同一 Goal 完成一次有效修改,provider revision 前进,来源注册表不变。 同一 Monitor 提案重试返回 200,用新的修订号保存计划并新增 Monitor;同条件的 Todo 返回 409,仍未写入。两个后端结果一致。 对应代码
动机 / Motivation
在 Goal 里从对话确认待办或定时检查(Monitor)的人。 提交待办或定时检查后,即使首次确认中断,也应能重新打开原确认并完成它。 以前确认中断后,提交人得自己核对后台有没有留下任务,避免重复创建。 要么只能等待,要么离开当前确认查询后台,核对是否已写入和是否会重复。 可捕获的准备写入错误保留原提案;未变更修订时只创建一次。写入已发生但回执丢失时,重试仍保留原 Todo/Monitor 和回执。 验证使用源码构建的打包 Chat、实际本机 HTTP 和隔离 File/SQLite;PostgreSQL 用例跳过,未验证装机 App、实际飞书租户或托管 CI。
改动思路 / Approach
以原提案 ID 找回操作。Hub 的 action store 保留确认和步骤,Goal 对应的 Todo 存储判定能否写入并生成回执,Dashboard 展示原记录和恢复入口。 复用现有操作锁和已经有来源绑定的 provider 写入。保留一个类型化恢复条件,并将阶段与原确认依据配对;不再建独立重试协议或让界面从错误文案推断写入。 服务判定恢复和初次写入权限;provider 拥有唯一写入和原回执;页面只展示原确认和读回结果。 先保存同一提案的准备信息,再交给已注册 Goal 的权威存储;已有原回执时按同一 ID 核验和更新展示。 首份计划和写入后的回执是两道不同的持久化边界;真实请求退出不会留下可捕获的错误记录。
具体改动 / Changes
关键代码讲解
apply;compileActionReviewPlan;_apply_monitor_create。原提案的锁会阻止仍在执行的同一次 apply;锁释放后,服务能够安全读回或重放。当前抽屉在首份记录缺失时要求先看到已经持久化的失败,遗漏了请求先退出的分支。Monitor 的动作指纹只有注册表和来源路径;记录尚未保存时,服务给它的计划写入了恢复时的存储版本,无法证明这个版本是操作者审阅过的。生产 24 文件,+749/-289;测试和夹具 8 文件,+1023/-9;其余为文档和工作流。
基线写入约束、打包任务验收及本提交的原确认约束:original-create-result 已验证;preplan-interruption-recovery 未满足;reviewed-monitor-revision 未满足。
对主干的风险 / Risks
首份计划和写入后的回执是两道不同的持久化边界;真实请求退出不会留下可捕获的错误记录。 桌面和窄屏的原卡都还在,抽屉只显示禁用的「正在应用」,没有重试。File/SQLite 两种存储都没有首写;同一 apply 能成功完成一次,第二次返回原回执。 同一 Monitor 提案重试返回 200,用新的修订号保存计划并新增 Monitor;同条件的 Todo 返回 409,仍未写入。两个后端结果一致。 由已有 action owner 区分仍有持锁请求和可恢复的原确认;为后者开放同一 apply,继续由原锁、来源和修订号把关,保留对取消和重新生成的限制。 在原 Monitor 确认依据中保留已审阅的修订号及存储见证,在首写前比较;有原提交回执时仍按原 operation ID 恢复,不让重试静默采用当前修订。 恢复权限只覆盖所选的原提案。服务须区分活请求、陈旧依据和可安全重试的原确认;有历史回执时沿用原结果。基线 HTTP 已能恢复 applying 提案,本轮缺口是被当前交付承诺覆盖的页面和未写 Monitor 的修订检查。
语义与 CI 对齐
PASS:82 项定向 Python、26 项定向 TypeScript、19 项标准 premerge(含 diff 与双语 RFC 检查)。 标准门首次因本机示例导入路径和临时目录位于私有树下失败,换用隔离的系统临时目录并保留示例导入路径后通过。全量类型检查在基线和候选的同一位置报 TS2345。 验证使用源码构建的打包 Chat、实际本机 HTTP 和隔离 File/SQLite;PostgreSQL 用例跳过,未验证装机 App、实际飞书租户或托管 CI。 旧 wire 的降级和缺见证对照仍由同一验证入口拒绝;原 Todo 修订过期后可重新审阅且不写任务。用打包 Chat/HTTP 在标为 applying 后、保存准备记录前中断请求,刷新后从原抽屉重试;核验只创建一次、返回同一回执,另设仍持锁和旧修订对照。 让首份 Monitor 计划写入报错,再通过另一项 Todo 写入推进修订,重试同一提案;断言原请求拒绝首写,且此前已提交的 Monitor 仍可读回原回执。
我的整体评价 / Verdict
请求修改。正常的已提交恢复能继续工作;请求退出前后仍有两处未满足的约束,持续恢复和页面可操作性都还未验收。桌面保留原卡,窄屏抽屉中长标题可完整换行;首份准备前的中断在两处都只显示禁用的「正在应用」。 优先修复共同的确认 owner,再重跑上述两个对照和原回执、浏览器恢复;沿用现有状态字段和权限边界。验证使用源码构建的打包 Chat、实际本机 HTTP 和隔离 File/SQLite;PostgreSQL 用例跳过,未验证装机 App、实际飞书租户或托管 CI。
English verdict: REQUEST_CHANGES at 54c8d23. A persisted applying confirmation can stay disabled after reload, and an unwritten Monitor can adopt an unreviewed revision on retry.
| && textValue(createPlan.expected_store_identity); | ||
| // The server may fail before any checkpoint is saved. The same apply still | ||
| // owns the reviewed source and must revalidate it before the first write. | ||
| const interruptedBeforePlan = createAttemptFailed && failure?.retry_safe === true && proposal.checkpoint == null |
There was a problem hiding this comment.
[P2] 首次保存准备记录前中断后,原确认无法恢复。 确认请求在保存首份创建计划和写入任务之前退出;前端刷新后只读回 applying,且没有 checkpoint 或失败结果。 桌面和窄屏的原卡都还在,抽屉只显示禁用的「正在应用」,没有重试。File/SQLite 两种存储都没有首写;同一 apply 能成功完成一次,第二次返回原回执。 由已有 action owner 区分仍有持锁请求和可恢复的原确认;为后者开放同一 apply,继续由原锁、来源和修订号把关,保留对取消和重新生成的限制。 用打包 Chat/HTTP 在标为 applying 后、保存准备记录前中断请求,刷新后从原抽屉重试;核验只创建一次、返回同一回执,另设仍持锁和旧修订对照。
| def _monitor_create_fingerprint(source: ChatTodoSource) -> str: | ||
| from ..chat_actions import _digest | ||
|
|
||
| return _digest({ |
There was a problem hiding this comment.
[P2] 首次计划丢失后,Monitor 重试使用未审阅的新修订。 Monitor 确认遇到首份计划保存错误,未写入;有人对同一 Goal 完成一次有效修改,provider revision 前进,来源注册表不变。 同一 Monitor 提案重试返回 200,用新的修订号保存计划并新增 Monitor;同条件的 Todo 返回 409,仍未写入。两个后端结果一致。 在原 Monitor 确认依据中保留已审阅的修订号及存储见证,在首写前比较;有原提交回执时仍按原 operation ID 恢复,不让重试静默采用当前修订。 让首份 Monitor 计划写入报错,再通过另一项 Todo 写入推进修订,重试同一提案;断言原请求拒绝首写,且此前已提交的 Monitor 仍可读回原回执。
|
This pull request has merge conflicts with Choose the remote for the base repository, not an out-of-date fork. git fetch upstream
git rebase upstream/main
# Resolve each conflict, git add the resolved files, then git rebase --continue.
git push --force-with-lease origin HEADFor a same-repository clone whose Keep the DCO |
loopx-agent
left a comment
There was a problem hiding this comment.
Reviewer: model_agent · gpt-6.1-sol · OpenAI · runtime_reported · reasoning_effort=max
Request changes conclusion (author-owned PR; GitHub blocks formal self-review)
评审提交 af94fe6d0b9acdcc0375bd39149a8c5ad4fdd301,对照目标基线 cbdc7655482e3db8cc70c5b0c4e49451d902a360。
[P2] 无关提交后恢复已创建 Todo 的指派。 确认时选择了具体 Agent。Todo 已写入并保存 todo_created 回执,另一个请求随后写入同 Goal 的无关 Todo。 当前提交在两个后端首次 apply 和原提案 retry 均返回 HTTP 400,报告修订已变化。原 Todo 已持久化且未指派,用户重试时仍使用刚创建时的 Goal 修订。 指派的修订前置
动机 / Motivation
在 Chat 中确认创建并指派任务或设置 Monitor 的操作者。 两种恢复触发点包括首份创建计划之前的请求中断,以及创建落库和后续指派之间的独立 Todo 写入。 原确认在首份计划前失联时,基线刷新后仍卡在 Applying。修复了首份计划前中断后的恢复。打包 Chat 刷新后能从原卡片继续,File 与 SQLite 都只创建一次。无写入的 Monitor 在另一个写入推进修订后返回 409,保持原数据。 独立覆盖打包 Chat 的桌面路径、真实 HTTP 与隔离 File/SQLite。真实安装宿主和 PostgreSQL 集成不在本次证据范围。
改动思路 / Approach
Chat 执行者持有原 proposal 和创建 checkpoint,canonical Todo provider 管理修订、权限和回执。 复用已有原生 create、update、claim 和 source witness。分配继续用原操作身份,缺少有依据的目标冲突恢复。 当前恢复读取原 checkpoint 后复用其 creation receipt,再发出原指派操作。保留原创建和指派 operation id。遇到 Goal 修订推进时,在原 store 内确认同一创建回执和目标 Todo 的指派前置条件,再用原子校验继续指派;目标、归属或来源实质变化时返回可操作的冲突。
具体改动 / Changes
关键代码讲解
原确认的中断恢复已经核验。File 和 SQLite 的两种视口、三种中断位置共 12 个打包浏览器用例都回读了原提案和单条新增记录。确认持有者检查
Monitor 的首写修订已经核验。File 与 SQLite 的基线在其他 Todo 更新后仍创建 Monitor,候选在原审阅过期时返回 409,两个实际存储的行和修订均保持不变。Monitor 首写检查
创建后的指派恢复仍有反例。在精确目标基线,File 和 SQLite 两个完整 HTTP 入口都返回 200,原 Todo 保持一次创建并完成指派。 候选把创建时修订留在 checkpoint。每次重试都传同一旧值,无关 Todo 的提交令该指派始终被拒绝。指派调用 · 现有 Todo 写入契约
对主干的风险 / Risks
用户已确认指定 Agent,也已经有了可见任务。两个后端的原提案重试均没有新写入,目标任务仍未指派。无关 Todo 自己的写入保持有效。现有用例覆盖了来源替换与恢复原来源,仍要验证来源有效而同 Goal 修订已推进的情况。
语义与 CI 对齐
112 项 Python 检查、26 项 TypeScript 检查和本地 standard premerge 的 25 项执行通过。PostgreSQL 用例跳过。 对照试验只在原 Todo 创建和指派之间增加一次无关提交。直接检查通过覆盖不了这段顺序,目标及无关记录已经分别读回。
我的整体评价 / Verdict
请求修改。原确认的恢复和过时 Monitor 拒写已在当前提交复验。把原任务的指派恢复加到真实 HTTP 回归中,覆盖无关更新与目标或来源变化两种情况,并核对同一任务、同一操作及实际归属。1200 × 854 和 390 × 854 的打包 Chat 都能读全原任务标题。刷新后可找到原卡的重试入口,成功后显示原回执和查看更新后的 Goal。高级诊断保留在折叠区。 独立浏览器覆盖桌面和窄屏,实际安装宿主仍待验证。
English verdict: REQUEST_CHANGES at af94fe6. An unrelated Todo update leaves the original confirmed assignment permanently retrying a stale Goal revision.
| agent_id=str(agent_id), | ||
| authority_reason="owner-confirmed typed Chat action", | ||
| update_operation_id=f"chat-assign:{proposal_id}", | ||
| update_expected_provider_revision=binding["provider_revision"], |
There was a problem hiding this comment.
[P2] 无关提交后恢复已创建 Todo 的指派。 确认时选择了具体 Agent。Todo 已写入并保存 todo_created 回执,另一个请求随后写入同 Goal 的无关 Todo。 File 和 SQLite 的基线都返回 200 并完成指派。候选首次提交和原提案重试都返回 400,原 Todo 已写入且没有归属。保留原创建和指派 operation id。遇到 Goal 修订推进时,在原 store 内确认同一创建回执和目标 Todo 的指派前置条件,再用原子校验继续指派;目标、归属或来源实质变化时返回可操作的冲突。
loopx-agent
left a comment
There was a problem hiding this comment.
Reviewer: model_agent · gpt-6.1-sol · OpenAI · runtime_reported · reasoning_effort=max
Request changes conclusion (author-owned PR; GitHub blocks formal self-review)
评审提交 379e72aad527319a5c10d18903b29f064f5a131f,本地和 GitHub 比较以当前 main f3bdad2387397e078fbc887e7cb7ed80adaf7510 为参考,实际 merge base f3bdad2387397e078fbc887e7cb7ed80adaf7510。净改动 40 个文件(+2197/−334)。GitHub PR 面板的 157 个文件沿用较早的 cbdc765548。 精确比较
[P2] 保留旧版原确认的安全接续
在基线运行实际 Chat HTTP,确认创建并指派一个 Todo。在 provider 提交和 todo_created checkpoint 持久化后、指派前注入一次中断。用候选代码重启同一隔离服务,保留同一 registry、provider、Todo 和 proposal,读回原 native create receipt,再对原确认提交两次。 File 和 SQLite 都能读回唯一的原 create receipt,原 Todo 内容保持一致。候选两次都返回 HTTP 400,任务始终未指派,provider cursor 不变。主干原版本重开同一场景,两次返回 200 和同一回执,只提交一次指派。 候选的取消与重新生成也都返回 409。File 打包 Chat 显示无新动态,原卡在列表 API 中仍为 failed。从原 proposal 的旧操作 ID、原创建回执和可验证的原存储谱系核对同一 Todo,保留当前注册与 CAS 校验。来源证据不足时,读回可证实的已创建结果,并显式引导对同一个 Todo 重新审阅指派。 代码
动机 / Motivation
面向确认创建 Todo 或 Monitor 后需要重新打开 Chat 的使用者。 主干先写 provider,再保存 Chat 的创建 checkpoint。候选先持久化创建准备,让原确认能在丢响应后重放。 File 和 SQLite 的四个无关修订/直接续做对照都只指派一次。末次换源下 Todo 与 Monitor 共四例返回 409,两库零新增。1200 px 英文 File 和 390 px 中文 SQLite 两个打包 Chat 场景在已提交突然中断后从原卡片恢复同一回执,两次点击合计只有一次创建。 独立验证覆盖本地 File、SQLite 与打包 Chat。 既有来源契约 恢复验收
改动思路 / Approach
增加原提案执行锁与写入前 checkpoint,沿同一 native create 重放,复用 #8717 的 source resolver 和创建后的原指派。 同一原卡片继续使用一个已确认操作,减少重新发起请求和再次选择 Agent 的成本。更小的 UI 刷新方案拿不到第一次创建回执;当前持久方案能恢复新版记录。对主干旧确认仍需完成既有结果读回并恢复指派。引用作者列出的 #8727 展示回执后续,单独保留该阶段的资格。 Chat 保留确认和界面状态,native Todo owner 处理权限、比较并交换(CAS)与命令回执,隔离 provider 状态用同一入口读回。 同一原操作要求
具体改动 / Changes
关键代码讲解
-
_save_create_plan原 proposal 和 reviewed CAS、source、store 持久化后才尝试首写,Monitor 保存首次时钟。 恢复沿用同一命令。 代码 -
_apply_todo_create按原效果回执接续当前指派。 旧记录缺 source 时抛出 400,同一原记录无法推进。 代码 -
createLocalCoordinationTodo在当前 authority guard 内核对 expected store 并进入同一 TS receipt/CAS owner。 同源已提交可以重放,换源零写入。 代码 -
compileActionReviewPlan从 typed proposal 支撑原卡恢复,核对已创建记录的来源字段。 旧安装确认为 failed 且缺新字段时,列表仍有记录,Goal Chat 不展示原卡。 代码
验收覆盖 same-confirmation 已验证,monitor-source 已验证,retained-confirmation 未达到,original-assignment 已验证。40 文件包括 #8717 的来源路由、确认及写入恢复、打包 Chat 的卡片和前端 schema,以及浏览器 smoke 的工作流接线。 来源契约 上一轮评审
对主干的风险 / Risks
当前恢复覆盖新版 checkpoint。存量确认仍可留下已创建、未指派的任务,刷新后无法从同一 Goal Chat 取得原确认操作。 任务恢复契约
语义与 CI 对齐
150 项 Python 通过,62 项 TypeScript 通过、1 项跳过。运行时仍有旧记录跨版本失败,精确候选的原生 bundle 构建、差分检查与两种视口恢复通过。 三个精确版本的 Python 摘要归属检查均出现同一历史失败,关联 #8836。托管 CI 未查询。旧 v2 到 witness-only v3 原命令回执重放测试通过。独立打包验证覆盖已提交的确认。 摘要归属修复
我的整体评价 / Verdict
请求修改。从原 proposal 的旧操作 ID、原创建回执和可验证的原存储谱系核对同一 Todo,保留当前注册与 CAS 校验。来源证据不足时,读回可证实的已创建结果,并显式引导对同一个 Todo 重新审阅指派。 让旧版已创建但未指派的确认在原卡显示可执行恢复。保留有来源绑定的新确认的防重和原时钟。 用实际 HTTP、File 和 SQLite 从旧版本持久化后的同一记录重放两次,验证原回执、目标归属和单次写入。把旧文件和 SQLite 投影纳入打包 Chat 的同一原卡路径。
Ego Lite 中 File 1200 × 854 和 SQLite 390 × 854,恢复入口和原任务标题完整可见,成功回读同一原回执。旧记录的 File 1200 × 854 对照保留在 API,Chat 显示无新动态。 旧格式确认的窄屏、展开的诊断信息、安装 App 和实时飞书未验证。
English verdict: REQUEST_CHANGES at 379e72a. Fresh prepared confirmations recover correctly, but an interrupted confirmation kept from main cannot resume its original assignment and disappears from packaged Goal Chat.
| raise ValueError("Todo creation checkpoint is malformed") | ||
| binding = todo_step.get("source") | ||
| if not isinstance(binding, dict): | ||
| raise ValueError("The created Task predates source binding. Recover the original Todo before assigning it.") |
There was a problem hiding this comment.
[P2] 保留旧版原确认的安全接续
在基线运行实际 Chat HTTP,确认创建并指派一个 Todo。在 provider 提交和 todo_created checkpoint 持久化后、指派前注入一次中断。用候选代码重启同一隔离服务,保留同一 registry、provider、Todo 和 proposal,读回原 native create receipt,再对原确认提交两次。 File 和 SQLite 都能读回唯一的原 create receipt,原 Todo 内容保持一致。候选两次都返回 HTTP 400,任务始终未指派,provider cursor 不变。主干原版本重开同一场景,两次返回 200 和同一回执,只提交一次指派。 候选的取消与重新生成也都返回 409。File 打包 Chat 显示无新动态,原卡在列表 API 中仍为 failed。从原 proposal 的旧操作 ID、原创建回执和可验证的原存储谱系核对同一 Todo,保留当前注册与 CAS 校验。来源证据不足时,读回可证实的已创建结果,并显式引导对同一个 Todo 重新审阅指派。
Signed-off-by: LoopX Agent <337587101+loopx-agent@users.noreply.github.com>
loopx-agent
left a comment
There was a problem hiding this comment.
Reviewer: model_agent · gpt-6.1-sol · OpenAI · runtime_reported · reasoning_effort=max
Request changes conclusion (author-owned PR; GitHub blocks formal self-review)
评审提交 40de5dd70711026730d44a165670997aea7bc5d7。以当前主干 f3bdad2387397e078fbc887e7cb7ed80adaf7510 为 merge base,本地 Git 与 GitHub 对照均为 41 个路径(+2204/−340)。上一轮评审后只修改了打包浏览器的 Goal 创建 smoke。产品文件逐字一致,以下旧确认结果已在新 head 复验。 完整差分
[P2] 让升级前已创建的确认续做原指派
在精确主干上用 Chat HTTP 确认创建并指派。等原创建提交及老格式 todo_created 持久化,在指派前中断。候选重开同一 registry 和实际存储,提交原确认两次。File 和 SQLite 都能读回唯一的原创建回执,Todo 与原内容一致。候选两次返回 HTTP 400,指派与 provider revision 均未推进,取消和重新生成返回 409。基线两次返回 HTTP 200 和同一个原回执,只写一次指派。File 的 1200 px 打包 Chat 读到同一个 Goal,原确认仍在 HTTP API,页面没有恢复卡片。 验证同一原创建回执及历史来源,按当前注册和原 Todo 前置条件安全续做指派。证据不足时,让原卡展示已创建的结果并引导用户为同一个 Todo 重新审阅指派。保留替换库拒绝、原操作防重和首次确认时钟。 代码
动机 / Motivation
检查使用现有 Chat HTTP、隔离 File/SQLite、实际构建的 Chat 和受控外部会话。已安装桌面 App、生产飞书运输和独立 PostgreSQL 资格仍待验证。
上一轮评审后只修改了打包浏览器的 Goal 创建 smoke。产品文件逐字一致,以下旧确认结果已在新 head 复验。
重新打开已注册 Goal 的使用者,要完成原 Todo 创建、指派和展示。
用户确认创建 Todo 或 Monitor 后遇到中断,可以重开同一确认,取回原任务并安全接续其工作。 新格式确认从原卡恢复同一 Todo 和 receipt。独立 Todo 推进同源修订后,两库各只写一次指派。末次换源使 Todo 和 Monitor 首写均返回 409,两库零新增。 来源契约 R1 恢复验收
改动思路 / Approach
Chat 保留确认和界面状态,native Todo owner 处理权限、比较并交换(CAS)与命令回执,隔离 provider 状态用同一入口读回。
先保存原确认计划,使用同一 native create 重放,并由 #8717 的来源 resolver 和后续指派的 CAS owner 处理权限。 单独重载页面只能再次读已有投影,后续还需要从 canonical 回执和原操作恢复。让旧确认使用同一个来源及指派校验能避免产生新的业务操作。 来源契约 R1 恢复验收
具体改动 / Changes
关键代码讲解
-
_save_create_plan先持久化 reviewed source 和 provider CAS,再执行首写。 代码 -
The created Task predates source binding旧确认要求指派时仍在此拒绝恢复。 代码 -
createLocalCoordinationTodo核对原提交、注册来源和 provider CAS。晚来源变化使原库与替代库均保持零新增。 代码 -
workspaceProposal消费 typed 状态决定原卡操作。 代码
逐项验收为 same-confirmation 已验证;monitor-source 已验证;retained-confirmation 未达到;original-assignment 已验证。
对主干的风险 / Risks
升级期间中断的旧确认可能长期留下未指派任务。用户刷新后无法在对应 Goal Chat 继续原动作。新版已审核来源和实际写入的防重通过独立复验。 来源契约 R1 恢复验收
语义与 CI 对齐
150 项 Python、62 项 TypeScript 与 8 个独立 File/SQLite HTTP 观察通过,TypeScript 跳过 1 项。两份候选和基线的全树摘要规范检查均有相同一项历史失败。打包 Chat 构建通过。 托管 CI 未查询。 检查使用现有 Chat HTTP、隔离 File/SQLite、实际构建的 Chat 和受控外部会话。已安装桌面 App、生产飞书运输和独立 PostgreSQL 资格仍待验证。 摘要归属修复
我的整体评价 / Verdict
请求修改。验证同一原创建回执及历史来源,按当前注册和原 Todo 前置条件安全续做指派。证据不足时,让原卡展示已创建的结果并引导用户为同一个 Todo 重新审阅指派。保留替换库拒绝、原操作防重和首次确认时钟。 用升级前真实写入、升级后两次 apply 及打包 Chat 校验原 receipt、原任务、指派次数和替代库拒绝。
在 1200 × 854 File 和 390 × 854 SQLite 打包 Chat 中,长标题和原确认重试可触达。保留旧 File 确认的 1200 px 页面显示「无新动态」,HTTP 仍可读该确认。
独立截图为取样的桌面和窄屏打包 Chat。未实测已安装桌面、真实飞书运输和旧版 PostgreSQL。
English verdict: REQUEST_CHANGES at 40de5dd. An interrupted assignment from the previous code still cannot resume on its original confirmation after upgrading.
| raise ValueError("Todo creation checkpoint is malformed") | ||
| binding = todo_step.get("source") | ||
| if not isinstance(binding, dict): | ||
| raise ValueError("The created Task predates source binding. Recover the original Todo before assigning it.") |
There was a problem hiding this comment.
[P2] 让升级前已创建的确认续做原指派
在精确主干上用 Chat HTTP 确认创建并指派。等原创建提交及老格式 todo_created 持久化,在指派前中断。候选重开同一 registry 和实际存储,提交原确认两次。File 和 SQLite 都能读回唯一的原创建回执,Todo 与原内容一致。候选两次返回 HTTP 400,指派与 provider revision 均未推进,取消和重新生成返回 409。基线两次返回 HTTP 200 和同一个原回执,只写一次指派。File 的 1200 px 打包 Chat 读到同一个 Goal,原确认仍在 HTTP API,页面没有恢复卡片。 验证同一原创建回执及历史来源,按当前注册和原 Todo 前置条件安全续做指派。证据不足时,让原卡展示已创建的结果并引导用户为同一个 Todo 重新审阅指派。保留替换库拒绝、原操作防重和首次确认时钟。
loopx-agent
left a comment
There was a problem hiding this comment.
Reviewer: model_agent · gpt-6.1-sol · OpenAI · runtime_reported · reasoning_effort=max
Request changes conclusion (author-owned PR; GitHub blocks formal self-review)
评审 head 193e6d53d8594eb00425440978c2480a3eb14e0c,对照主干 f3bdad2387397e078fbc887e7cb7ed80adaf7510。本次也核对了合入的 #8717 依赖和 #8725 自身的原卡恢复路径。 依赖评审
动机 / Motivation
在 Chat 创建 Goal 任务并选择 Agent 的用户,需要在确认响应丢失后从原卡读回同一个 Todo、原指派和原回执。 主干已经可以在 todo_assigned 保存后重放原成功回执。 新版修复了升级前只来得及保存 todo_created 的情况,重开原卡会为原 Todo 补做一次指派。 已完成指派的旧确认仍需从同一张卡读回成功。 本次检出的是本地 File、SQLite 和打包 Chat,实际安装 App、线上 Feishu 和 PostgreSQL 的旧记录升级仍待验证。
改动思路 / Approach
继续由 Chat proposal 保存确认状态,由 canonical Todo owner 核验存储谱系、操作回执和写入版本。 保留源与 CAS 校验,将历史成功读回和新指派的写入条件分别应用到对应阶段。 当前 PR 的交付边界是打包 Chat 按原 proposal 恢复已证实的创建与指派,换源或真实任务变更仍由原权限边界拒绝。 在已经提交指派的情况下,原始创建快照和当前 Todo 全量相等无法证明恢复安全。缺失的是对已完成阶段的回执核验。
具体改动 / Changes
在 f3bdad2387397e078fbc887e7cb7ed80adaf7510 的 docs/project-agent-todo-contract.md 中,Canonical registry source witnesses 要求源摘要在新写入前仍有效,并允许按原操作读回历史回执。准备记录固定原 operation、源、provider revision 和 Monitor 首次应到时间。File 与 SQLite 的晚换源和替换库场景拒绝写入。 docs/architecture/rfcs/loopx-overall-roadmap-v0.md 中的 R1 要求原确认在响应丢失后可恢复,且由发起界面显示原结果。在 todo_created 持久化、指派尚未执行的情况下,File 与 SQLite 连续两次返回 200,读回同一个原始创建回执、一个 Todo,只新增一次指派。打包 Chat 的 390 px 界面从原卡完成这一恢复。 来源契约 · R1 · 确认执行 · 原卡交互
对主干的风险 / Risks
- [P2] 已完成指派的旧确认在升级后无法读回成功。 用主干的真实 Chat HTTP 提交创建和指派,在
todo_assigned保存后中断确认响应,再重启服务,复用原 proposal、registry 和 store。主干在 File、SQLite 上都连续两次返回 200 和原回执,当前 head 两库均连续两次返回 409,原 Todo、指派和版本均未变化。Native 查询分别读回原创建回执和它的前序 revision。创建快照与当前 Todo 只在claimed_by和last_actor_agent_id有差异,这是原指派造成的变更。代码在检查已完成的todo_assigned之前就因这两个字段拒绝恢复。独立的 1200 px 打包 Chat 检查中,原卡要求恢复来源,重载后仍显示同一个 gate。 全量比较 · 已指派分支 - 建议先核验原库谱系、创建回执和已保存指派的 Todo、Agent 及提交证据,再从原卡读回同一确认。对确实需要补做的指派,继续要求相同来源、当前注册和 Todo 和 revision 校验。回归应同时覆盖主干写成的
todo_assigned,以及真实换源、撤销注册和编辑后的拒绝。 - 本轮通过 162 项 Python 检查、63 项 TypeScript 检查、Chat 构建和差异检查。TypeScript 的一项 PostgreSQL 用例跳过。摘要所有权检查在主干与候选均为 243 通过、1 失败,命中相同三处既有规则,相关源码没有变更。 已有摘要规则修复
我的整体评价 / Verdict
请求修改。 继续沿用现有原操作回执、可恢复的 Chat checkpoint 和 canonical writer,当前 PR 内修齐已完成指派的旧确认读回。 文件库 1200 × 854 第一屏显示原卡和重试。已有原指派时仍显示来源无法验证,重载后提示和原卡仍在。 SQLite 390 × 854 的原卡可进入重试,补做指派后抽屉显示结果,读回原创建回执。 实际安装 App、线上 Feishu、稠密混合状态和展开的诊断内容未在本轮验证。桌面误报来源问题会让用户在已经指派完成后继续重试或检查原任务,该分支尚未达到 R1 的原结果读回要求。
英文简短结论
English verdict: REQUEST_CHANGES at 193e6d53d8594eb00425440978c2480a3eb14e0c. The retained card recovers pre-assignment confirmations, but a persisted legacy assignment is compared to the pre-assignment Todo and both File and SQLite return HTTP 409 and leave the original receipt unreadable.
| }, | ||
| }, timeout=15.0) | ||
| todo = next((row for row in authority["todos"] if row["todo_id"] == todo_id), None) | ||
| if not isinstance(original, dict) or original.get("status") != "loaded" or result["receipts"][0]["todo"] != todo: |
There was a problem hiding this comment.
[P2] 已完成指派的旧确认在升级后无法读回成功
用主干真实 Chat HTTP 创建并指派 Todo,在持久化 todo_assigned 后丢失响应,重开同一 registry、store 和 proposal。原创建回执为 found,旧前序为 loaded。两库当前 Todo 只在 claimed_by、last_actor_agent_id 与原创建快照有差异,已保存的 todo_assigned 与实际指派一致。基线两个 provider 各两次返回 200 和原回执。候选各两次返回 409 gated。原 Todo、指派、store 和版本保持相同。 这里的全量比较比 todo_assigned 检查先执行。核验原创建回执和存储谱系后,识别并核验已持久化指派,再读回原确认。只对需要继续执行的写入沿用当前注册、Todo 内容和 CAS 校验。
…ing a Goal The Goal dialog only showed what the composer had chosen, so a draft in a directory the core had not granted could not be created at all. The dialog now lets the person pick the agent (those that can hold a Goal), the permission and one of the granted directories, starting from the first granted one when the composer's is not. The granted-directories endpoint now says where each one is on this machine. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: maxliux5 <5958865+maxliux5@users.noreply.github.com>
The two were the wrong way round: Codex had the warm near-white page and terracotta accent, Claude the cool white. Codex is now white and neutral greys with an ink accent; Claude is ivory and warm greys with its clay. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: maxliux5 <5958865+maxliux5@users.noreply.github.com>
…rth opening The steward's three buttons floated in their own row under the topbar, and the conversation scrolled inside a column narrower than the pane, so its scrollbar sat in the middle of the page. The buttons now sit at the topbar's right edge like a thread's, and the conversation scrolls at the pane's edge with its text still in the reading column. The menus get a heading with the count, one quiet row per Goal (a dot, the name, what it is on), and a way to All Goals; the brief shows its three numbers side by side. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: maxliux5 <5958865+maxliux5@users.noreply.github.com>
…eir group "Running 5" and "Needs you 0" in the steward's menus, and the three numbers in the brief, now open All Goals filtered to that group. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: maxliux5 <5958865+maxliux5@users.noreply.github.com>
The Chinese name for starting a conversation reads 新对话 everywhere it appeared as 新线程: the sidebar button, the empty-list hint and the directory and section hints. The browser tests look for the new name. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: maxliux5 <5958865+maxliux5@users.noreply.github.com>
Reject mismatched retained debit identity before replay or receipt repair. The shared TypeScript reader checks persisted bindings and billing terms for modern, legacy file and inline evidence. Both existing CLI entrypoints retain legitimate historical omissions and repair a missing receipt from the original debit without charging twice. Exact reviewed head: 75d0ccd. Independent exact-head approval: review 5920595494. Approval closeout is clear. Validation: 267 CLI cases passed (10 platform skips), 49 TypeScript spend/void cases, control-plane typecheck and standard premerge (6 direct + 18 selected) passed. Independent review additionally passed 45 runtime transaction cases and 20 isolated wheel cases covering same-second, nested and inline contradictions with unchanged retained bytes. No intervening target-main changes affect the five reviewed paths. Required bilingual diff and hygiene checks pass at the unchanged head. The configured workflow skips hosted CI waiting. The owner authorized related reviewed fixes to merge; any admin bypass applies only to GitHub's author-approval/branch policy. Native Windows, real PostgreSQL, installed-host adoption and complete C1 cost/benefit qualification remain separate. Signed-off-by: LoopX Agent <337587101+loopx-agent@users.noreply.github.com>
Signed-off-by: LoopX Agent <337587101+loopx-agent@users.noreply.github.com>
… connection (#8876) * fix(effect-runtime): retire an idle runtime that still holds a silent connection server.close() waits for every connection. A caller that connected and died without sending a request left the idle runtime alive but no longer listening, while its locator stayed published, so every later CLI call failed until the process was killed by hand. Co-Authored-By: Claude <noreply@anthropic.com> Signed-off-by: xiaobaifly7 <196511160+xiaobaifly7@users.noreply.github.com> * test(effect-runtime): reap the exited runtime child before probing it On POSIX an exited but unreaped child still answers the liveness probe. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: xiaobaifly7 <196511160+xiaobaifly7@users.noreply.github.com> --------- Signed-off-by: xiaobaifly7 <196511160+xiaobaifly7@users.noreply.github.com> Co-authored-by: xiaobaifly7 <196511160+xiaobaifly7@users.noreply.github.com> Co-authored-by: Claude <noreply@anthropic.com>
* fix(canary): retain timeout diagnostics and keep incomplete checks failed Signed-off-by: LoopX Agent <337587101+loopx-agent@users.noreply.github.com> * fix(canary): require a completed ratchet failure for advisory Signed-off-by: LoopX Agent <337587101+loopx-agent@users.noreply.github.com> * fix(canary): require a complete maintainability report Signed-off-by: LoopX Agent <337587101+loopx-agent@users.noreply.github.com> --------- Signed-off-by: LoopX Agent <337587101+loopx-agent@users.noreply.github.com>
* fix(quota): fence historical ACK writeback order by GoalRef Same-goal instances share one rollout log. The historical ACK reader bound the first matching refresh, so a retired instance's earlier writeback could clear writeback_follows_admission and drop the current same-second ACK. Keep only the quota owner's own events and run-index rows. Signed-off-by: Duang777 <duangjl007@gmail.com> * fix(quota): ignore another instance's ACK during replan recovery A retired instance's newer confirmation was compacted without goal_ref, so receipt recovery treated it as this owner's latest ACK and stopped the original replan. Keep goal_ref on the run projection, drop unowned recorded ACKs before rebuilding the open review, and use the supplied candidate list even when that list is empty. Signed-off-by: Duang777 <duangjl007@gmail.com> * fix(quota): move receipt ACK selection out of the frontier builder The receipt-owned ACK choice and empty-list fallback stay on the semantic history reader. The frontier builder keeps its existing decision-point budget. Signed-off-by: Duang777 <duangjl007@gmail.com> * fix(quota): keep progress-review context when restoring a replan Rebuilding the open review after a retired ACK must use the same assist policy and effective-turn cadence as the original status projection. Signed-off-by: Duang777 <duangjl007@gmail.com> --------- Signed-off-by: Duang777 <duangjl007@gmail.com>
Signed-off-by: LoopX Agent <337587101+loopx-agent@users.noreply.github.com>
Signed-off-by: LoopX Agent <337587101+loopx-agent@users.noreply.github.com>
Signed-off-by: LoopX Agent <337587101+loopx-agent@users.noreply.github.com>
Signed-off-by: LoopX Agent <337587101+loopx-agent@users.noreply.github.com>
Signed-off-by: LoopX Agent <337587101+loopx-agent@users.noreply.github.com>
…pts (#8647) * fix(scheduler): keep damaged history behind one receipt fence The Python compatibility route and the native owner now agree on a single freshness rule. A damaged row can no longer retire a valid ACK or failure return, a committed successor still blocks an older return when only its summary rotted, and the originating Turn plus Goal instance scope decide both before the write and again under the state lock. Re-applied on top of the merged #8708 minimal passthrough. This commit keeps only the extra hardening: tolerant scheduler reads with strict settlement and status reads, committed-successor scope proof, instance-range checks, and the lock-time re-read before replay. Verified on the current base with the six TypeScript suites (248 tests), the control-plane type check, and the Python receipt-fence (30), follow-up hint transport, single-request and rollout-log suites (53). The merged #8708 regression file passes 17 tests on the same tree. Windows, the installed host UI, PostgreSQL and hosted CI were not exercised. Signed-off-by: gcl-coder <gcl-coder@users.noreply.github.com> * fix(scheduler): read an unreadable successor identity as superseding A committed quota_should_run row can lose one byte of a nested display string (the opening or closing quote of a details value) while every identity member and the terminating newline stay intact. The tolerant scanner resynchronised on the next plausible quote, swallowed the members that followed the damaged nested value and reported the row as unrelated, so an older heartbeat return regained write authority and both compatibility commands wrote again. Receipt recovery now keeps the row's container state while it reads members. A damaged byte where a value must start is that value's opening quote; a damaged byte inside a string is the lost closing quote only when the separator its role requires follows it, and a lost key name is reported as such. Recovery also reports whether a separator or boundary was unknowable. A same-scope row whose identity cannot be attributed now answers "unknown", and the freshness check treats "cannot prove the successor is unrelated" as stale. Rows whose readable identity proves another Agent, another Goal lifetime or another event kind stay tolerated, and a diagnostic row with all key names intact still cannot supersede a Turn. Verified with real File and SQLite scheduler-ack-current and scheduler-fail-current commands: opening-quote, closing-quote and in-text damage of a committed turn-2 receipt all exit 1 with heartbeat_receipt_stale, scheduler_state_mutated=false and unchanged log bytes, while another Agent's damaged row still lets the current turn write. TypeScript receipt log, commit, followup and CLI suites pass; the Python receipt-fence suite passes on File and SQLite; control-plane typecheck passes. Windows, hosted UI and hosted CI were not exercised here. Signed-off-by: ChengLong GU <88359731+gcl-coder@users.noreply.github.com> * fix(scheduler): keep a duplicated scope identity from hiding a successor A stray rewrite can repeat a top-level scope member of an already committed quota_should_run receipt (for example agent_id) while the only damaged byte is in its display summary. Recovery now reads every value a member carries, so a duplicate that repeats the same identity merges into one scope instead of discarding the whole row. Duplicates whose values disagree keep the row inside this scope and answer "unknown", which the freshness check already treats as stale; only a row whose every scope value is foreign stays tolerated, together with diagnostics whose key names survived. Verified with real File and SQLite scheduler-ack-current and scheduler-fail-current commands: a committed turn-2 receipt with a repeated agent_id and one damaged summary byte exits 1 with heartbeat_receipt_stale, scheduler_state_mutated=false and unchanged log bytes, while a foreign Agent's row with the same shape still lets the current turn write. The full Python receipt-fence suite passes on File and SQLite, the six scheduler TypeScript suites pass, and control-plane typecheck passes. Signed-off-by: ChengLong GU <88359731+gcl-coder@users.noreply.github.com> * fix(scheduler): only a readable foreign identity may exclude a successor A stray rewrite can duplicate a top-level scope member of a committed quota_should_run receipt, name another Agent first and leave this scope's copy where a single damaged byte makes it unreadable. The recovery reader reported both the readable foreign value and the damaged key, but the successor check trusted the foreign value alone and returned "unrelated", so an older Turn regained write authority. The exclusion branch now consults the same recovered key list: a member whose every readable copy is foreign still proves the row unrelated only when no copy of that member is unreadable. If the same member is also damaged, the row answers "unknown", which the freshness check already treats as stale. Fully readable foreign rows and surviving diagnostics stay tolerated. Verified with real File and SQLite scheduler-ack-current and scheduler-fail-current commands: a committed turn-2 receipt whose first agent_id names another Agent while the trailing local copy holds a raw 0xff now exits 1 with heartbeat_receipt_stale, scheduler_state_mutated=false and unchanged log bytes. The mirrored order (local readable first, foreign copy damaged) also keeps the old Turn retired. The full Python receipt-fence suite passes (62 tests, File and SQLite), the six scheduler TypeScript suites pass (99 tests), and control-plane typecheck passes. Signed-off-by: ChengLong GU <88359731+gcl-coder@users.noreply.github.com> * fix(scheduler): finish the scope check before reporting uncertainty The previous commit treated an unreadable duplicate of a scope member as uncertainty immediately, before examining the remaining members. A row whose event kind was readable as a validation event and whose duplicate copy was damaged therefore returned "unknown" even though its Agent identity was complete and clearly another Agent's. That refused valid ACK and failure writebacks on File and SQLite. committedSuccessor now examines every identity member first. A member whose every readable copy is foreign and whose own copies are all readable settles the row as unrelated, regardless of damage in a different member. Uncertainty stays local to the damaged member: a readable foreign value with an unreadable twin of the same member still cannot exclude the row, and the unreadable copies of this scope's members still answer "unknown". Verified with real File and SQLite scheduler-ack-current and scheduler-fail-current commands: another Agent's validation row with a readable event kind and one damaged duplicate of that member no longer blocks the current writeback (exit 0, scheduler_state_mutated=true, log bytes unchanged), while this scope's damaged duplicates still return heartbeat_receipt_stale with zero writes. The full Python receipt-fence suite passes (66 tests, File and SQLite), the six scheduler TypeScript suites pass (99 tests), and control-plane typecheck passes. Signed-off-by: ChengLong GU <88359731+gcl-coder@users.noreply.github.com> * fix(scheduler): read only committed rows when deciding freshness A log write can stop between its complete JSON row and the newline that commits it. The snapshot exposed that final record, so a nearly written successor retired the old Turn and a nearly written original counted as the Turn's receipt. The next append drops those bytes, which made the same command flip from stale or missing to writable. The snapshot now names the unterminated record line, and the scheduler freshness read ignores that row for both the originating Turn and any successor. The two Python append entry points also deduplicate against newline-terminated rows only: retrying an interrupted append drops the abandoned bytes and commits the row once, while a committed row still returns the same record. General diagnostic reads keep the decodable tail. Verified with real File and SQLite scheduler-ack-current and scheduler-fail-current commands: an unterminated turn-2 successor no longer retires turn-1 (exit 0, scheduler_state_mutated=true), and an unterminated turn-1 row reports heartbeat_receipt_missing both before and after the next append drops it. The full Python suite passes (106 tests), appended-row dedup covers both entry points, the scheduler TypeScript suites pass (105 and 157 tests), and control-plane typecheck passes. Signed-off-by: ChengLong GU <88359731+gcl-coder@users.noreply.github.com> * fix(scheduler): ignore unterminated rows in prior-turn recovery A complete JSON row without its committing newline is abandoned by the next append. Use the same committed-only view for strict settlement, receipt qualification and prior-host-Turn recovery so an interrupted receipt cannot make a later admission ask for recovery before the next append and then disappear. Verify File and SQLite real CLI prior-turn recovery with committed and unterminated guard receipts, plus TypeScript settlement preflight and control-plane typecheck. Signed-off-by: ChengLong GU <88359731+gcl-coder@users.noreply.github.com> * Use committed receipts for tolerant settlement history Filter unterminated scheduler settlement rows through the existing committed rollout view in quota settlement readback and replanned history. Add TypeScript and real File/SQLite CLI regressions for interrupted tails while preserving damaged-row tolerance. Signed-off-by: ChengLong GU <88359731+gcl-coder@users.noreply.github.com> --------- Signed-off-by: gcl-coder <gcl-coder@users.noreply.github.com> Signed-off-by: ChengLong GU <88359731+gcl-coder@users.noreply.github.com>
…und Stop hooks (#8874) * fix(antigravity): guide fixed-session in-place schedule patrol and bound Stop hooks Signed-off-by: LoopX Agent <337587101+loopx-agent@users.noreply.github.com> * fix(antigravity): fix exact-surface unbind readback, default registry hooks, and skill isolation Signed-off-by: LoopX Agent <337587101+loopx-agent@users.noreply.github.com> --------- Signed-off-by: LoopX Agent <337587101+loopx-agent@users.noreply.github.com>
Signed-off-by: LoopX Agent <337587101+loopx-agent@users.noreply.github.com>
Signed-off-by: LoopX Agent <337587101+loopx-agent@users.noreply.github.com>
…structural moats (#9035) Signed-off-by: LoopX Agent <337587101+loopx-agent@users.noreply.github.com>
#8955) Verify #8577 (77eece8) and #8879 (3d6fd2e) on main against DSH pause/unsettled-recovery regressions across file and sqlite providers: - Paused Goals (goal_stopped and compute_quota_zero) must keep should_run=False and pause_cause without projecting unsettled_host_turn_recovery, setting closeout_required=True, setting a settlement_effect_id/recovery_effect_id, or consuming the bounded 3-admission recovery budget. - Resuming restores unsettled_host_turn_recovery with admissions_without_progress == 0. Closes #8505 Closes #8563 Signed-off-by: LoopX Agent <337587101+loopx-agent@users.noreply.github.com>
…livery (#9040) Align developer workflow docs and PR review/merge skills so managed Goals that configure wait_for_ci=false explicitly pass --registry and --goal-id across worktree/subagent invocations and complete review, readiness, and eligible self-merges locally without polling asynchronous GitHub CI. Signed-off-by: LoopX Agent <337587101+loopx-agent@users.noreply.github.com>
…plore_visual_sink (#8925) * refactor(lark): slim explore_results below 1,350 lines and extract explore_visual_sink Signed-off-by: LoopX Agent <337587101+loopx-agent@users.noreply.github.com> * fix(lark): preserve dynamic visual sync status in _sync_packet Signed-off-by: LoopX Agent <337587101+loopx-agent@users.noreply.github.com> --------- Signed-off-by: LoopX Agent <337587101+loopx-agent@users.noreply.github.com>
) * refactor(lark): slim kanban presentation and extract board module Signed-off-by: LoopX Agent <337587101+loopx-agent@users.noreply.github.com> * fix(lark): filter null view_ids during partial kanban setup Signed-off-by: LoopX Agent <337587101+loopx-agent@users.noreply.github.com> --------- Signed-off-by: LoopX Agent <337587101+loopx-agent@users.noreply.github.com>
…pair-8725-legacy-assignment-20261119 Signed-off-by: LoopX Agent <337587101+loopx-agent@users.noreply.github.com>
Signed-off-by: LoopX Agent <337587101+loopx-agent@users.noreply.github.com>
Signed-off-by: LoopX Agent <337587101+loopx-agent@users.noreply.github.com>
|
This pull request has merge conflicts with Choose the remote for the base repository, not an out-of-date fork. git fetch upstream
git rebase upstream/main
# Resolve each conflict, git add the resolved files, then git rebase --continue.
git push --force-with-lease origin HEADFor a same-repository clone whose Keep the DCO |
loopx-agent
left a comment
There was a problem hiding this comment.
Reviewer: model_agent · gpt-6.2-sol · OpenAI (runtime_reported, max)
Request changes conclusion (author-owned PR; GitHub blocks formal self-review)
[P2] 已提交的旧指派在丢失 Chat 阶段后无法从原卡恢复
旧版 Chat 完成创建和原生指派后,在保存 todo_assigned 阶段前中断。切换到当前 head 打开同一原卡,File 和 SQLite 都连续两次返回 HTTP 409 并要求恢复原来源。原注册库、创建回执和已指派的 Todo 仍存在,重试没有写入新提交。 相关代码
动机 / Motivation
维护者确认创建并指派 Todo。进程若在提交后中断,需要从原卡找回已完成的操作。新实现把原 Todo、注册库和回执绑到同一来源,并支持找回已经留存的原执行。仅丢失 Chat 阶段时,原卡应读回已证实的历史和当前负责人,并避免重复提交。本次以隔离的 File 和 SQLite 验证;真实飞书、PostgreSQL 升级和安装态浏览器仍待验证。
改动思路 / Approach
Chat 保留确认和阶段。原 Goal 来源路由核验注册库和存储,原生 Todo owner 负责回执、指派和 revision 检查。按实际共同祖先核对 43 个净改动,与上一受审版本相比有 20 个净路径变化。服务端与原卡展示两端都有变化。用旧版真实 HTTP 分别在指派前和原生指派提交后中断,再用当前 head 对同一个原提案重试两次,核对回执、负责人和提交数。
具体改动 / Changes
精确 head 4ed3a60432f084a35902bbaaf25f327dc9e4361f,diff base 41ea75e89892ed88d0ae543e97e31aad0a181c47,43 个文件 +2610/-372。逐项复核 source-bound-write、interrupted-original-assignment、retained-task-execution。验收约定
关键代码讲解
- 按原生创建回执及前序 revision 验证同一来源。源码
- 缺少 Chat 指派阶段时,已指派的当前记录会与创建时的未指派记录比较。源码
- 有明确原操作 ID 的 managed claim 可以证明已提交。源码
- 原提案卡继续承担打开恢复和操作入口。源码
对主干的风险 / Risks
在 File 和 SQLite 上,旧卡只差保存指派阶段时会持续收到 409。原创建回执和 agent-a 都仍在。同一情形在最新主干能读回 200,但会额外提交一次指派。指派尚未发生的对照在新头只写一次并正常恢复。14 项相关 Python 与 64 项 TypeScript 通过,另有 1 项 TypeScript 跳过。扩大 Python 复验时本地卷空间不足,当前只采用已完成的范围。
语义与 CI 对齐
新版保留原注册库和 provider 验证。仅凭缺失的阶段和变化后的任务快照无法判断已经发生的操作。最新主干 5cfd6ae 试合并在 tests/control_plane/test_native_user_completion_update.py 有冲突。远端 CI 按本轮设置留待既有流程。
我的整体评价 / Verdict
请求修改。原生证据能绑定原确认时读回原指派;历史证据不够时,在原卡分别展示已证实的创建和当前负责人。把需要新写入的操作继续交给现有来源和 revision 检查。
复看了上一受审版本的 1200px 和 390px 首屏。桌面保留原提案卡,窄屏恢复抽屉将「重试原操作」放在展开诊断前。 本次失败从实际 HTTP 复现。当前 head 的 409 状态、展开诊断和长任务混排尚无独立浏览器复核。
English verdict: REQUEST_CHANGES. The original completed native assignment becomes unrecoverable after the legacy Chat checkpoint is lost. File and SQLite return 409 twice while retaining the original creation receipt and assignee.
| # Assignment changes the Todo snapshot after creation. A completed | ||
| # assignment only needs receipt recovery. An unfinished one still | ||
| # requires the current Todo to match the originally reviewed write. | ||
| if not assignment_completed: |
There was a problem hiding this comment.
这里在没有 Chat 指派阶段时,拿创建回执中的未指派快照比对当前 Todo。旧确认已经提交原生指派但来不及记录 todo_assigned 时,正常的负责人变化会让真实 File/SQLite 原卡连续两次返回 409。请优先从原确认绑定的证据恢复指派。历史证据不足时,保留已验证的创建和当前负责人供原卡读回,新写入继续校验来源和 revision。
When Chat loses a Todo or Monitor confirmation, the user can reopen the original card and recover its original proposal, Todo and receipt. The canonical writer checks the selected Goal's source and reviewed provider before creating work. The prepared confirmation retains the reviewed source, first schedule and exact operation for replay.
A retained Todo confirmation from before source binding uses the original native create receipt and the reviewed revision from that store's history. If the Todo remains unchanged and the Agent is still registered, retry completes assignment on the original Task. If the source or assignment cannot be verified, the original card shows the recorded Todo and gives the user a recovery path after refresh. A prepared or committed operation keeps its original confirmation. An unsubmitted stale preview can still be regenerated.
Chat 确认中断后可从原卡继续。升级前留下的确认以原创建回执和历史来源为依据,恢复同一个 Todo 的指派。来源无法验证时,原卡仍显示已经记录的任务和恢复指引。新格式确认继续复用原操作和首次排期。
Review range
The exact source head is
4ed3a60432f084a35902bbaaf25f327dc9e4361f. The currentmaincomparison at41ea75e89892ed88d0ae543e97e31aad0a181c47contains 43 files, +2,610 and −372. The stack includes the original #8717 work. The GitHub PR page still uses snapshotf3bdad238739and displays 1,822 files. The full verified comparison is linked here. Verified compare Prior reviewHistorical validation at
193e6d53d8Visual evidence
Packaged Chat with synthetic records persisted by the previous server. The previous head keeps the old confirmation on disk and shows an empty activity view. The candidate exposes that confirmation and its retry in the existing Chat layout.
193e6d53d8· File · 1200 × 854193e6d53d8· SQLite · 390 × 854Legacy assignment recovery at
97bcd88b0dAn older Chat confirmation can have a saved assignment even though the caller lost the final response. The original card now verifies its historical creation receipt, source lineage and matching saved assignment. A pending assignment still checks the current Todo and Agent registration before writing.
41ea75e898. The six direct and 19 selected premerge checks completed with one inherited maintainability advisory. That revision still required a change-quality receipt and independent review. The existing Use the selected Goal runtime for Chat task reads and writes #8717 changes are retained in the branch.Original task execution recovery at
4ed3a60432Confirmations retained from 2.3.0 can now recover the original completed Session and Turn after upgrade. Confirmations with a newer managed claim can recover the same claim while its execution proof is current. A missing claim, a changed owner or lease, and a changed Todo before any claim still prevent a new dispatch.
升级后,从原确认卡可读回同一个已完成的 Turn。已有执行认领时,待启动的请求继续检查当前 Goal、认领者和租约。未取得认领且任务被外部修改时,仍会挡住新启动。
cqr_69bb86b5266974276d6a. The repository premerge gate passed with 6 direct and 19 selected checks executed. It classified 1 failure as an inherited advisory. Independent review of the current head is pending. The live desktop App, Feishu and PostgreSQL recovery of a retained 2.3.0 Chat record remain unverified.