test(chat): align operation fixture and recovery smoke

[LIHUA919]

Goal and gap

The standalone operation fixture emits prepared and delivered before confirmed, while its Python assertion expected only four later stages. In exact-head CI, the Personal Workspace browser smoke also timed out waiting for a recovered Todo card. That scenario armed its transient preview failure after reopening Goal Chat, allowing recovery to race the injection; its 5-second card deadline was tight for the history retry's 3-second backoff under CI load.

Change

• Assert the complete six-stage operation fixture lifecycle.
• Arm the transient preview failure before reloading, verify it was exercised, and give recovery up to 15 seconds to display the card after retry.
• Preserve the existing product UI, persisted state and authority behavior. Signed merge d701822be integrates current main 850268bff; the contribution diff is limited to the two test files.

Validation

• tests/test_chat_operation_actions.py: 41 passed on the integrated source.
• Chat bundle build and verification passed.
• Focused chat-todo-proposal real browser scenario passed in development with coverage and in the packaged Chat build.
• The complete development browser suite was attempted: 16 scenarios passed, then team-evidence timed out waiting for its active-mode label; that scenario passed when run alone. This separate load-sensitive failure remains unclassified, so the complete suite is not claimed green.
• node --check and git diff --check passed. Exact-head remote CI passed the four Python shards and dashboard acceptance, but aggregate pytest failed while reporting coverage for a disposable semantic-probe fixture (No source for code). #5440 owns that narrow repair. CI is red; maintainer review and merge remain pending.

中文

独立操作夹具会先输出 prepared、delivered，原 Python 断言却只接受后四个阶段。上一精确版本 CI 还在恢复后的 Todo 卡片可见性检查处超时：浏览器场景直到重新打开 Goal Chat 才启用一次性预览失败，可能与恢复完成竞争；在 CI 负载下，5 秒等待也不足以覆盖历史重试的 3 秒退避。

本 PR 校正六阶段断言，并在重载前启用预览失败，确认故障确实被触发，给恢复重试后的卡片最多 15 秒显示。产品界面、持久化和权限行为未修改。签署合并 d701822be 纳入最新 main 850268bff，相对主干仅改动两个测试文件。

整合源码上的 Python 文件测试 41 项通过；Chat 打包构建与校验通过；chat-todo-proposal 在带覆盖率的开发浏览器和打包浏览器均通过。完整开发浏览器套件曾运行，前 16 个场景通过，随后 team-evidence 等待运行状态标签超时；该场景单独运行通过，因此不能宣称完整套件已通过。语法和差异检查通过。精确版本远端 CI 的四个 Python 分片与 dashboard acceptance 已通过，但覆盖率汇总读取已清理的语义探针临时源码时失败（No source for code）；独立的 #5440 负责该修复。CI 仍为红色，维护者评审与合并待完成。

[Duang777]

Exact-head CI attribution for d701822beb6892cdf7d9062b24bcc5ccc6779730 on current main@850268bffc6f7123e88f578d3743f8f6da5c5c73:

• Run 36958279407 passed all four Python shards, Windows PowerShell, Dashboard acceptance, every TypeScript job, and all other direct checks.
• The only direct failure was the pytest coverage aggregate. Coverage combined all four shard files, then stopped on No source for code: /tmp/.../test_cli_inspects_disjoint_ind0/loopx/a.py.
• test(ci): isolate semantic probe fixture coverage #5375 fixes that inherited coverage leak by removing the outer pytest-cov environment from the synthetic subprocess. Its earlier exact-head run 36825867638 passed every job.
• Current-main run 36914904063 independently fails the stale four-stage operation-fixture assertion that this PR updates. The same main run also hit the unrelated Windows process-cleanup WinError 5 seen intermittently on other branches.

The clean order is to review and merge #5418 first, then sync #5375 so its coverage fix runs with the six-stage assertion. #5409 and #5389 can sync after those baseline fixes land. No merge action was taken.

[Duang777]

Updated CI triage for d701822beb6892cdf7d9062b24bcc5ccc6779730:

• All four Python shards, Windows, TypeScript, Stage 2C, and static checks passed on the existing head.
• The pytest aggregate failed only while combining coverage from a deleted synthetic checkout; merge-gate then failed downstream.
• Main now contains the coverage-isolation fix through perf(authority): copy journal JSON without repeated primitive allocation #5251 (d290f8ecf), so the previous test(ci): isolate semantic probe fixture coverage #5375/test(chat): align operation fixture and recovery smoke #5418 dependency cycle is resolved.
• This PR still needs its owner to sync current main@3dbde34e15122b31d447a8a640139dd676e41d11 and rerun CI. No changes were made to this PR branch.

[huangruiteng]

No blocking finding. No product behavior, authority or protocol change. The remaining uncertainty is the reported load-sensitive browser race: the immutable base also passed the focused browser scenario in this run, so I do not claim to have reproduced that timing failure or qualified the entire browser suite.

动机

本次按 pull_request_review_execution_contract_v2（policy 13）独立评审完整 base-to-head：850268bffc6f7123e88f578d3743f8f6da5c5c73 → d701822beb6892cdf7d9062b24bcc5ccc6779730。目标是修复已有验收的陈旧期望和故障注入竞争，不是新增恢复机制。实际独立 fixture 已输出六阶段；基线的四阶段断言失败，因此修正期望有明确仓库价值。相比新增 runner 或重写生产恢复 owner，延续两个已有测试是更小、完整且可逆的修复。

改动思路

失败开关在 reload 前启用，避免恢复完成后才注入；卡片等待上限从 5 秒调整到 15 秒，以容纳已有历史重试，而非修改产品退避或承诺新的性能指标。新增断言要求开关已被消费，防止场景“绿色但没有测试失败”。Python 测试继续用 -S 执行独立 fixture，不引入测试框架依赖。

具体改动

• test_browser_operation_fixture_needs_no_test_framework（tests/test_chat_operation_actions.py:81）校验 prepared → delivered → confirmed → waiting → unknown → reconciled，保留 Goal、executor 和未知结果/对账断言。浏览器 confirmed-operations.mjs 也实际消费前两阶段，并非为适配偶然输出添加字段。
• chatTodoProposalScenario.run（examples/personal-workspace-browser/chat-todo-proposal.mjs:124–150）仅改变故障注入位置、等待预算及消费检查；仍检查唯一幂等键、原 Goal、用户确认及后续 reload 不重复 apply。
• 已沿生产 projectRecoveredTurnProposals 和 completed-history effect 检查恢复路径：读取已有 action，失败返回未完成并由既有有限退避重试；preview 不授予 apply 权限。模拟 API 只提供输入/503，恢复与展示执行真实前端代码。

独立本地验证：

• 精确 head 的 uv run --extra test python -m pytest -q tests/test_chat_operation_actions.py：41 passed；同命令不可变 base：40 passed / 1 failed，仅上述陈旧四阶段断言失败。
• 原生 chat-todo-proposal 浏览器验收：base、head 开发版均通过；head 打包版通过。head npm run build:chat 和 uv run --extra test python scripts/chat_bundle.py verify --source 通过。
• 独立负向探针只在内存中绕过 fixture 的故障分支、保留未消费开关，产品代码未改；真实 Chromium 场景准确在 Recovery did not exercise the transient preview failure 断言失败。这证明新增检查能拒绝“故障未发生”的伪通过，不证明历史负载问题已复现。
• 改动 Python 的 Ruff、JavaScript 语法检查和 whole-diff whitespace 检查通过。

对主干的风险

贡献 diff 两文件 +8/-3，都是现有测试；无生产状态分类、默认开关、通用控制面措辞、guidance/obligation 或 actor authority 变更。未新增并行决策 owner，不需要语言迁移。长期重复恢复、离开 Goal 后返回及下一次 reload 的既有幂等/确认断言保留，用户操作路径和权限保持不变。

浏览器证据边界是“真实开发/打包前端 + 隔离合成 API/stream”，不是真实模型、部署后端或已安装版本验收。未跑完整 pytest、完整浏览器套件或 PostgreSQL；没有读取、等待或用远端 CI 代替本地评审。作者描述的全套浏览器失败和覆盖率问题不宣称已由本 PR 修好。

耐久性/重复扫描：检查了两个 immutable revision 的 fixture、相邻恢复场景、现有 runner、package smoke 命令和 Python workflow 调用，并查看作者近期 30 个 PR（包括 #5440 的独立 coverage 边界）。本改动复用既有生产验收，没有新建同形 smoke 或临时工具。future-facing pass 考虑了是否需要按请求绑定故障计数器或 TS 抽象；当前局部场景已有明确时序和消费断言，没有足以支持额外共用机制的缺口，暂不扩展。

我的整体评价

APPROVE。明确的陈旧断言修复，加上能经负向探针验证的既有恢复场景加固，构成完整测试维护结果；不把更大的等待值等同于产品性能修复。不替整个恢复路线、全套 CI 或上层路线图宣布完成。批准后应单独读回 effective blocking reviews，只有逐项独立确认旧 finding 已解决且拥有撤销权限时才撤销；本结论不授权 merge。

English verdict

English verdict: APPROVE

APPROVE — independently reviewed exact head d701822beb6892cdf7d9062b24bcc5ccc6779730. The stale lifecycle expectation is reproduced at the immutable base and corrected at head. Native development and packaged browser acceptance pass; an independent guard mutation demonstrates that the new assertion rejects an unexercised transient failure. The base browser also passed, so the historical timing failure is not claimed reproduced. This is existing-test maintenance, not a production latency, authority, protocol or default-policy change. Synthetic API/stream boundaries and unrun whole-suite/backend validation are explicitly retained. No merge authorization.