test(workspace): restore delivery-check recovery browser journeys

[loopx-agent]

Goal And Delivered Outcome

The Team recovery browser smokes still searched for 重新读取证据 after #6473 renamed the action to 重新检查交付 to disclose that it runs configured checks. The packaged-browser failure reproduces on current main; three adjacent recovery journeys fail at the same obsolete accessible name.

Point all four journeys at the current action and reuse the existing recheck locator within the evidence-return smoke. The complete failure/recovery assertions remain intact. Intended base: main; related: #6473. This fixes test discovery of the existing operation.

Author Declaration

• Written by: model_agent (Codex, GPT-6 family, OpenAI).
• Implemented against: Live Team Workspace RFC, revision 7c1f12a47c75a39cb37514e26b3570f3ef22f3f3.

Criterion	Disposition	Existing path / decisive validation
§7: retain current authority and recovery boundaries	implemented within the test repair	Original feedback identity/deduplication, unchanged Host-call counts and immutable history assertions retained and passed
§9: show stale, rejected and lost observations honestly	implemented within the test repair	Four packaged recovery journeys now reach and pass their original withdrawal/recovery assertions
§9: real-model objection → independent acceptance → requester adoption	deferred, unchanged	These browser/fixture runs do not qualify the full live-team outcome

Self-check: compared the actual component action with all active smoke callers, reproduced all four old-name failures, reviewed the complete diff and inspected generated desktop/390px screenshots. The small future-facing pass reuses one existing locator; separate fixtures need no shared abstraction. No new capability, protocol vocabulary or decision owner.

Scope And Continuation

Complete within this test-only scope: evidence return, uncertain-feedback recovery, original-task revalidation and result-use recovery. Product UI, default behavior, backend and first screen have identical source blobs to the base. Full model/installed-App/L1 qualification remains in the existing feature scope; this proposal does not close it. No new successor is needed for the selector repair.

Validation

• Tested revision: 66d1bbae178db514ea7feb7632d891d1d0df96c5 test bytes; the production bundle was built from 7c1f12a47c75a39cb37514e26b3570f3ef22f3f3, whose product sources are unchanged by this PR.
• Run state: finished. Input class: synthetic.

Check kind	Result	Evidence / limitation
regression_parity	passed	All four base journeys reproduced 30s old-name locator failures; all four fixed journeys passed without timeout changes or removed assertions
real_entrypoint	passed	uv run --extra test python scripts/chat_bundle.py build --install and verify --source; actual packaged Team UI
integration	passed	team-evidence-return-smoke.mjs: checked output versions, stale/revoked source/consumer, recovery, keyboard return, list pagination, 390px/reduced motion, no work dispatch
real_backend	passed	team-feedback-recovery-smoke.mjs: disposable production inbox HTTP/store, lost acknowledgement, same-id retry, scoped input and paused admission
real_backend	passed	team-task-revalidation-smoke.mjs: disposable production Chat/delegation worker, original failure and response-loss recovery without a second Host call
real_backend	passed	team-result-use-recovery-smoke.mjs: disposable production Chat/delegation worker, source withdrawal/restoration, unchanged history and no redispatch
static	passed	Node syntax checks for four files, git diff --check, public-boundary scan; semantic advisory found no supported vocabulary carriers (MJS is outside that probe)
manual	passed	Generated desktop/390px screenshots inspected; no visual change introduced
integration	not_run	Full hosted CI and PostgreSQL suite; no product/provider/store code changes. No installed-App upgrade or paid model evaluation performed

The bundle build emits the existing chunk-size warning; the build succeeds. Initial failures are retained as before/after evidence. Production authority is exercised only by the named disposable backend fixtures; the evidence-return transport fixture is presentation coverage, not independent semantic acceptance.

Frontend / Visual Evidence

UI impact: none. Existing desktop/390px layouts, keyboard focus and failure/recovery screenshots were inspected during the smokes; there is no before/after presentation delta or first-screen change. The existing single explicit check keeps its effect-disclosing label and adds no user interaction.

Type of Change

• Test update

LoopX Area

• Build, packaging, installer, or CI

Direction: S5/R2 live-team evidence/recovery qualification. Shared-authority RFC fixture impact: N/A; no state rule, provider or compatibility change.

Boundary Checklist

• Public-safe diff and PR body; private state, raw logs, local paths and generated screenshots excluded.
• No duplicated benchmark work.
• Scope limited to the reproduced selector failures.
• UI impact marked none.
• Commit carries DCO sign-off.

[loopx-agent]

Reviewer: model_agent; model=gpt-6.1-sol; provider=OpenAI; declaration_source=runtime_reported; reasoning_effort=max

Approval conclusion (author-owned PR; GitHub blocks formal self-approval)

动机

使用团队工作区的用户需要在证据失效、反馈回执丢失或原执行恢复后，再次检查同一份交付；维护者需要浏览器回归真正走到这些恢复断言。旧脚本仍寻找“重新读取证据”，而实际按钮已改为“重新检查交付”，因此在检查版本和恢复结果以前就超时。当前提案修正定位名称，使四条既有旅程重新执行完整断言，不改变用户操作或执行权限。本 PR 不证明真实模型纠偏、独立语义验收、请求方采用或已安装 App 的完整链路。

评审目标为 6546@66d1bbae178db514ea7feb7632d891d1d0df96c5，实际基线 7c1f12a47c75a39cb37514e26b3570f3ef22f3f3。先读取同版本 docs/architecture/rfcs/live-team-workspace-v0.md、设计和前端交付规范，再读实现；依据为 §7 的权限及兼容边界、§9 的真实状态与恢复要求。准则 V1、V3、V6 在本测试修复范围内得到覆盖，V4 的真实持续协作仍是既有功能验收，不能由合成脚本代替。经验建议仅用于核实实际恢复路径，没有继承旧结论或声称记忆带来净收益。

改动思路

GoalTeamEvidence 已用效果更明确的“重新检查交付”描述现有 read 操作，操作仍经过原来的验收及当前授权边界。此处只让测试调用方跟上这个名称。没有重命名产品按钮、扩大匹配范围、增加等待时间或删除失败断言。

我比较了不改、修复既有定位器、引入公共测试封装三种路径：不改会持续在错误入口停止；现有定位器已足够；为六个名称引用增加框架反而扩大维护成本。team-evidence-return-smoke.mjs 复用本地 recheck，另外三份独立 fixture 保留各自的后端与恢复语义。当前边界是修复四个测试调用方，生产组件、状态所有者和执行控制保持原样。

还核对了相关工作：#6547 已于 2026-10-15 合并，当前 main 的四个文件与此提案逐一同 blob。因此这是正确但已经被主干吸收的等价修复，建议维护者关闭重复提案；不需要为了交付相同结果再合并一次，也不需要增加后继任务。

具体改动

完整差异只有四个文件、6 行替换及 1 个签署提交。team-evidence-return-smoke.mjs 的 recheck 改为精确匹配当前按钮，并在采用撤回和核心版本恢复两处复用。team-feedback-recovery-smoke.mjs 的 command 仍读真实收件箱；其恢复点击只更正名称。team-task-revalidation-smoke.mjs 和 team-result-use-recovery-smoke.mjs 的 recheck 仍调用原 HTTP 及 delegation owner，没有另造状态或校验器。

在正常构建的基线页面上，未修改的 evidence-return 脚本于第 67 行等待旧名称 30 秒后失败；正常构建当前头后，四份完整脚本全部通过。对反馈旅程，提交后故意丢失响应，导航及刷新不发送新反馈，显式重试保留原身份、正文和版本，真实 store 中只有一条输入；暂停后仍拒绝新投递。对原任务恢复，失败保持可见，修复及提交后丢响应的恢复保留原操作且 Host 调用计数仍为 1。对结果使用，撤回源版本会同时清除报告，恢复后读取当前版本，历史保持不变，三条 Host 计数均不增加。

返回旅程还保留错版本与缺版本、核心证据及下游采用丢失、显式恢复、键盘返回、分页游标、390px 和 reduced motion 的原断言。源文件逐项比较确认所有断言、超时、fixture 及产品代码都未改变；图片只展示合成内容，桌面及手机截图已查看。

对主干的风险

当前头的构建、verify --source、四条打包旅程、语法检查与语义 advisory 均通过。原生 canary premerge --from-git-diff 针对实际基线得到 3 项直接检查及 5 项目录选中检查通过，失败和人工 hold 均为 0。基线旧名称失败作为回归敏感性证据保留，没有把它隐藏成当前通过，也未查询或等待 GitHub CI。

没有新配置、协议词汇、状态判断、默认开关、提示或权限。默认关闭、只读导航及拒绝路径仍由同一产品代码及真实 fixture 验证；浏览不会启动工作，收件箱 delivered 也没有被解释为 applied。证据返回脚本的业务观察为合成 API；另外三份脚本使用一次性生产 HTTP/store 或 delegation worker，不能扩大为真实模型、PostgreSQL 或原生 App 资格。此次没有产品视觉或首屏变化。

我的整体评价

APPROVE：测试调用方的修复正确，全部受影响恢复断言得到当前头的独立执行。 有界维护性检查已经应用：复用现成 recheck 即可，没有必要抽象不同 fixture。相关 #6547 已交付完全相同的文件，因此建议关闭此重复 PR；本结论不授予合并权限，也不把测试通过当成 §9 完整协作验收。

English verdict: APPROVE - 6546@66d1bbae178db514ea7feb7632d891d1d0df96c5. All four packaged recovery journeys pass without weakened assertions; the old-name baseline fails as expected. The identical fix is already on main via #6547, so closing this duplicate is preferable to another merge. No live-model or installed-App qualification is claimed.