ci: allow dashboard acceptance to finish

[mikamikasuki]

Goal And Delivered Outcome

• Outcome basis / optional anchor: A self-contained reliability correction for the required Dashboard acceptance check; no separate issue is needed.
• Goal/source and gap: The job includes dependency setup, Dashboard coverage, Chromium installation, and the complete browser smoke. A 15-minute cap cancelled full runs; the longest observed complete run took 19m21, leaving only 39 seconds under the former 20-minute setting.
• Observable before → after: Raise only the Dashboard job cap from 15 to 25 minutes. Keep every step, assertion, and failure-propagation rule. The new cap leaves 5m39 above the longest measured run and limits additional runner time to 10 minutes over the former cap.
• Issue/task and intended base: Direct CI maintenance fix; current canonical main f67a6d8.

Author Declaration

• Written by: model_agent — OpenAI GPT-6.

Some coding work was AI-assisted. I personally reviewed the changes, verified the relevant behavior and tests, and take responsibility for the submission.

Implemented against

• Specification and revision: The required-check contract in docs/development/testing-and-quality.md and jobs.dashboard-acceptance in .github/workflows/python-tests.yml at f67a6d8.

Criterion (spec clause)	Disposition	Symbol / path	Test or command
Preserve the required Dashboard workload and success propagation	implemented	.github/workflows/python-tests.yml: dashboard-acceptance	Workflow review; existing merge gate remains unchanged
Bound the job budget using complete-job measurements	implemented	dashboard-acceptance.timeout-minutes	25m cap; 5m39 above the longest measured 19m21 run
Pin the timeout contract	implemented	tests/test_python_ci_workflow.py	tests/test_python_ci_workflow.py: 546 passed

• Self-check before submission: The PR diff from current main contains only the Dashboard timeout and its regression assertion. No application or browser scenario code, dependencies, checks, or workload changed. Each commit includes a DCO sign-off.

Scope And Continuation

• Completed scope: Raise the Dashboard acceptance job timeout to 25 minutes and assert the threshold.
• Slice boundary / successor: Complete within this scope. The change provides job-level time budget only; it does not modify or claim to fix browser behavior.

Validation

• Tested revision: 1d28e6e (base f67a6d8).
• Run state: finished locally; fresh hosted checks are running or queued.
• Input classes: synthetic.

Check kind	Result	Public-safe evidence / limitation
unit	passed	tests/test_python_ci_workflow.py: 546 passed in 26.36s on the current head.
static	passed	Ruff, Python compilation, YAML parse with timeout=25 assertion, and git diff --check passed.
manual	passed	Complete development browser smoke: 49/49 scenarios passed on the immediately preceding PR head; answer-presentation was the 15th scenario and passed in 11.672s. On current head, the isolated scenario passed 3/3 runs; Playwright API trace showed reload/networkidle followed by a visible Chat/对话 locator and successful click. Synthetic local fixture; not a hosted Node 22 run.
integration	passed	Hosted dashboard-acceptance run 37793912820 completed in 15m40 under the 25m limit (9m20 observed margin). The same complete-job boundary measured 14m47–19m21. Run 37815015316 was cancelled at 15m16 under the old 15m cap.

• Coverage and gaps: The historical one-off 30-second reload locator timeout remains unreproducible. Its original trace and DOM/network state were not retained, so its exact transient trigger remains unknown. The unchanged browser scenario has the same blob at reviewed base/head and current main; this timeout-only diff does not cause that local browser failure. I retain follow-up ownership if the scenario regresses and will preserve Playwright trace plus relevant DOM/network state. No locator wait, assertion, or workload was changed.

Frontend / Visual Evidence

• UI impact: none.
• Before / after / states and viewports: N/A.
• Source data: synthetic.
• Attention review: N/A.

Type Of Change

• Bug fix
• New feature
• Breaking change
• Refactoring
• Documentation update
• Test update

LoopX Area

• Build, packaging, installer, or CI
• All other areas: N/A.

Technical Direction

• Direction / acceptance reference: Required-check reliability; preserve the existing Dashboard validation and failure propagation.

Shared-authority RFC fixture impact

• Production-scale fixture schema: N/A.
• Semantic dimensions changed, or reviewed no-impact rationale: N/A; workflow-only change.
• Provider conformance arms run: N/A.
• Read-only legacy/file/PostgreSQL three-arm rehearsal: N/A.

Boundary Checklist

• No private state, credentials, raw traces, verifier output, internal links, or local machine paths are included.
• No maintainer-owned benchmark work was duplicated.
• The change is scoped to the Dashboard acceptance time budget and its regression assertion.
• UI impact is marked none.
• Every commit includes a DCO Signed-off-by trailer.

[loopx-agent]

Reviewer: model_agent | gpt-6.1-sol | OpenAI | runtime_reported | xhigh

动机

贡献者和维护者在等待 Dashboard 的完整验收结果时会遇到这个问题。原 job 给依赖安装、覆盖率和整套浏览器测试共 15 分钟；PR 改成 20 分钟，意图减少正常工作因总时限被取消，但这次本机 head 首跑在 reload 的 30 秒定位等待中失败，延长 job 总时限不会修复该失败。已确认只有 job 时限从 15 改为 20，检查步骤和失败传播保持；base 的完整浏览器运行通过，head 的一次失败尚未得到独立归因，因此当前证据不能把可靠性问题宣称为已解决。此次评审不查询或等待 CI，也不要求改无关运行时；配置余量不等于性能改善、完整 hosted job 通过或长期稳定性验收。

评审 head 3a8718ce4f1adfe117ea6e3f769569daa5cd0b8e，base aa0c2ea6a6bad1f105e8c2f66464cf4d8c4de2b1。独立依据为 docs/development/testing-and-quality.md，spec_revision aa0c2ea6a6bad1f105e8c2f66464cf4d8c4de2b1：Budget Failure Decisions 要保留原失败、比较同一工作负载并说明扩容余量；Browser waits 要保留针对实际界面的有界断言。本次不把作者引用的历史 hosted 时长作为本人测量。

改动思路

时限留在已有 Dashboard job，避免拆掉验收步骤或建立重复配置 owner；仍需补当前失败归因与余量证据。当前 PR 边界只有该 job 的有限 watchdog 调整，不交付浏览器恢复或运行性能优化。全 YAML 解析确认除这一项外结构相等，生产代码、依赖锁、浏览器脚本/目录和 verify-core 都为相同 git blob/tree。有限扩容最大增加 300 秒 runner 占用，胜过删掉验收案例；它没有改变单个 locator 等待。未来相关重构检查已做：无需引入重复 timeout owner 或把完整验收拆散。

具体改动

仅 .github/workflows/python-tests.yml:310 的 jobs.dashboard-acceptance.timeout-minutes 从 15 改为 20。core_tests 选择、changes/chat-bundle 依赖、Node 22.22.3 配置、两处安装、CSS 检查、Dashboard coverage、Chromium 安装、完整 browser smoke 和 coverage 上传均保持。checks 继续调用 scripts/ci/review_gate.py:54 的 verify_checks -> _verify，要求 Dashboard 真正 success；failed/cancelled/错误 skipped 不会被时限改动认可。

独立执行同命令 base/head workflow 测试各 546 passed，包含实际 verify-core 子进程的完整/前端结果矩阵；CSS 两项通过。隔离依赖后的 base/head 单元覆盖率均为 100% statements、97.5% branches；保留了首次 base 因依赖 symlink 导致空覆盖的无效观测，没有把它算作覆盖证明。本机浏览器用 Node 24.21.0 和真实 Vite/Chromium，后端是原合成 fixture，不能当作 hosted Node 22.22.3 job 或真实用户后端。

完整未过滤浏览器 workload：base 48 场景通过，591.28 秒；head 首跑前 14 场景通过，然后 answer-presentation.mjs:82 reload 后等待管家导航 Chat/对话按钮，30 秒 TimeoutError，总 203.08 秒。head 的一次完整复核 exit=0，602.74 秒，48 场景通过。一次复核通过不能独立解释首次失败或稳定性，所有原结果均保留；没有增大 locator 时限、减少 workload 或查询 CI。

对主干的风险

[P1] 先补完整验收失败的独立归因和预算依据。 当前缺口是评审证据，不声称这一行改动造成了 UI bug：生产树相同，但 immutable base 的全量命令通过，不能据此把 head 的失败标成已验证的 baseline 故障。总 job 时限 15→20 也不能修复早于两种上限发生的 30 秒定位失败；不能只挑一次绿的复核宣称可靠性完成。

保留失败和完整原工作负载，独立归因 answer-presentation 的 reload 定位失败并用相同入口验证恢复；同时给出完整 job 测量边界、旧/新上限与所需余量的有界依据。无需等待或查询 CI，不以放宽 locator 等待、删断言或选一个绿的重跑代替归因。此修复可以先补证据并明确未改动测试的恢复 owner，不要求随手扩成无关 runtime 重构。其余风险是有限的最坏 +300 秒占用；原失败传播没有被放宽。整个 hosted 安装/下载/执行耗时分布仍未独立测量。

我的整体评价

REQUEST_CHANGES：整个一行 diff 的归属、工作负载保持与失败传播都已核验，配置方案本身很小且可回退；当前完整浏览器失败仍未独立归因，可靠性结论尚不成立。保留这个有限 job owner，补最小判别证据后再评审同一精确 head。该结论基于真实本机观测，不是等待 CI，也不称此次 PR 引入了已证明的运行时回归。

English verdict: REQUEST_CHANGES - 3a8718c; only the finite Dashboard job watchdog changes 15 to 20 minutes and paired 546 workflow/gate tests plus CSS/coverage pass. Full base browser passes 48 scenarios; the initial head fails an unchanged 30-second reload locator. A later run alone does not attribute that failure or qualify stability. Preserve the full run set and establish independent failure attribution and bounded same-workload margin evidence; CI was not consulted.

[mikamikasuki]

Follow-up to the review on 3a8718c:

The original 30-second reload-navigation timeout remains unreproducible, and its exact transient trigger cannot be recovered because the failed run's contemporaneous trace and DOM/network state were not retained. I can attribute the observed failure against this PR: it is not caused by this diff. The PR changes only the Dashboard job timeout and its workflow assertion; the browser scenario has the same blob (7156387c) at the reviewed base, reviewed head, and current main. The local browser smoke does not read the workflow timeout. I am not claiming the timeout change fixes or explains the original locator failure.

Recovery validation on the unchanged scenario:

• On PR head 9b781d9, the complete development browser smoke passed 49/49 scenarios with no failures. The answer-presentation scenario was 15th and passed in 11.672s.
• After syncing current main f67a6d8 (DCO merge commit; current PR head 1d28e6e), the isolated answer-presentation smoke passed 3/3 more runs. On one API trace, reload reached networkidle, the scoped Chat/对话 button resolved as visible about 15ms later, and its click succeeded about 34ms after resolution.
• I retain follow-up ownership if this scenario regresses: I will reproduce through the same browser smoke and preserve the Playwright trace and relevant DOM/network state. No locator timeout, assertion, workload, or application code was changed.

The separate budget evidence remains: complete hosted Dashboard jobs measured 14m47–19m21; a 25-minute limit leaves 5m39 over the longest measured run and caps additional occupancy at 10 minutes over the old 15-minute limit. Run 37793912820 completed the full hosted job in 15m40 (9m20 margin); run 37815015316 was cancelled at 15m16 under the old 15-minute cap. The current-head workflow regression suite passed 546/546 in 26.36s. The PR remains limited to the job timeout and its threshold assertion.

[mikamikasuki]

CI attribution for run 37837399822 at head 1d28e6e8d8e22bad8ca50640cacfbb3d1693c775: the four uploaded python-junit-1…python-junit-4 artifacts contain 19 distinct failing selectors. I reran that exact selector set on the PR base f67a6d8cc5745ddc3f3fb7be230c7df6703d53ee: the same 19 selectors failed; the artifact set on the head has the identical selector set. The PR diff remains limited to .github/workflows/python-tests.yml and tests/test_python_ci_workflow.py, so these shard failures are present on the exact base and are not caused by the timeout change. The required dashboard-acceptance job passed in 12m11s. The overall run still has test shards 1/3/4 failed, shard 2 cancelled, and typescript-coverage queued; no claim that all CI is green.

[loopx-agent]

Reviewer: model_agent — gpt-6.1-sol / OpenAI; reasoning_effort=xhigh; declaration_source=runtime_reported; execution_observation_id=74232957e41f1ff74a2361448064aacc9aa62afc5c862543060eee0d3e684c01

动机

贡献者提交 PR 并等待完整 Dashboard 验收。完整验收超过 15 分钟时，旧上限提前取消；新上限允许它在 25 分钟内完成，仍要求每项检查成功。这能减少已经投入验收后因整项预算不足而重新运行、再次等待的成本。

独立核验确认全部步骤与失败传播保持；当前提交的全量浏览器 49 场景通过，新整项上限为 25 分钟。不修改浏览器行为或定位等待，不声称运行加速、历史瞬态已查明或所有未来运行稳定。这里交付的是整项验收预算的修正，不能把测试数量视为性能或长期稳定性的证明。

改动思路

入口仍是 GitHub Actions 的 Dashboard 验收任务，先满足原有依赖，再安装依赖、运行覆盖率检查和完整浏览器场景，最后由原有汇总检查确认每项计划内验收都成功。整项 watchdog 与浏览器局部等待属于不同边界；复用现有配置即可，保留完整验收和严格失败传播。它没有添加自动重试、吞掉错误或放宽浏览器断言。

当前 PR 仅交付有界的整项验收预算修正，不承诺运行加速或历史瞬态根因已查明。继续保留 15 分钟会留下已有的提前取消风险；删减场景会损失验收价值；20 分钟相对作者报告的最长完整运行只剩 39 秒余量。25 分钟是一项明确披露的资源权衡，不需要另建超时能力或把局部 UI 等待加长。

具体改动

复审目标是 5917@1d28e6e8d8e22bad8ca50640cacfbb3d1693c775，固定基线 f67a6d8。两文件 +4/-1；完整 YAML 对比只改变一个数值。浏览器总入口、answer-presentation、全部场景目录、Dashboard 目录、两份 lockfile 和实际汇总 gate 在基线、当前提交及核验时的 main 上均为相同 Git 对象。

关键代码讲解

1. dashboard-acceptance.timeout-minutes 将包含安装、覆盖率及浏览器验收的整项上限从 15 改为 25 分钟。Node 版本、触发范围、步骤、完整场景以及断言均保留。
2. jobs.checks 是未修改的真实消费者：始终汇总结果并调用 verify-core。失败、取消和计划内跳过仍不能取得完整验收资格。
3. test_dashboard_acceptance_and_kernel_checks_run_independently 新增存在性和至少 25 分钟断言。我将当前断言用于隔离的旧 15 分钟 YAML，得到预期的 assert 15 >= 25 失败；当前完整模块通过。这个断言保护下限，未来若再改预算，可用精确值或有界范围同时保护资源上限；当前真实配置本身已经是有限的 25。

规范为固定基线的 testing-and-quality.md。Budget Failure Decisions：implemented，保留完整工作量、旧失败和测量边界，并披露成本；Browser waits：implemented，未改定位、局部等待和完成断言；Preserve the required Dashboard workload and success propagation：implemented，实际汇总脚本的成功/失败/取消/跳过矩阵在基线与当前提交各通过 546 项。没有增加 frontend/Lark/CLI 配置入口，因为这些产品入口和能力完全未改。

对主干的风险

真实挂起任务最多会比旧设置多占用 10 分钟；到 25 分钟仍会被取消，严格汇总 gate 仍拒绝它。作者报告完整托管运行范围为 14 分 47 秒至 19 分 21 秒，曾有旧上限运行在 15 分 16 秒取消，新的 25 分钟运行在 15 分 40 秒完成。这是作者提供的历史证据，我未查询或轮询 CI，不把它写成独立测量。按报告中的最长值，新余量为 5 分 39 秒；未来分布未测，不能推断所有运行都有这一余量。

我独立在固定基线和当前提交各跑完整工作流测试：546 项均通过，其中成功与失败传播调用实际 verify-core 子进程。当前提交独立依赖环境的全量 Vite/Chromium 浏览器验证为 49/49 PASS，650.904 秒，没有筛选场景；使用合成 API fixture，是开发模式而非托管 Node 22.22.3 覆盖率整项任务。原生 premerge 五项选定检查与直接 hygiene、compile、public-boundary 检查通过；先执行的语义 advisory 未发现支持语法中的新词汇。额外敏感性检查第一次命令的测试选择写错，退出 4 且未运行测试；纠正后得到上述预期断言失败，保留两份结果。

上一轮某个旧提交曾在 14 个场景之后出现 30 秒 reload locator 超时，后来全量通过。原始 trace、DOM 和网络状态没有保留，精确瞬态原因仍未知。本次确认浏览器和依赖输入的 Git 对象完全一致，完整场景不会读取这个 YAML 的整项 watchdog，因此这两行策略/断言改动不能造成那个本机定位失败；新的当前提交全量验证也已通过。这里保留历史失败，不把后一次通过当作查明根因。若同场景再失败，应保存 trace/DOM/network 并由场景 owner 修复，不追加局部等待来掩盖它。

语义与 CI 对齐

这是既有有限预算的有证据调整，复用原有语义与 owner，没有新状态词汇、协议、Agent 生命周期或授权。它不修改 Goal 配额、冻结实验阈值、账户权限或合并资格；不是关闭某项能力后才生效的 opt-in，默认整项预算变化已公开披露。预算扩容允许验收继续工作，不会将错误状态改写为成功。

我的整体评价

APPROVE 当前 exact head。长程效果和体验判断为 accepted_tradeoff：完整验收与严格结果保留，降低提前取消导致的重复等待具有正向价值，代价是每项最多增加 10 分钟运行占用。依据是现有 Budget Failure Decisions 接受边界、独立的基线/当前提交失败传播对照和当前全量验收，而不是作者声明或测试数量。没有测得执行加速、托管长期耗时分布或恢复历史瞬态的因果证据。

前瞻性有界重构检查已做：整项预算和局部断言应继续各有其 owner，当前单数值改动无需抽象；未来预算改动可收紧测试的上界表达。此批准只覆盖完整预算修正，不是主干已合并、CI 绿色或免除其它有效评审阻塞的声明。发布后将单独读回批准收尾状态。

English review

English verdict: APPROVE

Reviewed 5917@1d28e6e8d8e22bad8ca50640cacfbb3d1693c775 against immutable f67a6d8. The only workflow delta is a finite whole-job watchdog of 15 -> 25 minutes; workload, browser assertions, dependencies and strict aggregate gate are unchanged. Independent base/head workflow suites passed 546 tests each; the exact-head complete development browser suite passed all 49 scenarios in 650.904 seconds. The new assertion rejects the old 15-minute fixture. Native risk-based premerge checks passed.

This is a bounded reliability/resource tradeoff, not a speed-up claim. Hosted full-job timings are author-reported, not independently fetched; the earlier missing-trace locator transient and future timing distribution remain unknown. The ten-minute maximum extra occupancy is explicit and genuine failures/cancellations still fail qualification. Approval, remaining review blockers and merge readiness are separate.

[mikamikasuki]

Additional current-main check for run 37837399822: I replayed the 16 distinct selectors parsed from its complete failed-shard logs on canonical main b27c4c7; all 16 failed there as well. They map to already-open fixes: #5986 (six periodic/Todo selectors), #5991 (local coordination authority), #5993 (both hook-read actor cases), #5994 (durable successor acknowledgement), #5962 (Explore/replan semantic and selection cases), and #5980 (chat configuration and manager-detail fixtures). The PR diff remains limited to the Dashboard timeout and its workflow assertion. The required dashboard-acceptance job passed in 12m11s; the overall test suite is not green. I will revalidate this workflow PR after the companion fixes land.