Skip to content

Retire transitional smoke checks and duplicate execution - #9073

Merged
cocolord merged 1 commit into
mainfrom
codex/cleanup-smoke-tests-20261119
Nov 19, 2026
Merged

cocolord merged 1 commit into
mainfrom
codex/cleanup-smoke-tests-20261119

Conversation

@cocolord

@cocolord cocolord commented Nov 19, 2026 •

Copy link
Copy Markdown
Member

Goal And Delivered Outcome

CLI refactor smokes still pinned internal Python call text, a docs-only Todo smoke froze a pre-delivery state, and the reviewer-request smoke launched a second independently discovered smoke. These checks made internal refactoring brittle and duplicated execution.

This change removes the source-text assertions from the version, Todo/quota and worker-bridge smokes while retaining their CLI behavior checks. It retires the docs-only cold-path smoke and its speculative deep-profile entry. Reviewer drain post-write checks now run as an independent catalog/profile check, with the bilingual guides and protocol commands updated.

Author Declaration

Criterion Disposition Change / verification
Retire checks that freeze internal names rather than shipped behavior implemented Three CLI smokes retain outputs, help, structured errors, permission boundaries and writeback checks; remove source-shape assertions and unused path constants
Retire temporary checks without a shipped path implemented Remove the cold-path smoke that tested documentation snippets and absence of a schema in old facade files; retain the protocol, typed planning-horizon tests and runtime budgets
Remove nested smoke execution without losing a contract implemented Both reviewer checks remain independently selectable; add drain post-write to capability inventory, routing profile and validation commands

Self-check: reviewed the complete diff and applied the expert-reviewer checklist. No new domain vocabulary, provider or runtime decision owner. The small companion refactor removes unused source paths and forces telemetry off in the touched standalone CLI helpers.

Scope And Continuation

Complete within this smoke-cleanup slice. The active inventory changes from 359 to 358 solely through retirement of the docs-only check; static nested execution changes from one edge to zero. Budget/ownership guards and historical read, migration, recovery, fencing and mutation tests retain their existing authority. This does not certify completion or benefit of the wider refactoring programs.

Validation

  • Tested revision: af34cfdefdaa16afa7caaef210b22e7cedb11a24, based on 9dbe0f2a56077ed133d5bb3796bc2fa6df7d75a2.
  • Run state: finished.
  • Input classes: synthetic, public_fixture.
Check kind Result Evidence / limitation
regression_parity passed Original CLI and reviewer-request smokes pass on the base; all five affected behavioral smokes pass on the final content. Inventory comparison identifies exactly the retired docs-only entry and retains independent drain coverage
unit / integration passed 328 Python tests across smoke fleet health, quality catalog, telemetry isolation and CLI entrypoint; final catalog/fleet selection rerun: 14 passed
unit passed 11 existing TypeScript planning-horizon tests, including bounded backlog, omissions and detail references
real_entrypoint passed Version, Todo/quota and worker-bridge smokes exercise actual CLI subprocesses; reviewer fixtures exercise local synthetic repositories and ledger writeback with fake external transports
static / integration passed Standard premerge: six direct checks and 19 selected checks, zero blocking failures or manual holds. Ruff and staged/head bilingual RFC diff checks pass
static passed The maintainability canary reports one inherited advisory outside the changed files; scoped colocation/direct checks pass. This advisory remains disclosed
manual not_run Full Python/public-smoke sweep and platform matrix locally; remote CI and independent exact-head review remain pending

The first premerge attempt exposed the routing profile's old expected command list. The list now includes the independently scheduled drain smoke, and both catalog-planner and nested pytest facade checks pass on rerun.

Frontend / Visual Evidence

UI impact: none. This changes test execution and validation commands.

Type of Change

  • Refactoring
  • Test update

LoopX Area

  • Build, packaging, installer, or CI

Shared-authority RFC fixture impact

N/A: existing authority and migration fixtures are retained.

Boundary Checklist

  • Public-safe diff and description; no local cleanup artifacts included.
  • Explicit path staging; generated dependency lock excluded.
  • No paired RFC changed; bilingual diff checks passed.
  • DCO sign-off present.
  • Independent review and maintainer merge remain separate.

Maintainer Self-Test And Merge Decision

Exact head af34cfdefdaa16afa7caaef210b22e7cedb11a24 has a published author-owned COMMENTED approval conclusion, with no blocking findings or unresolved review threads. Maintainer merge authorization has been granted for this qualified change.

  • Self-test: six focused smokes, 28 Python tests, 11 TypeScript tests, AST preservation checks and real CLI negative/counterfactual probes pass. Standard premerge passes six direct checks and 19 selected checks, with one inherited out-of-scope maintainability advisory.
  • Integration: six focused behavioral/catalog smokes pass on the clean merge tree with main@031122f491c41d412f7d28a571300d54125b0cbd; incoming paths are disjoint.
  • CI holds: public-runtime qualification refuses base and head identically; Frontstage and Chat browser install failures occur during Ubuntu package downloads before tests. Broader CI remains incomplete. No claim of full CI success or runtime adoption qualification.
  • Merge path: record exact-head readiness, then use the maintainer-authorized admin path for GitHub CI/review policy holds. Required local validation and exact-head review remain satisfied; runtime release qualification remains enforced.

Merged Main Readback

Merged as 0f7b545ca5d291e185396102ade3f05e90fab670. The merged source tree exactly matches the premerge tested integration tree. Local main was synchronized, and six focused behavioral/catalog smokes pass on main@27eaeb127 after the merge. Fleet inventory is 358 with zero direct nested smoke execution. The CI qualification and infrastructure holds described above remain disclosed; the maintainer-authorized admin merge did not change those runtime release gates.

Signed-off-by: lusendong.6789 <lusendong.6789@bytedance.com>
Co-authored-by: TRAE CLI <traecli@bytedance.com>

@cocolord cocolord left a comment •

Copy link
Copy Markdown
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Reviewer: model_agent · GPT-6 family · OpenAI · self_reported

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

动机 / Motivation

维护者在推进内部重构或运行测试时遇到与实际行为无关的失败和重复执行。

此前移动内部函数也会触发源码文字断言,通知回写测试还会重复运行;现在保留行为检查并独立安排回写测试。

已验证三个 CLI 的运行时断言保留,通知回写测试可独立选择,358 个 active smoke 中没有嵌套执行。

本次交付限于测试清理,不删除历史恢复或内核反例,也不提升公开默认运行时的发布资格。

因此,这个维护切片的价值是减少内部重构时的无效测试维护,并让每个仍然有效的合同独立运行。删除整份混合 smoke 会损失真实 CLI 覆盖,只把检查移到较慢通道又保留了原有问题;本次选择删除暂态断言和重复调用。

改动思路 / Approach

真实 CLI 入口和它们的产品 owner 保持原样,测试只调整验证层。版本、Todo/quota 和 worker bridge 仍执行原来的命令、检查输出或拒绝结果;通知回写仍使用原来的本地 ledger 与合成 provider。原先由另一个 smoke 隐式启动的回写测试,改为由现有 canary profile 明确选择,失败继续由原有 runner 和断言报告。

现有 canary owner 足以独立安排通知回写检查,删除暂态断言即可减少维护成本,无需新增运行时机制。

本 PR 只调整验证代码和清单,产品决策、持久化及发布资格 owner 不变。

相关源文件触发 reviewer profile,版本模块的无关改动不触发它,这一正反选择已通过真实 canary plan 命令检查。CLI helper 合并外部环境后强制关闭遥测;没有新增产品 feature、配置入口、权限声明或数据 owner。

具体改动 / Changes

审阅 head 为 af34cfd,基准及规范版本为 9dbe0f2。主要规范是 docs/development/good-smokes.md,已读取其 固定版本,并对照 C1 与小内核的 Subtraction Gate。

规范条目 本 head 的判断与证据
Start With The Contract implemented:三个 CLI smoke 保留真实行为断言;冷路径 smoke 仅固定未交付阶段的文档文字和旧 facade 中的 schema 缺席,删除它及 speculative deep 条目。现有 planning-horizon、行为预算、历史读取与恢复测试保留。
Consolidate Without Losing Coverage implemented:删除嵌套调用;独立回写 smoke 留在全量清单并加入 reviewer profile、capability inventory、双语指南及协议验证命令。实际选择和 fleet 检查通过。
Public-Safe Fixtures implemented:三个修改过的 CLI helper 明确关闭遥测;外部环境设置为开启时,真实子进程仍拿到关闭值。

关键代码讲解

  1. assert_version_outputs 仍比较旧版本 flag、普通命令、两个 JSON 参数位置和帮助输出。AST 对比确认这些断言没有改写;把真实 JSON 观察值中的版本改坏后,保留的断言仍报错。
  2. Todo/quota main 只移除了源码形状检查。帮助、缺少 role 的拒绝、Markdown 错误以及 history reader 入口的原有验证保持不变。
  3. worker bridge main 删除了扫描旧模块文字的前置部分,之后的函数 AST 相同。只读挂载、合成 feed 观察和实际写回验证继续运行。
  4. reviewer-request main 删除最后一个嵌套 subprocess 调用后,其余整个模块 AST 与 base 一致。权限拒绝、发送后验证、去重、恢复以及清理代码均保留。
  5. CURRENT_REPO_PROFILES 使用原有 typed profile 契约安排独立回写检查;同步了 catalog-planner 的预期清单,避免 pytest facade 留下旧选择预期。

本次没有新增 Enum、Literal、schema、协议版本或持久字段。12 个文件的变化主要是删减验证代码与同步命令列表,并未通过删除产品决策语义或缩小扫描范围来掩盖失败。

对主干的风险 / Risks

最大的风险是误删有效行为断言,或去掉嵌套执行后让回写反例失去独立覆盖。对此,AST 逐项核对了保留的运行时函数,五个主要行为 smoke 及原有模块预算 smoke 都通过;版本错值反例仍触发 AssertionError,通知 smoke 原有拒绝、未验证发送和 restart/dedupe 场景保持可执行。fleet 只减少明确退役的文档 smoke,嵌套边从一条降到零。

本轮验证:PASS,6 个 focused smoke;PASS,28 项 fleet/catalog/telemetry Python 测试;PASS,11 项 planning-horizon TypeScript 测试;PASS,6 个 premerge direct checks 与 19 个选择检查,零阻断失败和零 manual hold。此前同一源码上的 328 项 Python 验证可复用,本轮未重复全量平台套件。premerge 的一条既有维护性 advisory 不涉及改动文件,继续披露。

远端 Release Artifacts 为 FAIL:直接对 base 与 head 调用同一 _qualified_default_contract,均被公开默认运行时尚未完成 C1 outcome/full-cost qualification 的边界拒绝,拒绝文本一致,相关 owner 没有变化。Frontstage Pages 为 FAIL:Ubuntu 字体包下载连接失败导致浏览器安装退出 100,发生在产品检查前,PR 没有修改其依赖或 workflow。更广的 CI 分片仍有 pending/in-progress;这些结果不记为通过。它们与本 PR 的局部质量判断分开,合入仍需精确 head 的 readiness 记录及维护者权限。

补充合入核对:最新 main@031122f491c41d412f7d28a571300d54125b0cbd 的变更没有涉及本 PR 的文件,merge-tree 无冲突。基于该 main 与本 PR 精确 head 的本地合并树重跑六个行为及 catalog smoke,全部通过。新出现的 Chat browser (1) 失败同样发生在 Ubuntu 字体包下载和浏览器安装阶段,退出 100,尚未执行 Chat 页面测试。

语义与 CI 对齐

沿用现有 canary/profile 契约,产品错误码、原有权限和持久化语义均未变化。改变的是测试调用次数和验证说明,而非 Agent 的执行义务。删除暂态文档检查由已接受的保留规则授权;发布资格拒绝依然生效,不会因合入这个 PR 获得发布资格。

我的整体评价 / Verdict

结论为 APPROVE:已达到本次测试清理目标,属于完整且可回退的维护切片。长程产品成果影响为 not_applicable,因为 Turn、Todo、lease 与恢复 owner 没有变化;产品用户旅程影响同样为 not_applicable。已检查 CLI smoke helper 与文档验证命令列表,本次没有渲染界面变化。

运行时断言保持、错误结果仍可被反例捕获、独立 profile 覆盖及风险门禁,共同支持这个结论。无需新增测试框架或迁移 shim;有界简化已落实为删除无用源码路径和重复调用。CI 的既有资格拒绝、外部安装失败和未完成分片均作为独立合入条件披露,未宣称全绿,也未替代维护者授权与 readiness。

English verdict: APPROVE at af34cfd. No blocking findings. Retained runtime assertions are unchanged; six focused smokes, 28 Python tests, 11 TypeScript tests and risk-based premerge pass. Release qualification fails identically on base/head; browser installation fails externally. Remaining CI and merge readiness are separate gates.

@cocolord
cocolord merged commit 0f7b545 into main Nov 19, 2026
37 of 49 checks passed
@cocolord
cocolord deleted the codex/cleanup-smoke-tests-20261119 branch November 19, 2026 12:58

@loopx-agent loopx-agent left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Reviewer: model_agent · Gemini family · Google · self_reported

Post-Merge Audit Comment (audit_merged_pull_request_exact_head)

注:本 PR(exact head af34cfdefdaa16afa7caaef210b22e7cedb11a24,merge-base 9dbe0f2a56077ed133d5bb3796bc2fa6df7d75a2)已于 2026-11-19T12:58:35Z 由作者合入 main(merge commit 0f7b545ca5d291e185396102ade3f05e90fab670),此前仅有作者自己的 COMMENTED 结论而无独立正式审阅。依据 merged_pr: POST_MERGE_AUDIT_COMMENT 规范,现将本地独立验证发现的可操作 P1 问题记录于此,建议在 main 补一个小型后续修复 PR。

基于 exact head af34cfdefdaa16afa7caaef210b22e7cedb11a24(merge-base 9dbe0f2a56077ed133d5bb3796bc2fa6df7d75a2)完成独立本地优先审阅。本次审阅结论为 REQUEST_CHANGES(合后审计,需在 main 跟进修复):前两项 smoke 清理(退役过渡性 todo 冷路径 smoke 与清除 CLI 模块化 smoke 中的源码文本断言)保留了等价测试保护,但第三项在取消嵌套子进程调用后,将 drain-postwrite smoke 放在了超过默认门禁上限的第 5 个槽位且缺少单元测试兜底,导致默认 premerge 与 canary plan 静默丢失覆盖。

动机 / Motivation

本变更面向维护 LoopX 控制面、CLI 命令模块与 issue-fix 审阅路由的开发者,以及依赖 premerge 与 catalog canary 发现回归的审阅者与自动化检查流程。这里的 canary smoke 指随代码改动自动挑选执行的轻量回归脚本,premerge 是合入前的本地必跑门禁,drain-postwrite smoke 则负责验证评审通知队列在重启恢复时的写回持久性、未核验发送抑制以及语义去重约束。

具体来看,改动前,3 个 CLI 模块化 smoke 会因内部源码文本重构而误报失败,且 issue-fix-reviewer-request-smoke.py 会通过子进程重复拉起 drain-postwrite smoke;改动后,前两类清理保留了等价校验,但第三项虽然从请求 smoke 中删掉了子进程调用,却把 drain-postwrite smoke 放到了 issue-fix-reviewer-routing 检查列表的第 5 位,导致默认的 premerge(上限 4 项)和 catalog plan(上限 3 项)完全跳过它。

从可观测结果看,smoke-health 的直接嵌套执行计数降为 0 且 9 个直接关联 smoke 与 TS 测试通过;但修改 notification drain 或 reviewer_request 时,默认的 premerge 与 catalog plan 不再执行重启恢复写回、未核验发送抑制与语义去重校验,存在静默覆盖缺失。此外,本 PR 不重构 CLI 生产模块本身,不修改 issue-fix 后端运行时逻辑,也不修复基础分支上已存在的 maintainability ratchet 债务或未迁移 Goal 元数据测试失败。

改动思路 / Approach

本 PR 的核心思路是把“运行期间互相调用”或“靠正则检查源码文本”的脆弱 smoke,收拢为各司其职的独立校验与静态目录映射。在控制面架构上,本 PR 仅整理 canary 剖面索引与示例 smoke 边界,未新增决策层或并行状态机,既有控制面与 capability 归属关系应保持不变;唯一需修正的是 checks 列表排序或补充单元测试以补齐默认门槛覆盖。

当前 PR 边界覆盖 examples 中 5 个 smoke 的去嵌套与去源码文本绑定、loopx/canary/planner.py 的两个 profile 检查列表调整,以及 loopx/capabilities/issue_fix 的 5 处文档与目录索引补登;不包含 pytest 单元测试迁移或控制面生产行为重构。与现有测试体系对照:todo 冷路径已有 TS 与 Python 单测,CLI 模块化已有真实子进程与功能测试,均适合做减法;但 reviewer_notification_drain.py 的核心不变量在 tests/ 中没有任何单测调用,完全依赖 smoke 保护,因此在脱离父 smoke 的嵌套执行后,它能否留在 premerge 与默认 plan 的前 3/4 个截断槽位内就直接决定了是否丢覆盖。

下表概括了基线 9dbe0f2a5607 与 exact head af34cfdefdaa16afa7caaef210b22e7cedb11a24 下三条路径的实际差异(以实测代码路径为依据):

清理切片 基线机制 (9dbe0f2a5607) 当前 HEAD (af34cfdefdaa) 实测覆盖结果
todo 冷路径 todo-detail-cold-path-contract-smoke.py + TS/pytest 删除过渡 smoke 与 todo-lifecycle 对应 deep 条目 PASS:TS 11+66、test_chat_todo_detail.py 与 4/4 控制面 smoke 覆盖完备
CLI 模块化 assert_source_shape 检查源码文本,缺 LOOPX_USAGE_PING=0 删除源码文本断言,为 run_cli 注入 LOOPX_USAGE_PING=0 PASS:3 个 CLI smoke 与 test_cli_entrypoint.py 等通过
issue-fix 去嵌套 reviewer-request(slot 4)内嵌子进程跑 drain-postwrite 删子进程并注册到 checks[4](第 5 位) FAIL:premerge(4) 与默认 plan(3) 全部截断跳过

具体改动 / Changes

全量 diff 涉及 12 个文件(+21/-197):6 个测试/smoke 文件(+8/-191,含整文件删除 examples/control_plane/todo-detail-cold-path-contract-smoke.py)、2 个规划与目录模块(+9/-6)以及 4 份中英文文档/协议(+4/-0)。以下对照规范 docs/development/good-smokes.md(基准修订 9dbe0f2a56077ed133d5bb3796bc2fa6df7d75a2)检验各项准则:

  • retire-superseded-todo-smoke -> implemented:全仓 rg 确认无剩余引用;tests/control_plane_ts/planning_horizon.test.ts(11 passed)、turn_envelope.test.ts(66 passed)、tests/test_chat_todo_detail.py 与 examples/control_plane/hot-path-interface-budget-smoke.py 完整承接冷路径契约。
  • remove-brittle-cli-source-shape-checks -> implemented:3 个 CLI 模块化 smoke 删除 assert_source_shape 后仍执行真实 CLI 子进程并对 run_cli 设置 LOOPX_USAGE_PING=0,且 tests/test_cli_entrypoint.py、tests/test_worker_bridge.py 等持续提供模块级回归保护;中英文 loopx/capabilities/issue_fix/README.md、README.zh-CN.md、catalog_entry.py 及两份协议文档同步补登 +1 引用无语义偏移。
  • remove-nested-smoke-without-coverage-loss -> not_met:取消嵌套后未确保新条目进入默认/合前切片上限,也没有对应单元测试替代。

关键代码讲解

  • loopx/canary/planner.py:748 CURRENT_REPO_PROFILES(issue-fix-reviewer-routing 与 558 行 todo-lifecycle):定义路径触发到 smoke 列表的静态顺序。todo-lifecycle 删去已退役条目正常;但 issue-fix-reviewer-routing 将 examples/issue-fix-reviewer-notification-drain-postwrite-smoke.py 追加到了第 800 行的索引 4(第 5 位),排在 4 个更早的检查之后。
  • loopx/canary/planner.py:1646 build_catalog_canary_plan:第 1655 行默认参数 max_checks_per_profile: int = 3,严格截取前 3 项,使 tests/test_smoke_suite.py 与默认 loopx canary plan 永远触达不到索引 4。
  • loopx/canary/premerge.py:909 build_premerge_validation_gate:第 986 行固定以 max_checks_per_profile=4 调用规划器,仅选取前 4 项(索引 0-3,截至 issue-fix-reviewer-request-smoke.py),再次把索引 4 截断在门禁之外。
  • examples/issue-fix-reviewer-request-smoke.py:357 main:移除了末尾通过 subprocess.run 触发 drain-postwrite smoke 的 11 行代码,切断了此前第 4 槽位顺带传递执行的路径;仅 examples/canary/catalog-planner-smoke.py:573 因显式传入 max_checks_per_profile=7 才看到 6 个检查。
  • examples/cli-control-plane-command-modularization-smoke.py:17 run_cli:通过注入 LOOPX_USAGE_PING=0 隔离遥测开关并在三个 CLI smoke 中只保留行为断言;另附非阻塞提示:该文件第 160 行 assert_history_reader_route 的 -c 子进程仍未显式附带 LOOPX_USAGE_PING=0。

对主干的风险 / Risks

P1 阻断项:drain-postwrite smoke 从 premerge 与默认 canary plan 静默失焦且缺少单测保护

当后续改动触及 loopx/capabilities/issue_fix/reviewer_notification_drain.py、examples/issue-fix-reviewer-notification-drain-postwrite-smoke.py 或 reviewer_request.py 时,build_premerge_validation_gate(上限 4)与 build_catalog_canary_plan(默认上限 3)仅执行 issue-fix-reviewer-routing['checks'] 的前 3-4 项,直接舍弃位于索引 4(第 800 行)的 drain-postwrite smoke;与此同时,全仓 tests/ 中没有任何单元测试调用 drain_issue_fix_reviewer_notification_queue。实测反事实检查(以这 3 个路径分别调用 build_premerge_validation_gate 与 build_catalog_canary_plan 共 6 次)结果全部未选中 drain-postwrite smoke;pytest -v tests/test_smoke_suite.py --loopx-smoke-suite catalog-plan --loopx-smoke-profile issue-fix-reviewer-routing 实测也仅运行了前 3 项检查(4 passed in 2.35s)。由于本 PR 已合入 main(0f7b545ca5d291e185396102ade3f05e90fab670),这意味着后续通知队列重启写回、sent_unverified 抑制和语义去重的退化风险已失去日常默认门禁保护。

最小修复与重跑验证:建议在 main 上提一个后续修复 PR,在 tests/capabilities/ 中为 drain_issue_fix_reviewer_notification_queue 补齐核心单元测试,或将 examples/issue-fix-reviewer-notification-drain-postwrite-smoke.py 调整至 issue-fix-reviewer-routing['checks'] 的前 3 位(并增加校验 max=3/4 下必选中的断言),随后运行 loopx canary premerge --tier standard --from-git-diff 与 pytest -v tests/test_smoke_suite.py --loopx-smoke-suite catalog-plan --loopx-smoke-profile issue-fix-reviewer-routing。

其他核查项边界清晰:本 PR 未引入新配置开关(default_off_isolation 不适用),协议文档补充不扩张 actor 生命周期或权限语义(authority_semantics 不适用),也未新增共享状态词表(semantic_alignment 不适用)。较宽范围诊断跑 pytest(794 passed, 8 skipped, 3 failed,其中 tests/canary/ 为 137 passed, 6 skipped, 1 failed)出现 3 条失败,均集中在本次未改动的文件(test_maintainability_ratchet.py 报告的 interaction_contract.py 与 test_t06_stopped_goal_refuses_at_effect_time[file,sqlite] 的 Goal 元数据升级提示),因未单独在基线 commit 重跑同 nodeid 而标为非必需诊断项,不构成对本 PR 的额外要求。

我的整体评价 / Verdict

基于 exact head af34cfdefdaa16afa7caaef210b22e7cedb11a24(已于 2026-11-19T12:58:35Z 合入 main)的本地独立验证已全部完成,不咨询或等待远程 CI。长程影响 long_horizon 为 regression,因为取消嵌套子进程后将关键恢复写回检查放在超出 premerge 与默认 canary plan 配额的第 5 位,使自动化门禁在主干上出现持续盲区;交互体验 user_experience 为 not_applicable(本次改动仅涉及 Python smoke 脚本、canary 剖面配置与 issue_fix 文档目录,不涉及前端页面或交互渲染面。,未核验范围 none)。可观测语义对比结论为 unintended_drift:todo 与 CLI 路径等价通过,但 issue-fix-reviewer-routing 默认选取语义发生非预期覆盖流失。代码量 code_volume 为 necessary(净删 176 行),整体改动比例 change_proportionality 为 proportionate(架构评估 retain),默认关闭隔离与权限语义均 not_applicable,且未引入或修改任何共享语义词表。

本地关键验证明细:

  • loopx canary smoke-health -> PASS(358 个 smoke,0 嵌套执行,0 重复组)
  • 定向 smoke(9 个改动及关联 smoke + 4 个控制面 smoke)-> PASS(9/9 与 4/4 全通过)
  • 替代测试与重点单元测试(TS 11+66、test_quality_surface_catalog、test_smoke_fleet_health、test_telemetry_isolation、test_worker_bridge、test_chat_todo_detail、test_chat_internal_module_layout)-> PASS(77 TS + 44 pytest 全通过)
  • 仓内必检门禁(premerge quick/standard、ruff check 7 个改动 .py、mypy 2 个改动模块源码、loopx check 11 个未删文件、RFC 双语与 diff 检查)-> PASS(其中 premerge standard 对控制面既有的 maintainability ratchet 失败自动降级为 advisory,新增词表检查通过)
  • 反事实默认门禁选测(drain-postwrite 在 premerge max=4 / plan max=3 及 tests/ 单测检索)-> FAIL(0/6 选中且 tests/ 中 0 单测)

剩余风险为上述未单跑基线复现的 3 条无关诊断测试以及 cli-control-plane 第 160 行遗留环境变量缺口。由于 PR #9073 已合入,建议通过后续小 PR 补齐 drain-postwrite 的默认门禁顺位或 pytest 单元测试后再复核关闭该覆盖缺口。

English verdict: REQUEST_CHANGES (POST_MERGE_AUDIT_COMMENT) — Retiring the todo cold-path smoke and brittle CLI source-shape checks is clean and covered, but removing nested subprocess execution leaves examples/issue-fix-reviewer-notification-drain-postwrite-smoke.py at checks[4] beyond the premerge (4) and default canary plan (3) limits with zero pytest unit coverage for drain_issue_fix_reviewer_notification_queue on merged head af34cfd.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants