Repository navigation
Fix stale canary checks and repository paths - #8737
Conversation
Signed-off-by: aihop <4408344+hhyykk@users.noreply.github.com>
Signed-off-by: aihop <4408344+hhyykk@users.noreply.github.com>
Signed-off-by: aihop <4408344+hhyykk@users.noreply.github.com>
Signed-off-by: aihop <4408344+hhyykk@users.noreply.github.com>
Signed-off-by: aihop <4408344+hhyykk@users.noreply.github.com>
…01a1d82f Signed-off-by: aihop <4408344+hhyykk@users.noreply.github.com>
Signed-off-by: aihop <4408344+hhyykk@users.noreply.github.com>
…ures Signed-off-by: aihop <4408344+hhyykk@users.noreply.github.com>
Codex Review Summary: ✅ CompletedThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
loopx-agent
left a comment
There was a problem hiding this comment.
Reviewer: model_agent; model=gpt-6.1-sol; provider=OpenAI; declaration=runtime_reported; effort=max.
动机
修复 canary 选择到已删除脚本的问题,并清除 Core 与 Dashboard 中不再使用的声明。 loopx/canary/planner.py 是当前选择入口。修复失效的验证入口并减少维护噪声;类型、路由、所有权与错误语义保持在本次验证范围内。 主干与候选各816项核心测试通过、6项PostgreSQL跳过,822条测试名称和状态一致。36条相同输入的完整输出/异常观察逐字相同(20,548字节)。候选52条未用诊断全部消失。 本地macOS/Node24验证;6项真实PostgreSQL用例因未提供隔离URL跳过,未认证该provider。未验证Windows、已安装App或完整浏览器布局;清理未改变布局。
改动思路
两个检查入口指向现存的模块化 bridge smoke,补回原安装契约断言,遍历 catalog 的脚本路径;清除未用符号但保留参数位置、初始化及摘要校验求值。 保留原 canary planner、bridge、类型 owner,不增加新的 runner;不能把未用摘要校验误删为无求值。 当前交付边界:Catalog and same-owner declaration cleanup。
具体改动
GitHub历史比较603条路径;相对已携带主干 c526242,提案独有40条。与当前主干 42483ac 的本地 merge-tree 无冲突。
相同目录探针在主干发现已删除脚本并退出1,在候选149项Python脚本检查(127种路径)中无缺失并退出0。 9项 catalog pytest和planner smoke通过;bridge smoke在本机默认双runtime路径冲突后,以仅增加明确隔离runtime参数的同断言夹具通过。
主干与候选各816项核心测试通过、6项PostgreSQL跳过,822条测试名称和状态一致。36条相同输入的完整输出/异常观察逐字相同(20,548字节)。候选52条未用诊断全部消失。 Core与Dashboard严格类型检查、Chat route smoke通过。当前目标的premerge共6项直接检查和19项选择检查通过。
发布前主干推进到 d210e4c,仅新增两处Todo storage compatibility路径。两文件覆盖到候选后的定向集成实验836项通过、6项PostgreSQL跳过;实验后已恢复原精确头。新主干的 merge-tree 也无冲突。
私有未用参数只改名,不删位置;fixture初始化和请求摘要求值保留。尤其 readOperationReceipt 仍调用 operationRequestDigest(request)。 两处 profile 改用现存 bridge smoke,其安装只读挂载、无上传、凭据不入记录等断言已并回。历史文档链接均有对应 Git blob。
对主干的风险
清理横跨共享类型与权威状态路径,主要风险是误删求值或参数位置。本次逐项核对并做相同输入的完整输出、拒绝诊断和共享测试对照。 PostgreSQL两处仅删未用import;六项集成跳过保留为未验证,不据此宣称provider资格。本地macOS/Node24验证;6项真实PostgreSQL用例因未提供隔离URL跳过,未认证该provider。未验证Windows、已安装App或完整浏览器布局;清理未改变布局。
我的整体评价
批准,未发现阻塞项。修复失效的验证入口并减少维护噪声;类型、路由、所有权与错误语义保持在本次验证范围内。 依据 docs/development/testing-and-quality.md 在 42483ac 的 REAL_PATH、BEHAVIOR_PARITY。没有新增公共状态、激活开关或决策owner。
English verdict: APPROVE. The maintained canary targets restore executable validation; the bounded cleanup preserves evaluated calls, parameter positions and checked behavior.
The merge-base changed after approval.
Signed-off-by: aihop <4408344+hhyykk@users.noreply.github.com>
loopx-agent
left a comment
There was a problem hiding this comment.
批准:未发现阻塞项。修复失效的 Canary 入口具有直接维护价值,清理未增加运行时机制。现存预算失败与合并验证保留为单独限制。
Reviewer: model_agent; model=gpt-6.1-sol; provider=OpenAI; declaration=runtime_reported; effort=max.
动机
运行 Canary 的贡献者和维护者。
选择 runtime connector 或安装验证配置时,旧清单会执行已删除脚本而中止;候选改为维护中的 bridge smoke,使同一检查能够运行。
149 项 Python 脚本引用全部可解析,完整安装边界断言与 active-user 观察通过;未用声明清理保持已验证的返回值、拒绝诊断和状态回读。
本次不改变运行权限、默认激活方式、持久化格式或 UI 布局,也不认证生产部署和 PostgreSQL provider。
审查精确头:274c4b65b4b70dcccb0abcd2206429e263abd763。基线采用真实分支 merge-base d210e4cc1d002f92498b2811c565b9430e79f4c1。本次判断针对检查入口与未用声明的维护任务,不把通过的检查数量当成整个产品已完成。
改动思路
保留现有 Canary catalog、worker-bridge CLI 和 typed owners;用 61 行断言补足现存 smoke,没有新决策 owner 或运行时分支。
当前 PR 的交付边界是修复失效检查与文档指向,并删除同一代码库的未用声明。
最小方案确实可以只改两处脚本名;我也重新评估了捆绑 40 条路径的成本。这里的生产差异仅 +23/-50,主要是 import、未用绑定和保留位置的私有参数改名。删除无调用 helper 比增加兼容层更简单。没有新 schema、开关、写入入口或人工同步状态,现有断言、初始化与摘要求值仍在。
两个配置仍保留原来的 default/deep 分级。CURRENT_REPO_PROFILES 继续由原 planner 选择,转向现存 smoke;不是新增第二套 runner。
具体改动
完整差异为 40 文件、+119/-69:26 个生产文件 +23/-50,12 个测试或 smoke 文件 +85/-13,2 个文档 +11/-6。上一正式审查绑定的是 57f1284854190dbd824abc9b7b7035b34cbfaa01;新头吸收了六处主干文件变化,本轮重新判断完整 PR,没有沿用旧批准。
- bridge smoke main 补回只读挂载、Python preflight、无上传、凭据不入记录和 active-user 挂载的断言。基线 catalog 探针确实报出两条已删除脚本,候选在同一 149 项检查、127 种脚本路径中没有缺失。
- readOperationReceipt 删除未用变量名,仍执行
operationRequestDigest(request);后续锁内摘要比较也保留。私有参数位置没有被删,测试夹具初始化仍执行。 - validateScope 只去掉未用解构绑定;全局 gate、单 agent 绑定及冲突诊断继续由原 typed owner 判断。36 组相同输入的完整返回值与异常字段逐字一致,共 31,858 字节。
两边相同 35 组 TypeScript 测试各 1243 通过、6 跳过、0 失败;包含嵌套用例的全部 1249 条名称和状态一致。 File/SQLite 路径覆盖真实持久化、lease 重放、claim、Todo 回读与 SQLite 历史迁移。6 个跳过项需要隔离 PostgreSQL 服务,未记为通过。Core 额外 unused 诊断从 43 降到 0,Dashboard 从 9 降到 0;普通与额外类型检查、9 项 catalog pytest、planner smoke、源码环境的 Chat build、route、delivery review、team report、4 项 Markdown 测试及变更 Python 文件的 Ruff 均通过。
依据 docs/development/testing-and-quality.md,以 d210e4cc1d002f92498b2811c565b9430e79f4c1 固定所审查文本。验收项 REAL_PATH:两棵物理源码树使用各自已安装的 Python 环境,验证实际 File/SQLite 与 CLI;BEHAVIOR_PARITY:完整输出、全部测试名称和状态及失败分支对照;REGRESSION_SENSITIVITY:同一个 catalog 路径探针在旧缺陷退出 1,在候选退出 0。三个文档历史链接的对应 Git blob 也已核实。
对主干的风险
最强风险是清理把原来会求值的代码误删,或让参数移位、fixture 初始化消失。上述实际入口、完整输出对照与原有断言覆盖这些风险。生产查询仍走相同 owner,未增加字符串分类规则、权限或持久化版本。
保留以下原始失败,不把后来的成功改写成从未失败:第一次集成运行遇到 ENOSPC 并退出 7;一轮启动器选到已安装 Python 的运行主动终止。正式对照固定到各自源码 Python,使用本轮隔离临时目录,并保留所有原断言与存储余量限制。独立 Bridge 回读还检查了实际 observation JSON 与 counter-trace:启动前消息未被算入,seq=2 的消息被记录。
原始 bridge smoke 在两边都因双默认 runtime 含状态而失败,CLI bootstrap 和该故障分支不在差异内。相同完整断言仅增加显式 fixture runtime 后,两边均通过,且完成独立文件回读。预算检查在同一长临时路径环境的两边均为 dashboard_status_json 25,294/25,000;预算与相关路径没有被本 PR 修改,没有放宽阈值。
源码环境标准 premerge 执行完成:六项直接检查通过,19 项选择检查中 18 项通过、上述预算检查失败,总门禁仍是 failed。它保留为合并验证限制,不能宣称全绿。没有查询、轮询或等待远端 CI。主干 ef099ae3ac3c08054b9d928fff48aefe2615b950 与候选的完整 merge-tree 无文本冲突;四个重叠文件均已检查,但没有把这一静态结果说成合流树运行通过。
我的整体评价
保留现有 Canary catalog、worker-bridge CLI 和 typed owners;用 61 行断言补足现存 smoke,没有新决策 owner 或运行时分支。 当前 PR 的交付边界是修复失效检查与文档指向,并删除同一代码库的未用声明。
这个维护切片已恢复可执行检查,并清除当前 52 条未用诊断,未发现 PR 引入的阻塞缺陷。预算失败在基线与候选相同,不要求此 PR 改写不相关预算来换取批准;合并仍须保留仓库验证与 maintainer 授权门禁。本轮没有合并或撤销其他人的审查。
所审查的 Dashboard 改动只移除未用 import、局部变量与私有无调用 helper;没有更改实际渲染分支、层级、样式或导航。
未验证 Windows、已安装 App 的页面及 PostgreSQL 端到端;本次不以它们作为新增资格声明。
English verdict: APPROVE. The maintained catalog targets and bounded cleanup preserve the independently tested behavior. The pre-existing local budget failure remains a separate validation hold.
Current delivery (2026-11-17)
Two Canary profiles still invoked a deleted worker-bridge script. They now use the maintained bridge smoke, which also covers read-only mounts, preflight, privacy and active-user observation. The regression checks Python script targets across all catalog profiles. Broken protocol paths are corrected, while retired examples link to verified historical revisions.
This branch includes main
d210e4cc1d002f92498b2811c565b9430e79f4c1, including upstream #8892 queue rendering/Lark boundary and #8902 historical Todo field compatibility. It clears the current extra TypeScript unused diagnostics: 43 to 0 in the control plane, 9 to 0 in Dashboard. Positional signatures, original expression evaluation, fixture initialization and all assertions are retained. The only deleted helper is private and has no callers. No abstractions or lint exemptions were added.Fresh validation
--noUnusedLocals --noUnusedParameters: passed with zero diagnostics.endpoint.tokenwarning in the unchanged ZCode request body.After latest-main reconciliation, standard premerge passed 6 direct and 19 selected checks, and both extra unused checks remained clean. Pushed-head CI is checked separately. No historical evidence, persistent user state or compatibility API was deleted.
Historical qualification notes (preserved)
Goal And Delivered Outcome
Two Canary profiles advertised a deleted worker-bridge script. They now select the maintained bridge smoke, which checks read-only mounts, preflight, privacy and active-user observation. A regression checks direct
pythonandpython3script paths across the Canary catalog profiles.Correct the broken event-ledger protocol paths. Link retired examples to their exact historical source and retain the original evidence. Remove 25 unused TypeScript imports without changing call signatures or assertions.
Related to the review in #8703; intended base is
main.Author Declaration
Implemented against
mainat185e0d90df7ef0d6a38a743958fe0ea9d695267f; the newermain3cf2f45552dc2fdad0a160e3e736c1b7c9be21eehas no overlapping files and its projected merge contains exactly the same 21 changed files.loopx/canary/planner.pytests/canary/test_quality_surface_catalog.py, both changed smokesdocs/concepts/interaction-pattern-catalog.md, course 04Scope And Continuation
Validation
Windows reader compatibility follow-up: trying the PowerShell parser first and falling back to POSIX is unsafe. A legacy POSIX-rendered value containing two consecutive U+201A quotes was accepted by that parser but decoded to one quote. A revised local candidate distinguished the tested producer forms using the quoted versus unquoted quota and context-read tokens, retained the existing pagination/identity assertions, and used the existing command_argv helper for a .ps1 launcher. Across the two relevant base revisions it passed 5,000 generated argument vectors per revision in both producer formats and eight six-page synthetic subprocess runs with mixed page formats; changed capture identity and five malformed expressions were rejected. The .ps1 route used a synthetic local pwsh shim, not PowerShell or Windows. Microsoft's pwsh documentation describes the File form. The exact failed CI argv was not logged, the candidate is not committed, and native Windows validation remains required.
Main 859a23e preflight: GitHub pull metadata reported base 3cf2f45; comparison with main found merge base 185e0d9. Complete inventories contain 21 branch paths and 272 main paths, with one overlap. Its in-memory git merge-file result had no text conflict and matched the local-authority source hash used by the earlier projected checks. All six PR commits have Signed-off-by trailers. This is a file-level comparison, not a full-tree merge or CI result.
Subsequent behavior follow-up at main 859a23e and this PR head: Node 26.12.1 ran the existing tests with project ESM source supplied by an in-memory projection. Both the projected PR and local candidate passed all 232 targeted tests, plus four governance cases selected by name. Test names and statuses matched between versions. A module-load audit verified the expected source hashes for all six changed files in each version; the loaded module path sets also matched. The runs used synthetic data in a network-isolated sandbox that allowed writes only in a temporary directory. This is not a real checkout, a complete project suite, or native CI; the candidate remains uncommitted.
Main 859a23e candidate follow-up: the existing six-file local-only patch applies to the projected PR tree in memory with exact context. With TypeScript 6.0.3, the ordinary typecheck reports zero diagnostics both before and after the patch; the additional no-unused checks report 16 before and 7 after, with nine diagnostic occurrences removed and none added. The comparison counts duplicate messages separately. The first local counting script deduplicated those messages and failed its own count assertion; the corrected multiset comparison passed. These are in-memory compiler results, not a committed change, fresh-checkout tsc, current-main behavior tests, or CI. The 236 targeted behavior tests below ran against the earlier baaf657 projection.
Current-main compatibility note: PR #8703 TypeScript core shard 2 exposed a duplicate SHA-256 matcher already present in its main parent. The old main base used by this PR did not contain the new inbox module. A virtual-source scan of main 97fb254 reproduces the guard failure; this PR has not been tested as a full merge against that main.
PR #8703 dashboard shard 3 also failed at 184 of 200 items. The same scenario passed on PR #8733 with 200 items and a 771 ms uninstrumented expansion. The scenario, key fixture and work-board sources and lockfiles match, and the jobs used Node 22.22.3 and Chrome Headless Shell 151.0.7922.34. The product trees and runners differ, so this is not a controlled attribution. Separately, a local static synthetic DOM using the unchanged digest produced five groups of 40 and visible counts 15, 52, 89, 126, 163 and 200 as disclosures opened. That limited Chrome 145 check did not run the application or reproduce the CI result; inspect actual group, disclosure and timing state before changing the test or UI.
Windows parser compatibility investigation: before commit 2ecfc26, the capture reader emitted POSIX shlex-joined commands; the later Windows renderer emits PowerShell literals. A local PowerShell-only parser prototype rejected a historical POSIX command containing an apostrophe, while POSIX shlex changed an apostrophe in a modern PowerShell path even when the command had no ampersand prefix. A separate in-memory prototype matched 1,500 generated argument vectors and exercised six pages through a real synthetic child process; it rejected a changed capture identity. These are local diagnostics, not a committed fix, native Windows validation or proof of the CI runner argv. A repair must preserve both historical and current command forms.
CI queue snapshot (2026-11-14 around 09:36 UTC): each of the six Python test shards was pending behind 23 entries in its repository-wide concurrency group, including one running shard from this main run. The GitHub concurrency API was queried for each job, including shard 1. Queue positions can change and are not an estimated completion time. This snapshot does not establish the result of other checks.
The checks aggregate failed because dashboard-acceptance was skipped after dashboard-browser shard 3 failed. It is a downstream gate result, not a third independent failure. In the failing scenario, the expansion loop samples the visible count until a five-second deadline and then samples it again for the assertion. The reported 184 alone does not establish whether the deadline was reached or visibility changed between samples; the prior local base/merge comparison did not reproduce 184. Investigate with timing and disclosure-state evidence while retaining the existing 200-item and two-second acceptance limits.
A separate Windows job on PR #8703 also failed in the base CLI differential with WinError 2. Its actual test merge and this PR use identical fixture helper bytes, but their product sources differ. A repair needs validation against both revisions; neither log records the missing argv.
Updated TypeScript audit: with TypeScript 6.0.3 and the actual source files from main
bd2718a485c86fe4f0ad4d1356fe9aeeccfed1baplus the clean projected PR changes, an in-memory compiler host reports 41 no-unused diagnostics on main and 16 on the PR projection. It reports no ordinary typecheck diagnostics in either tree. The host was calibrated against the earlier realtscresults at mainbaaf657: 40 on main and 15 on the PR projection, with no ordinary diagnostics. The one new main diagnostic is the unusedresearchIdentifierimport inloopx/control_plane/capabilities/explore_research_execution.ts; the PR adds no no-unused diagnostic relative to this main. This is an in-memory source projection, not a real-checkouttscrun. The local-only nine-diagnostic candidate below is not included in the 16.Windows differential follow-up: the probe copies the candidate checkout’s
tests/control_plane/cli_output_fixture.pyinto the shared fixture directory for both the base and candidate arms, while each arm loads its own product source. A fix to that helper can therefore reach the failed base measurement without replacing the baseline or skipping the comparison. In a separate parse-only check, the actual renderer from base3cf2f455was round-tripped through PowerShell 7.7 on Linux for an executable and capture path containing spaces, ASCII apostrophes and all four escaped smart single quotes; an appended unquoted--selectalso survived. The existing POSIXshlex.splitput&in executable position for that case. A bare executable case matched both parsers. These are synthetic parser checks, not a native Windows run or proof of the exact CI argv.Latest-main follow-up (2026-11-14): main at
f17ce83544f8b0420b6cb13dfdd086ca23cf6646changed 80 paths sincebaaf657e5df6b779be1401d143af6aa5970454ef; none overlaps the 21 files currently in this PR. A whole-tree merge and typecheck against the newer main have not run. The current CI test merge has parent3cf2f45552dc2fdad0a160e3e736c1b7c9be21ee; its failures are described below.Dashboard browser shard 3 failed on
goal-workspace-m2: 184 of 200 tasks were visible in its uninstrumented qualification pass. The previous packaged shard at this same PR head passed.Follow-up local Linux comparison used the exact CI base
3cf2f455and synthetic merge treed1072ce8, Node 22.22.3 and Chromium 151, with the existing development and coverage scenario and unchanged limits. Two alternating runs per tree each yielded one failure at a different 1-second lens-switch budget and one complete pass; every run that reached the count saw all 200 tasks. Both complete runs reported slower coverage-only diagnostics, which the existing test does not use as its qualification gate. The CI 184-item symptom was not reproduced; local results do not prove its cause or establish performance parity on the CI host.Follow-up at main
baaf657e5df6b779be1401d143af6aa5970454ef: the extra TypeScript checks report 40 diagnostics on main and 15 on the clean projected merge, with 25 removed and none added. Both used TypeScript 6.0.3.A separate local-only candidate removes nine more diagnostics without removing assertions or awaited calls. The normal typecheck and 236 targeted tests passed on both the unchanged projected merge and the local candidate. The candidate is not committed or included in this PR.
Windows CI failed while measuring the base, before the candidate:
read_pinned_cli_bodyraisedFileNotFoundError(WinError 2). The relevant producer and reader are byte-identical between that base and this PR. A local reproduction using those source versions and Windows command rendering makesshlex.splitplace the PowerShell&operator in executable position. The CI log did not record argv, so this mechanism is supported but not confirmed as that runner's exact missing executable. Removing only&is insufficient: the same local reproduction shows POSIX splitting also changes a PowerShell-quoted path containing an apostrophe.Tested revision:
4aee125ac805161b3b6136a6f1a39ca9470bd121for targeted checks after merging current main. The full TypeScript run below was on07413af94058695c39116a6d096c8b30f5a6334f.Run state: finished (the full local Python suite was interrupted).
Input classes: synthetic, public_fixture.
regression_paritypassedreal_entrypointpassedpython3 examples/canary/catalog-planner-smoke.py,python3 examples/cli-worker-bridge-command-modularization-smoke.py, andpython3 examples/worker-bridge-active-user-feed-smoke.py.unitpassedunitpassedstaticpassednpm run typecheck:control-plane, project-scoped Ruff, mypy and CLI output-budget smoke passed on the merged head.staticfailed--noUnusedLocals --noUnusedParameters: 40 diagnostics before, 15 after; no new diagnostic. This extra check is not enabled by the normal project typecheck.unitfailed07413af94058695c39116a6d096c8b30f5a6334f: 6,436 passed, 85 skipped, 2 failed. Host heartbeat failed to start within the test's two-second window; the exact test later passed three times on the candidate and once on an isolated main source copy. The SQLite rehearsal hit its 5 GiB free-disk reserve; the same test passed after task-owned temporary files were cleaned, with the threshold unchanged.unitfailed185e0d90df7ef0d6a38a743958fe0ea9d695267f. The previous-base run was interrupted when main advanced (734 passed, 232 skipped).integrationpassedloopx canary premerge --from-git-diff --git-diff-base origin/main: 10 catalog, 8 risk-profile and 1 public-boundary checks passed on the merged head. The first run on the prior base timed out the vocabulary smoke while other suites were running; the timeout was unchanged.staticfailedFrontend / Visual Evidence
Type of Change
LoopX Area
Technical Direction
Shared-authority RFC fixture impact
Boundary Checklist
none.Signed-off-bytrailer.Dashboard-browser local reproduction (2026-11-14)
The PR #8703 dashboard shard 3 job reported 184 visible work items instead of 200. In one local run of the original
goal-workspace-m2scenario, the 200-item expansion assertion passed in 953 ms. That run then failed a different existing performance gate: one median of three loaded-lens transitions was 1,028 ms against the 1,000 ms limit; the maximum individual sample was 1,081 ms. The uninstrumented qualification failed, so the original test did not run its coverage pass. No acceptance threshold was changed.The run used Chrome Headless Shell 151.0.7922.34, matching CI, on OpenCloudOS with Node 26.12.1. It used the original scenario and fixture, a local Vite development server, and an in-memory overlay of the ten PR dashboard source files; those sources match the failed CI merge
2fe7226cb2a83d0ab3dae224618670437cb7129c. The remaining files came from an older local checkout, and the host differs from CI. This single run does not establish the cause of the 184-item CI result or its relationship to the PR.Real PowerShell reader validation (2026-11-14)
The failed Windows runner image lists PowerShell 7.6.9. A local Linux test used that PowerShell version from the official release, verified against the release SHA-256 manifest.
For each of the two affected base revisions, 1,002 PowerShell-rendered argument vectors matched the real PowerShell parser and the candidate reader; 1,002 legacy POSIX-rendered vectors also round-tripped. The cases include apostrophes, smart quotes, Unicode, spaces and shell metacharacters. Four synthetic
.ps1paging runs completed six pages each, and identity tampering and five unsafe commands were rejected.A separate test used the actual LoopX
loopx.ps1andloopx_entry.pyscripts from an older checkout; both scripts match base87a736bd. The original reader reproduced a missing executable for the PowerShell call operator. The candidate used real PowerShell to read a 1,193-byte synthetic capture through five real LoopX pages, reproducing the original bytes without creating a registry. When the capture was changed after page one, the next read failed without returning partial content or a continuation command. The test restored its temporary capture.These runs were isolated to loopback-only networking and temporary writes. They do not prove which executable was missing in CI: its argv was not logged. The candidate is not committed, the checkout is not the full CI merge, and native Windows CI remains required.
Current-main overlap audit (2026-11-14)
A read-only tree comparison used main 8d7d234 and each PR's actual base and merge base. It checked every path changed on both sides with an in-memory three-way merge. This is not a full checkout, build or CI run.
For #8737, 21 PR paths and 281 main paths changed from merge base 185e0d9; the single overlapping path merged cleanly in memory. GitHub reports actual PR base 3cf2f45, which differs from that merge base; a direct base-to-head tree comparison has 31 differing paths and must not be treated as 31 PR changes.
The uncommitted two-line portion of the Windows reader candidate merged cleanly with #8266's fixture changes but conflicted with #5677. This is only a textual check of those two lines, not a merge or test of a complete fix. #8758 already owns the separate digest repair.
Python lint audit (2026-11-14)
On main 4604f93, Ruff 0.15.22 checked all 3,911 tracked Python files individually via stdin. The 2,071 files in the previously used project scope had zero diagnostics. The additional all-tracked scan reported 132 diagnostics outside that scope. All source blobs were checked against their Git hashes, and known F401/F821 inputs verified that stdin and direct-file checks agree. This is a per-file source audit, not a full-checkout Ruff run.
The four Python files changed by #8737 reported zero Ruff diagnostics at this head. Their merge-base versions match main.
Two of the extra findings are duplicate imports in loopx/status.py. An in-memory candidate removes only the second occurrence of each binding, keeps both public aliases, and reduces Ruff F811 from two to zero. The original and candidate passed the same 18 existing pytest cases, and each alias remained identical to its owner. The checked status and test files match main; other dependencies came from an older checkout. The candidate is not committed, and full CI is still required.
The candidate also merged cleanly in memory with the status.py changes in #8647. This is a textual merge check, not an integrated test. Ruff also marked two pytest fixture imports in benchmark/tests/test_checkpoint_storage.py as unused; pytest fixture discovery confirmed that the Docker-gated test uses both, so they must not be removed automatically. Docker was not invoked.
Current TypeScript unused audit (2026-11-14)
On main 4604f93, a hash-verified in-memory TypeScript 6.0.3 compiler host checked 528 control-plane source files. Normal type checking reported zero diagnostics for main, the projected PR, and the existing nine-diagnostic candidate. Extra noUnused checks reported 41, 16, and 7 diagnostics respectively.
One additional local candidate removes the unused researchIdentifier named import from explore_research_execution.ts while retaining the other imports from the same module. The three existing Explore research test files passed 24/24 both before and after, with the same test names and 14 hash-audited loaded modules. Applying that removal on top of the earlier nine-diagnostic candidate left zero normal type errors and six unused diagnostics. Neither candidate is committed.
The six remaining diagnostics concern name, root, identity, request, expectedRequestDigest, and goalId. They have not been removed automatically. In particular, the lifecycle receipt digest is checked later under the mutation lock; any cleanup there must preserve the existing validation and failure order.
These checks used source overlays and the installed node_modules, not a clean checkout or CI. The earlier 236-test qualification for the nine-diagnostic candidate used main 859a23e; the 24-test Explore check used the current main snapshot.
Continuation-note parameter audit (2026-11-14)
On main 4604f93, the unused name parameter in the private boundedString function has 15 call sites. A local candidate removes that parameter and the corresponding string-literal arguments. The function is not exported, and the AST check found no other references.
The original and candidate each passed all 37 existing continuation tests. A 156-case comparison of valid, invalid and boundary inputs produced identical full results and error messages. The target and test files match main; other dependencies came from an older checkout. An initial isolated run had 11 failures in both variants and in an unmodified control. Those failures disappeared after enabling loopback inside the isolated network namespace, which the Python-to-TypeScript runtime requires.
Combining this candidate with the previously described nine-declaration and one-import candidates produced zero normal TypeScript diagnostics and five extra noUnused diagnostics across 528 files. This used an in-memory compiler host; it is not a clean-checkout typecheck or CI. None of the candidates is committed.
Private parameter audit (2026-11-14)
On main 4604f93, the private functions reductionWithTurnOutcome and successEnvelope each receive one unused parameter at their single call site. An AST check verified both arguments are simple identifiers and are unused in the corresponding function bodies. Identity and request validation elsewhere remain in place.
Using the main sources and this PR's removal of the unused task-lease import, the original and local candidate each passed all 68 existing settlement, identity, task-lease and CLI tests, with no skipped tests. The source loader checked the hashes of 109 loaded modules. Runs used isolated temporary storage and installed dependencies. This is not a complete PR checkout or CI run.
Combined with the earlier local candidates, TypeScript 6.0.3 reported zero ordinary diagnostics and three additional noUnused diagnostics across 528 source files. The three concern root, expectedRequestDigest, and goalId; they remain unchanged. The earlier local candidates and these two parameter edits are uncommitted.