[tianxiaochannel-oss88]

Summary

• Add read-only CI timing telemetry in an independent best-effort job after the existing quality gate, with a 14-day ci-telemetry artifact and GitHub Step Summary.
• Capture the current workflow attempt's job/step observations and available shard reports, including failed reports and per-file results.
• Keep measurement labels explicit: job start delay includes dependencies, shard file sums are not parallel wall-clock time, and elapsed completed-job time is not a DAG critical path.
• Preserve all existing test selections, platform lanes, exact-once coverage checks, and required quality-gate semantics. No release, version, lockfile, or product-runtime changes.

Validation

• Relevant local CI, scanner, runner, native-verifier, and relay-cleanup tests: 292 tests, 291 passed, 0 failed, 1 pre-existing environment-gated skip (Node 22.22.1, serial file execution).
• Local workflow contracts and real extracted Bash commands cover successful paginated collection, API failures, invalid JSON/page shapes, and unavailable render fallback.
• Existing lane-policy regression tests still reject failed, cancelled, skipped, or missing required results.
• Markdown links, public-repository safety, and git diff --check pass.
• YAML parsing passes for all 11 CI jobs. actionlint is not installed locally; hosted Actions validation remains necessary.
• Hosted run 41343476793, attempt 1, on head dfc6a458ed02d93c42e18ea4544f504f1a525e21 failed. The PR is not ready to merge; see the investigation below.
• The new telemetry job succeeded on that failed run. Its downloaded artifact contains eight shard reports, including the failed daemon file. This validates failure-report collection, not overall CI success.

Hosted CI Investigation (Merge Blocked)

• The originating failed check is Daemon tests (shard 0), job 124958879262. The required aggregate quality gate also failed. Other daemon shards, package/core lanes, builds, typechecks, platform smoke checks, and CodeQL passed on this head.
• packages/daemon/test/channel-node-invocation-binding.test.mjs:113 expected the first real ESM command to return succeeded, but it returned timed_out with the existing timeoutMs: 4000 request. The subtest took 4896.695 ms; that file reported four passing tests and one failure, then exited normally. This is not a reproduced HTTP-server leak or a file-runner deadline failure.
• The workflow diff only appends the telemetry job. The failing test, product sources, dependencies/lockfile, and shard/process runner inputs are unchanged from base c472c62fe80ea52a579593c289e6bca42c450205. Telemetry depends on the aggregate quality job and started after it ended, so telemetry execution did not overlap this failure.
• Local source-revision comparison: 20 trials each on the base and PR head, alternating order, Node 22.22.1, two real shard workers, with acp-approval-journal.test.mjs alongside the victim. The original four-second command limit was unchanged. Package sources and runner inputs were identical; the baseline reused the head's built package outputs and dependencies via symlinks, so these are not independently built revisions or hosted-runner reproductions.

Revision	Failed / Trials	Subtest ms (min / p50 / p95 / max)	First command close ms (min / p50 / p95 / max)
Base c472c62f	0 / 20	477.015 / 489.344 / 526.345 / 551.252	29.037 / 29.958 / 32.801 / 37.423
Head dfc6a458	0 / 20	477.733 / 484.266 / 499.322 / 500.341	28.801 / 29.399 / 30.740 / 31.226

• Historical hosted samples: this subtest passed at heads a7594c74aab4 (1581.069 ms), 7e752ce119b1 (2679.953 ms), 63bc33230b44 (2309.817 ms), and PR v0.2.0 perf: speed up CI with bounded test shards and cleanup #630 head dd1212c2bb21 (1656.897 ms). The first failing head in this limited sample is dfc6a458; that observation does not establish a causal commit. Historical logs do not contain individual command startup/exit timing.
• Root cause remains unconfirmed. The local trials did not reproduce the failure; runner load, login-shell startup, process-tracking overhead, and cross-test pollution are not established causes. No timeout increase, skip, retry, hosted rerun, or merge was used to obtain a green result.
• Next diagnostic evidence needed: command spawn/first-output/exit timing and safe stderr, plus process-tracker timing on the hosted failure path. A delay-injection probe alone would demonstrate the deadline mechanism, not explain the original delay. Keep the existing gate and four-second limit intact while collecting that evidence.

Observation Limits

• Telemetry failures are isolated from the mandatory quality gate; they do not certify test success.
• Failure counts apply only to available reports. Missing or partial reports do not establish zero faults for the full run.
• The workflow was still in progress at telemetry capture time, so that snapshot's run duration is not its final completed duration. File-level timeout counts do not include an internal command timeout that surfaces as a failed test file. Observe 5-10 complete hosted runs before changing shard counts or worker concurrency; this PR claims no fixed speedup or billing reduction.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-11-01T19:48:57.975818Z	dfc6a45	PR opened
🔒 Security Review	✅ Completed 2026-11-01T19:51:54.905804Z	dfc6a45	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.