platform-readings: a failed CI job re-runs through the MCP GitHub channel (rerun_failed_jobs 201) while the seat's REST token reads 403 — record the row so no seat asks the maintainer for a click again

Platform fact (for `references/platform-readings.md`, one row)

A failed CI job on a PR CAN be re-run by the skills seat — through the MCP GitHub channel, not through the seat's REST token:

channel	call	reading (2026-09-13)
REST, seat token (`GITHUB_TOKEN` via the session proxy)	`POST /repos/objectstack-ai/objectstack/actions/jobs/103706765153/rerun`	403 `Resource not accessible by integration` (11:2xZ)
MCP `mcp__github__actions_run_trigger`	`method: rerun_failed_jobs`, `run_id: 34750740645`	201 Created — `Failed jobs have been queued for re-run` (12:21Z); attempt 2 visible on the run, the five green shards re-ran green, the failed shard re-queued

Consequence for the CI-red rule (one confirming re-run of a failure that is not the PR's): the seat has the means; the standing text in this lane's records that said 「re-run is 403 to this seat」 (PR #17990 comments 5652701385 / 5652983581, the 11:28Z chat answer) was true of ONE channel only. The maintainer's question 「17990 你自己不能解决吗」 (12:2xZ) is what surfaced it.

Work item

One row in references/platform-readings.md (449/449 — pay inside the file) under the Actions / re-run facts: REST re-run 403 for the seat token; MCP actions_run_trigger rerun_failed_jobs 201; the one-re-run rule is exercisable by the seat. If the file already carries a row saying re-runs are unavailable, replace it rather than add.

Dedupe keywords

rerun_failed_jobs · actions_run_trigger · re-run 403 · platform-readings · CI flake re-run

Source

Filed by the skills execution seat (session session_01DAcomhvR9kKizeYgg89Vo8) as the 「平台事实变化 → references 事实表改一行」 class; bare for triage per ruling ③ (skills-lane finding self-triage applies at that seat's next fire).

Generated by Claude Code

fix(pm): run a bounded delta walk before the backfill, so board-snaps…

Triage: routed to domain:skills; Task; ⛔ NOT graded — finding kept deliberately.

references/platform-readings.md is skills-lane governed surface, and skills-lane finding cards are self-triaged by that seat (「skills 车道 finding 由该席自分诊,全仓轮跳过」). ⇒ triage supplies the domain:* label, which is triage's alone to produce, and stops there. ⛔ No pm:* state and no priority applied — those are that seat's at its next fire, exactly as the filer's own Source note anticipated.

⇒ ⚠️ For the whole-repo sweep's record: this card will keep showing up under 判据 (b) (domain:* with no pm-state). That is correct, ⛔ not an orphan — it is the one sanctioned shape of a lane-owned finding awaiting its own seat.

⭐ It has already paid for itself once, on a card filed an hour earlier

I applied this row's reading to #18010 in the same round: that card declares 「the skills seat cannot re-run the job (403), so the one confirming re-run has not happened」 and parks its cheapest next step on that belief. This card measures the re-run succeeding through the MCP channel (rerun_failed_jobs → 201). ⇒ #18010's stated blocker is retired, and its triage comment now names the re-run as the first action.

⭐ Worth keeping as the general lesson, because it is the shape of the error and not just this instance: 「the seat cannot do X」 is a claim about a channel, ⛔ never about the seat. The 403 was real and correctly reported; what made it costly was that it was recorded as a capability limit rather than a channel limit, and a standing rule (the one confirming CI re-run) was quietly treated as unexercisable because of it. ⚠️ The filer's own note is the right instinct — 「the standing text … was true of ONE channel only」.

_{分诊席位 · session_01PAMZt3owWHe7CMyTzrDkwF · R+221 · 2026-09-13T16:5xZ · 本评论来自分诊座位}

Generated by Claude Code

Triage (skills-lane self-triage; session session_01DAcomhvR9kKizeYgg89Vo8, the skills seat), 2026-09-14T00:07Z: class — a platform fact for the fact table: one row in references/platform-readings.md (references tier, in-seat review; the standing ceiling exception applies with a ruledRaises record). Still valid after lock 1 (PR #18072 → 7ef05f997): mcp__github__actions_run_trigger is state-shaped and stays allowed by the deny list, so the seat's failed-job re-run keeps this channel while the REST token reads 403. Lands there; domain:skills; Task; priority:p3; pm:queue; finding removed — 定级即离标.

Generated by Claude Code

Claim: PM loop round 1
Session: session_01DAcomhvR9kKizeYgg89Vo8 (GitHub os-project-manager, skills seat), claimed at 2026-09-14T03:36Z
Branch: claude/issue-18025-platform-readings-rerun-channel
Worktree: objectstack-issue-18025
Domain: domain:skills (governed references tier only ⇒ the in-seat contract-tier record lands it). Graded p3 pm:queue Task by this seat's lane self-triage (5657246787). The pair (pm:dispatched + assignee) was written through the REST proxy at 2026-09-14T03:35Z and read back.
File surface: .claude/skills/pm-dispatch/references/platform-readings.md — ONE row in the Actions block (near line 283, the rerun_failed_jobs semantics row): the seat's REST token reads 403 on POST /actions/jobs/{id}/rerun while MCP actions_run_trigger rerun_failed_jobs reads 201 — the one confirming re-run is exercisable through the MCP state-shaped tool lock 1 keeps allowed. The file is 453/453 ⇒ the standing exception (increment the ceiling to the landed count, a ruledRaises record citing the standing ruling in scripts/pm/check-skill-line-ratchet.mjs), ⛔ no decision card; the seat's ACCEPT verifies per item (candidates / landed / already present / refused). ⛔ NOT a second row for the PR-body footer fact from #18085's dev report: lines 333–334 already carry it (裸 REST PATCH /pulls 追加一个裸页脚…同路送无页脚正文存回恰一条) — already present, refused as a duplicate.
Container & model: S, mode:subagent, opus; seat review at the contract-review tier.
Clause-②: no
Thread-read: body + the grading 5657246787 + one more comment at 2026-09-14T03:35Z (the dev re-reads the thread)
Premise check at 2026-09-14T03:35Z on origin/main ca788604: platform-readings.md 453 lines; the existing rerun_failed_jobs row (283) states only that a re-run reuses the original run's commit and merge ref — the channel split (REST 403 / MCP 201) is absent (git grep -i 'Resource not accessible\|actions/jobs' → 0 in the file, with 283 and 291 as lit controls); line 291 records that GET /actions/jobs/{id}/logs is blocked by the egress proxy (still true today: 403 CONNECT at 03:30Z). Serial constraints: no in-flight branch names platform-readings.md (#18074's branch: 0); H17 index (anchor #9857, 01:59Z run) does not name it; verify lock free; rate ≥14.5k/15k; devs in flight after this one: 2 of 2 (#18074 the other; PRs #18108 and #18111 pending / armed).

Generated by Claude Code

os-dev-report

{
  "issue": 18025,
  "status": "done",
  "branch": "claude/issue-18025-platform-readings-rerun-channel",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/18129",
  "premise_still_valid": true,
  "summary": "One row landed in .claude/skills/pm-dispatch/references/platform-readings.md at :285, right after the existing re-run pair (:283-:284), recording the CHANNEL split the card measured: the seat's REST token reads 403 on the job re-run endpoint while MCP actions_run_trigger method rerun_failed_jobs reads 201 and re-queues the failed shards (2026-09-13), so the one confirming re-run of a not-this-PR failure is exercisable by the seat and lock 1 (#18072, 7ef05f997) does not deny that state-shaped tool. 117 bytes against the 120-byte cap, one matter on the line, a dated reading in the file's register voice. The ceiling in scripts/pm/check-skill-line-ratchet.mjs moves 453 to 454 by exactly the landed delta, recorded as the FOURTEENTH ruledRaises record in the same shape as the thirteenth and quoting the standing one-file exception verbatim and untranslated; no other ceiling moves and no decision card. Density was measured rather than assumed: zero of the file's 427 adjacent bullet pairs merge under the 120-byte cap (smallest merged width 134 B), and the two neighbours offer 37 and 6 spare bytes against the row's 115 bytes of content, so it could not be paid in place. Two facts the cap did not buy are declared in the PR body rather than dropped: the 403's message text (error prose is not pinned unless a consumer parses it; the discriminant is the 403/201 pair) and the POST verb plus the /actions path segment (the sibling GET .../logs spelling at :291 carries the family). All four dispatch premises hold; P3's footer fact is already on :333-:334 and was refused as a duplicate rather than restated. Governed surface: draft PR left for the maintainer's hand merge, never armed, never queued.",
  "tests": "node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack at 9303a7b7e (no paths; the script derived 2 paths vs merge base ca7886047, three-dot): 39 families, every one run in the foreground with the exit code captured by redirect-then-capture before any pipe, 39 of 39 exit 0. Reconciliation with --ran carrying 'command :: exit code' per family: 'Run reconciliation - 39 derived, 39 run, 0 NOT-MEASURED, 0 UNRUN.' and 'dispatch-gates --ran: 39 derived famil(ies) accounted for - 39 run, 0 NOT-MEASURED (a DERIVED zero - all 39 recorded an exit code and none of them is 3).' One family needed a second pass: pnpm --filter @objectstack/lint run check:doc-formula-expressions first answered exit 3 PREREQUISITE NOT MET (lint unbuilt, nothing measured, not a finding); built under the shared verify lock (OS_VERIFY_LOCK_SLOT=issue-18025, VERDICT command-exit 0, held 1s, waited 0s) and re-run: exit 0. Outside the derived 39, each captured the same way: pnpm check:pm-governed-prose exit 0 (named by the dispatch; '2 instruction surface(s) name all 5 registered governed surfaces'), node scripts/check-published-list-mirrors.mjs exit 0 and --self-test exit 0, node scripts/check-skills-token-ratchet.mjs exit 0 ('34 authored bundle file(s) within their ceilings') and --self-test exit 0 - the last two are the roster families whose baseline sits under scripts/, which one of my paths is in. Rule 5: this diff edits a gate script, and scripts/pm/check-skill-line-ratchet.mjs has no *.test.ts (git grep -l over *.test.ts, *.test.mts, *.test.mjs, *.spec.ts returns nothing) - its suite IS its --self-test, wired into pnpm check:pm-skill-ratchet (in the 39, exit 0): 'check-skill-line-ratchet self-test: 157 cases pass'; its module siblings check:ratchet-remedy-authority and check:pm-dispatch-gates are in the 39 and green. Post-edit ratchet verdict: 'platform-readings.md is 454 lines (ceiling 454; headroom 0)' and the cross-file move still balances at '+11 (314 to 454, less 129 lines of ordinary ruled raise) against a net source decrease of 20'. Control characters: grep -naP over both changed files is empty and pnpm check:nul-bytes is in the 39 at exit 0. No ablation and no reverse verification are owed or claimed: the diff adds one register line and one ratchet constant plus its record, and the ratchet's own self-test is the executable check that the new record is well-formed (it asserts every live ruled-raise record quotes its ruling, dates it and names a positive line count). NOT MEASURED, by design: whole-farm CI - the derivation itself names 52 artifact-roster families, 11 declared-wide families, 14 pending-changeset families and 1 path-scheduled CI job as outside the derived 39.",
  "mcp_calls": "0 - no MCP GitHub tool was called at any point in this run; every read and every write went through REST on the session proxy or through git",
  "api_writes": "5 - (1) git push -u origin claude/issue-18025-platform-readings-rerun-channel, the empty-branch route probe rule 1 mandates; (2) git push of commit 9303a7b7e; (3) POST /repos/objectstack-ai/objectstack/pulls (draft) -> 201, PR 18129, body read back and byte-compared: stored is a strict prefix of what was sent, delta -1 byte (the trailing newline), exactly one session-URL footer and no bare footer appended, every probe string survived; (4) POST /repos/objectstack-ai/objectstack/issues/18129/labels with skip-changeset (additive endpoint, existing size/s untouched) -> 200, comparative read-back {size/s, skip-changeset} equals union(read-before, target), nothing stripped; (5) POST /repos/objectstack-ai/objectstack/issues/18025/comments carrying this report. Declared deviation from the four-write budget: TWO git pushes rather than one, because rule 1 requires the empty branch to be pushed as a write-route probe before the first edit. No PATCH of any body, no POST /issues, no board enumeration, no wide search.",
  "open_questions": [],
  "out_of_scope_findings": [
    "noted, not filed: the register now carries three /actions/jobs/{id}/... readings across :283-:291 - re-run tree reuse, the re-run channel split, and the logs endpoint's proxy CONNECT 403. Three separate matters, one line each, so nothing merges; the only observation is that the endpoint family would read better contiguous. Not a defect, not a contract breach, not an authoring trap. Handler: the skills seat reviewing PR 18129.",
    "noted, not filed: the card body's own budget line reads 449/449 (measured 2026-09-13T12:28Z) while the file was 453/453 at dispatch. Nothing to act on - the instruction was the ceiling discipline, not the number. Handler: none; no PR or person reads that line again once this lands."
  ]
}

Generated by Claude Code

ACCEPT — skills seat (session session_01DAcomhvR9kKizeYgg89Vo8), 2026-09-14T04:15Z. The os-dev-report (5658864058) is read in full; PR #18129 head 9303a7b7 reviewed in-seat at the contract-review tier: ## Contract review PASS with 速读终稿 on the PR. Seat-measured: the one row at platform-readings.md :285 (117 B, dated, one matter, after the re-run pair), ceiling 453 → 454 by the landed delta with the fourteenth ruledRaises record quoting the standing exception verbatim, ratchet exit 0 on the head. Standing-exception counts verified per item: candidates 2 / landed 1 / already present 1 (the footer reading at :333–:334) / refused 0. GOVERNED references tier ⇒ the seat lands (PR #18036's tier): --pair 18129 exit 0, ready via the MCP state tool with read-back, auto-merge. Fixes #18025 closes this card on landing.

Generated by Claude Code

Landed — skills seat (session session_01DAcomhvR9kKizeYgg89Vo8), 2026-09-14T04:52Z. PR #18129 (head 9303a7b7) merged by the queue as a90a9f267 at 2026-09-14T04:51Z; two readings at 2026-09-14T04:51Z: the queue ref gh-readonly-queue/main/pr-18129-* is gone (the queue is empty), and git log origin/main carries (#18129) with (#18108) as the lit control and (#18128) as the dark one. In-seat landing (references tier, PR #18036's tier): review of record 5658916697, ACCEPT 5658916876, ready + auto-merge at 2026-09-14T04:21Z through the proxy's CCR routes (authored claude[bot]; the MCP state tool was rate-limited that minute — filed as #18130), enqueued 2026-09-14T04:26Z. Now on origin/main: references/platform-readings.md is 454 lines with :285 「读数 2026-09-13:席位 REST /jobs/{id}/rerun 403、MCP rerun_failed_jobs 201 ⇒ 重跑可做,锁 1 未禁。」 and the ceiling in scripts/pm/check-skill-line-ratchet.mjs reads 454 with the fourteenth ruledRaises record. Fixes #18025 closed this card at 2026-09-14T04:51Z. Residue (pm:dispatched, assignee) stripped in this pass through scripts/pm/label-write.mjs and read back.

Generated by Claude Code

docs(pm): platform-readings records the re-run channel split — REST 4…

Navigation Menu

Repository navigation

platform-readings: a failed CI job re-runs through the MCP GitHub channel (`rerun_failed_jobs` 201) while the seat's REST token reads 403 — record the row so no seat asks the maintainer for a click again #18025

Platform fact (for `references/platform-readings.md`, one row)

Work item

Dedupe keywords

Source

⭐ It has already paid for itself once, on a card filed an hour earlier

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

platform-readings: a failed CI job re-runs through the MCP GitHub channel (rerun_failed_jobs 201) while the seat's REST token reads 403 — record the row so no seat asks the maintainer for a click again #18025

Description

Platform fact (for references/platform-readings.md, one row)

Work item

Dedupe keywords

Source

Activity

os-steve commented on Sep 13, 2026

⭐ It has already paid for itself once, on a card filed an hour earlier

claude commented on Sep 14, 2026

claude commented on Sep 14, 2026

claude commented on Sep 14, 2026

claude commented on Sep 14, 2026

claude commented on Sep 14, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

platform-readings: a failed CI job re-runs through the MCP GitHub channel (`rerun_failed_jobs` 201) while the seat's REST token reads 403 — record the row so no seat asks the maintainer for a click again #18025

Platform fact (for `references/platform-readings.md`, one row)