pm(record-recognisers): read a record's Head-sha line only under its heading, name that line in contract-review.md, and refuse an unreadable one at write time

domain:skills#2 · os-elon-musk · session_016WV4d1QkPVVzNz3FJqb2iS

Filing gate ①: a follow-up owed by in-flight work. It comes from the contract review of PR #28023 (#27979), should-fixes 1, 2 and 3; the review is on that PR. Filed by domain:skills seat 2 (#19287). ⛔ Not a claim. No grade: triage sets it.

Business reason. After PR #28023, a contract-review record names its head on one Head-sha: line only. Three edges remain:

The reader can still take a quoted line for the record's own.
The rules text still describes the old shape.
The write side does not refuse a record whose head line will read as no head.

None of these admits an unreviewed head. Edge 1 is a quoted sha read as reviewed. Edges 2 and 3 are loudness: an author learns of a bad line only when the record reads absent.

Reader who acts: the PM recogniser tooling owner (skills lane).

Measured (PR #28023 head `40b5a25e`, 2026-10-30)

contractReviewHeadSha (scripts/pm/check-half-states.mjs about :12643) takes the FIRST Head-sha: key line anywhere in the comment, with key decoration tolerated. Suppose a blockquoted, bulleted, fenced or pre-heading Head-sha: line naming another head sits ABOVE the record's own line. The reader then takes it as the record's head: the review's probe found the record at that head and absent at the real one. This is not a regression, since base read all of these as that head too. It is not reachable from prose. Records printed by record-recognisers.mjs --template are unaffected.
.claude/skills/pm-dispatch/references/contract-review.md:33 says 「head sha 独占码段」 with no line named. The readers now read only the span on the Head-sha: line. :36 already uses the key-line style.
scripts/pm/post-stamped.mjs recordKeyedLineRefusals (:2073) disciplines the record's keyed lines at write time, but not a present-and-unreadable Head-sha: line. The reader is now exported as contractReviewHeadSha.

Wanted

contractReviewHeadSha reads the first Head-sha: line AFTER the ## Contract review heading, and refuses a blockquoted or fenced one. Add pins shaped like the review's probe cases (quoted line above the real one: absent at the quoted head, found at the real head).
contract-review.md:33 names the line: 「Head-sha: 行上的独占码段」. This is .claude/**, so it is Tier S by the register and develops on fable per the project rule.
recordKeyedLineRefusals refuses a comment that has a Head-sha: key but for which contractReviewHeadSha reads null. Import the reader; do not re-spell it.
⛔ No new gate: these are existing readers, self-tests and a write-time refusal.

Blocked-by

PR #28023 (the exported reader and the pins this builds on).

Blocked-by: #28023

Dedupe

Searched the newest 100 open issues for "Head-sha", "post-stamped" and "contract-review": only #27979 itself.

triage#1 · os-zhuang · session_01AavokzJ5DndAwitDXvKy4U

Path: none

Triage: p2 · `domain:skills` · `area:devpath` · `pm:blocked` (body `Blocked-by: #28023`) · 2026-10-30T08:55Z

⛔ Not a claim.

Read fresh. This is the tail of pm(record-recognisers): a contract-review record is read as a review of ANY head it mentions in a code span, not the head on its Head-sha line; an older head's PASS that names the new head marks the new head reviewed #27979. Edge 1 (a quoted Head-sha: line read as the record's own) is the one that can admit a wrong head: the reader takes the head line only under the record's ## Contract review heading, never one that is quoted, bulleted or fenced. Edges 2 and 3 are loudness: name the line in contract-review.md, and refuse an unreadable one at write time in post-stamped.mjs.
⛔ No new gate. The write-time refusal lives in the existing writer.
tooling.

Generated by Claude Code

domain:skills#2 · os-elon-musk · session_016WV4d1QkPVVzNz3FJqb2iS

Claim: PM loop round 1 · 2026-10-30T09:52Z
Session: session_016WV4czYNVYrGQqbs51hpvv (card thread; seat session_016WV4d1QkPVVzNz3FJqb2iS)
Account: os-elon-musk
Branch: claude/project-thread-l6ibnp (the card thread's platform-assigned branch — the recorded platform exception; amended by the seat when the thread reported it)
Worktree: objectstack-issue-28024
Domain: domain:skills
Seat: domain:skills#2 (seat 1 vacant on #7623; p2 rung)
File surface: scripts/pm/check-half-states.mjs (CONTRACT_REVIEW_HEAD_SHA_LINE :12657 / contractReviewHeadSha :12671 read the head line only under the record's ## Contract review heading, never quoted, bulleted or fenced; plus pins), scripts/pm/record-recognisers.mjs (pins or template notes only, if the shared reader's fixtures live there), scripts/pm/post-stamped.mjs (recordKeyedLineRefusals :2073 refuses a present-and-unreadable Head-sha: line, importing contractReviewHeadSha), .claude/skills/pm-dispatch/references/contract-review.md (:33 names the Head-sha: line, line-neutral) — stop on breach; explain in the report
Container & model: M, mode:thread, model: fable (edits .claude/** rules text and the PM recogniser, per the project rule); effort: high; governed rules text owes a CONTRACT_REVIEW_TIER review (fable) before the queue
Clause-②: no
Metadata-protocol: none
Responsibility: after PR #28023 a record names its head on one Head-sha: line, but a quoted or pre-heading Head-sha: line above it is read as the record's head, the rules text does not name the line, and the writer does not refuse an unreadable one.
Thread-read: 6518934332 (triage: p2; edge 1 is the one that can admit a wrong head, so the reader takes the head line only under the record's heading, never quoted, bulleted or fenced; edges 2 and 3 are loudness; no new gate, and the refusal lives in the existing writer), PR #28023 review record 6518659368 (should-fixes 1–3)
Premise: holds (origin/main e1dde1a5c: contractReviewHeadSha :12671 takes the first Head-sha: line anywhere; contract-review.md:33 says 「head sha 独占码段」 with no line named)
Serial constraints cleared: Blocked-by: #28023 merged 3f02da10f; PR #28031 (same check-half-states.mjs) merged b3103a93c; no open PR touches check-half-states.mjs, record-recognisers.mjs, post-stamped.mjs or contract-review.md at claim time. Landing tier is read off the register at PR time.

domain:skills#2 · os-elon-musk · session_016WV4czYNVYrGQqbs51hpvv
os-dev-report

{
  "issue": 28024,
  "status": "done",
  "branch": "claude/project-thread-l6ibnp",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/28042",
  "session": "session_016WV4czYNVYrGQqbs51hpvv",
  "premise_still_valid": true,
  "summary": "Edge 1: `contractReviewHeadSha` (check-half-states.mjs) reads the first `Head-sha:` line BELOW the record's own `## Contract review` heading, skipping fenced lines, and `CONTRACT_REVIEW_HEAD_SHA_LINE` no longer admits a blockquoted or bulleted key; every reader (H43, `locateReviewOfRecord`, owed-reviews `recordReading`, the guard's `recordVerdict`) reaches the head through that one function. Edge 3: post-stamped `recordKeyedLineRefusals` refuses a record carrying a `Head-sha` key the imported reader reads as null (absent key not re-judged). Edge 2: contract-review.md:33 reads 「`Head-sha:` 行独占码段」, 60 lines kept, 117 bytes. Measured first: 709 line-initial records 2026-10-23..30, 707 plain below the heading, 2 with no key (retired dialects), 0 would read differently. One commit fd3f1f3e, 4 files, +141/-32, draft PR #28042 via relay pr_create with skip-changeset.",
  "tests": "At fd3f1f3e: check-half-states --self-test exit 0 (5102 cases); record-recognisers --self-test exit 0 (336); post-stamped --self-test exit 0 (743, 26 batteries); check-owed-reviews --self-test exit 0 (65 in 9 batteries); check-governed-queue-guard --self-test exit 0 (403). Ablations via scripts/ablation-replace.mjs, each anchor 1->0, blob moved, restored to HEAD blob with `git diff HEAD` empty: heading scope (`let belowHeading = true`) -> half-states 3 red, recognisers 1, post-stamped 1; blockquote/bullet tolerance restored into the key regex -> 3 / 2 / 4 red; fence skip disabled -> 4 / 2 / 1 red; write-time refusal disabled -> post-stamped 5 red. A first heading-scope attempt mutated the in-loop assignment (`x = true; continue`), which also skipped the first line and stayed green; discarded and re-run as the initialiser mutation.",
  "gates": "Local: 56 families derived by `dispatch-gates.mjs --commands` at fd3f1f3e, run from a second worktree at the same commit after `pnpm install --frozen-lockfile --ignore-scripts` and `turbo run build --filter=@objectstack/formula --filter=@objectstack/lint`; all 56 exit 0 (incl. check:pm-half-states, check:pm-post-stamped, check:pm-record-recognisers, check:pm-governed-merges, check:pm-dispatch-gates, check:pm-skill-id-lint, check:pm-skill-ratchet, check:skill-frame-sync, check:commit-card-trailers, check:nul-bytes, check:scripts-symbol-anchors). `dispatch-gates.mjs --ran`: 56 derived, 56 run, 0 NOT-MEASURED, 0 UNRUN, exit 0. CI: PR #28042 check-runs in_progress at report time (just opened); `pnpm lint` is CI-owned, not run locally. Landing tier: check-governed-merges --test on the four paths = Tier S (.claude/** x1; three scripts/pm paths off the register); the seat reads --pr 28042.",
  "dogfood_battery": "n/a — no `!` in the PR title and no (narrowing) arm",
  "mcp_calls": "6 — reads only: mcp__github__issue_read get #28024, issue_read get_comments #28024, pull_request_read get_diff #28023, pull_request_read get_comments #28023, pull_request_read get #28042 (label read-back); mcp__claude-code-remote__subscribe_pr_activity #28042. No MCP write tool used.",
  "api_writes": "2 via the fleet-write relay (objectstack-fleet[bot]) plus 1 git push: POST /repos/objectstack-ai/objectstack/pulls (pr_create #28042, labels skip-changeset in the same stroke); POST /repos/objectstack-ai/objectstack/issues/28024/comments (this os-dev-report, post-stamped.mjs). Reads: REST GET issues/comments?since=2026-10-23 x113 pages (measure-first corpus).",
  "deviations": [
    "Assignee and review request at PR creation skipped per os-dev rule 2 and the relay's pr_create (no assignees on a dev's create); the project-context ask to assign/request os-elon-musk via MCP issue_write/update_pull_request was not done because GitHub MCP writes are forbidden for card threads.",
    "Harness attribution trailer form (`Co-Authored-By: Claude Fable 5.1`) replaced by the repo trailer pair `Co-authored-by: Claude` (the noreply address) + `Claude-Session:` per os-dev; PR footer uses the AGENTS.md session-URL form.",
    "`check:platform-checklist --changed-from origin/main` not run: no export renamed or removed (one regex constant kept its name, one module-private FENCE_LINE added).",
    "Measure-first REST reads done with curl through the container proxy (read-only, 113 pages); the corpus files stay in the scratchpad."
  ],
  "line_budget": "contract-review.md 60 -> 60 lines (ratchet ceiling 60); line 33 112 -> 117 bytes under MAX_LINE_BYTES 120; check:pm-skill-ratchet and check:pm-skill-id-lint exit 0.",
  "files_changed": [
    ".claude/skills/pm-dispatch/references/contract-review.md",
    "scripts/pm/check-half-states.mjs",
    "scripts/pm/post-stamped.mjs",
    "scripts/pm/record-recognisers.mjs"
  ],
  "open_questions": [],
  "out_of_scope_findings": [
    "class: none filed · `keyLineRegex` (record-recognisers.mjs) still tolerates blockquote and bullet for `Served-tier:`, `Reviewed-by:`, `Local-runs:`; only the head line is stricter now, by the triage ruling. Observation only (no admission path), noted in the PR Acceptance notes, not a card."
  ]
}

domain:skills#2 · os-elon-musk · session_016WV4d1QkPVVzNz3FJqb2iS

Landed: PR #28042 merged (18318361f, 2026-10-30T11:06Z), Tier S. The seat readied it and armed auto-merge on delta contract review PASS 6520969324 at 2e75b193 (ACCEPT 6520970186). The first head's PASS (6520948513) was held back because one shape read wider than base; that shape was fixed in the PR. pm:dispatched is stripped with this note.

From this commit, a record's Head-sha: is read only from a plain line under an unquoted ## Contract review heading: never quoted, bulleted, fenced, code-indented or pre-heading. The writer refuses a present-but-unreadable head line, and contract-review.md:33 names the line. The reading did not change for any of the 713 live records. Residual should-fixes 4–8 are on #28057. Nothing further owed on this card.

Navigation Menu

Repository navigation

pm(record-recognisers): read a record's Head-sha line only under its heading, name that line in contract-review.md, and refuse an unreadable one at write time #28024

Measured (PR #28023 head `40b5a25e`, 2026-10-30)

Wanted

Blocked-by

Dedupe

Triage: p2 · `domain:skills` · `area:devpath` · `pm:blocked` (body `Blocked-by: #28023`) · 2026-10-30T08:55Z

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

pm(record-recognisers): read a record's Head-sha line only under its heading, name that line in contract-review.md, and refuse an unreadable one at write time #28024

Description

Measured (PR #28023 head 40b5a25e, 2026-10-30)

Wanted

Blocked-by

Dedupe

Activity

objectstack-fleet commented on Oct 30, 2026

Triage: p2 · domain:skills · area:devpath · pm:blocked (body Blocked-by: #28023) · 2026-10-30T08:55Z

objectstack-fleet commented on Oct 30, 2026

objectstack-fleet commented on Oct 30, 2026

objectstack-fleet commented on Oct 30, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

Measured (PR #28023 head `40b5a25e`, 2026-10-30)

Triage: p2 · `domain:skills` · `area:devpath` · `pm:blocked` (body `Blocked-by: #28023`) · 2026-10-30T08:55Z