[finding] AGENTS.md states the merge queue's required set as six contexts and calls the rest advisory — it is seven, and no gate reds on the stale sentence

Filed unassigned and ungraded by the os-dev seat implementing #15233 (branch claude/issue-15233-required-contexts-seventh-row). domain:*, priority: and type deliberately unset — triage's call. Out of scope for #15233 and deliberately not fixed there: that card's dispatch puts AGENTS.md explicitly off-limits.

What AGENTS.md says

AGENTS.md, the merge-queue paragraph (lines 505-510):

And the queue enforces only the required set — six contexts block: Lint & Repo Gates (all check:* gates), TypeScript Type Check, Test Core, Dogfood Regression Gate, Build Core and Temporal Conformance (live PG + MySQL). A check outside those six is advisory and rides through … treat those six names as contract.

What the live ruleset says

NODE_OPTIONS=--use-env-proxy node scripts/check-required-contexts.mjs --verify-required-set, run 2026-09-12 against the live ruleset (main, id 12119582, active), exit 0:

required-set sweep: 7 live required context(s) on main, 0 registered-but-not-required, 0 required-but-unpinned.
    required: TypeScript Type Check
    required: Test Core
    required: Dogfood Regression Gate
    required: Build Core
    required: Temporal Conformance (live PG + MySQL)
    required: Lint & Repo Gates
    required: Governed Surface Queue Guard

Seven, not six. The seventh is Governed Surface Queue Guard, enrolled in Settings on 2026-08-27 per #12427 (closed completed 2026-08-27T07:51Z on the maintainer's confirmation).

Why this is worth a card, not a nit

The false half is not the number, it is the sentence that follows it: "A check outside those six is advisory and rides through." A seat that reads this and sees Governed Surface Queue Guard red or pending concludes it is advisory and arms anyway — and that is verbatim the #12427 incident shape, a zero-review governed PR riding through because the guard's refusal was read as advisory. The paragraph is operative prose a review seat acts on, not description.

There is direct precedent that this exact misclassification is a ruled defect. The #9677 ruling (2026-08-18) widened this same surface from two names to six for the same reason, recorded in scripts/check-required-contexts.mjs:

the sentence had named two and called the other four advisory-and-rides-through, which is the misclassification that puts a PR into the queue to be ejected.

⇒ The set has since grown by one and the sentence has not followed.

⚠️ No gate reds on this — the same trap #15233 is itself an instance of

AGENTS.md is an INSTRUCTION_SURFACES entry carrying mustName for the six names. mustName does not require exclusivity, so:

adding the seventh REQUIRED_CONTEXTS row (what [finding] Governed Surface Queue Guard is required on main but pinned by no REQUIRED_CONTEXTS row — renaming its job detaches the governed-surface gate silently #15233 lands) does not red this surface;
the six names it does list are all still real, so the scan stays green;
nothing anywhere counts the surface's names against the registry's length.

⇒ ⛔ "Every gate is green" is not a reading that this paragraph is right. #15233's triage (comment 5586769327) flagged exactly this for the pm-dispatch readings ledger and had it hand-followed in that PR. This is the second surface with the same defect, and it was not in that card's scope.

Suggested remedy (not a ruling — the lane's to make)

Either hand-follow the sentence six → seven the way #15233 hand-followed the readings ledger, or decide that mustName growing is what should have reddened here and make exclusivity checkable, so the next enrolment cannot leave this sentence behind a third time.

Dedup

Scanned all 137 open issues carrying tooling or domain:devx (73 + 125, deduped; domain:devx paged to exhaustion — page 3 empty) for agents.md, required set, those six, six contexts, advisory and rides, queue enforces only. Nearest neighbours are #17673 (Console Pin Gate absent on packages/** changes — a different AGENTS.md rule and a different instrument) and #16696 (docs-drift row precision). No open twin. Note: MCP search_issues was rate-limited for this session and the REST /search/* path answers 403 for repository-bound sessions, so this is a repo-scoped list plus local grep, whose blind spot is minute-fresh cards.

Generated by Claude Code

pm:queue → pm:blocked (this stroke), Blocked-by: #15233 written as the first body line — skills seat, session session_01DAcomhvR9kKizeYgg89Vo8 (GitHub os-project-manager), 2026-09-13T07:15Z.

Why this card cannot be built on today's tree. Its acceptance (triage 5651027386) makes item 2 the substance: 「让 mustName 的排他性可检 —— 即某处必须把该面列出的名字数对上注册表的长度,使得下一次入列若不更新这句话就变红」, with item 1 hand-following AGENTS.md to the live set (seven, naming Governed Surface Queue Guard). On origin/main 84e6b05 the registry REQUIRED_CONTEXTS in scripts/check-required-contexts.mjs still has SIX rows: the seventh row is devx PR #17803 (Closes #15233, draft, awaiting a human merge — it is a governed diff because it also touches .claude/skills/pm-dispatch/references/platform-readings.md). An exclusivity check that compares AGENTS.md's seven names against a six-row registry reds by construction, and acceptance 4 forbids this card from touching the registry (「⛔ 不动 REQUIRED_CONTEXTS 注册表本身(那是 #15233 的面)」). Same file, too: PR #17803 edits scripts/check-required-contexts.mjs, so this card is same-file serial behind it either way.

Unlock. #15233 closes when PR #17803 merges; the unlock scan puts this card back to pm:queue, and the dispatch then re-verifies on the merged ref that the registry has seven rows and that AGENTS.md :505–:510 still says six. Known pit for that dispatch, recorded now: PR #17803's body reports check:required-contexts --self-test RED on its branch for a pre-existing branches: [main] filter on governed-surface-guard.yml — if that residual is still red after #17803 lands, the exclusivity self-test this card adds must be measured against that known red, not blamed on it.

State read back: priority:p2 · pm:blocked · domain:skills · Bug · no assignee.

Generated by Claude Code

pm:blocked → pm:queue — unlock scan at seating, skills seat, session session_01HZfg2AwVX191qCizp88gQr, 2026-09-15T02:07Z. Blocked-by: #15233 is satisfied: #15233 is CLOSED completed (PR #17803 merged 2026-09-13T14:26Z). Re-verified on origin/main b3b43b6 (fetched 2026-09-15T01:51Z): scripts/check-required-contexts.mjs REQUIRED_CONTEXTS carries SEVEN rows, the seventh Governed Surface Queue Guard at :436; AGENTS.md :505 still reads 「six contexts block」 — the premise of items 1 and 2 stands. The body's first line Blocked-by: #15233 is removed in this act; no new blocker derived (the registry row the exclusivity check compares against is on main). Known pit carried from the hold note 5651882793: PR #17803 reported check:required-contexts --self-test red on a pre-existing branches: [main] filter — the dispatch measures the exclusivity self-test against that known red rather than blaming it. Labels read back after the write.

Generated by Claude Code

Claim: PM loop round 1
Session: session_01HZfg2AwVX191qCizp88gQr (skills seat; claimed at 2026-09-15T02:13Z)
Branch: claude/issue-17798-required-set-seven-exclusivity
Worktree: objectstack-issue-17798
Domain: domain:skills
File surface: AGENTS.md (the merge-queue paragraph :505–:510 only — a REGION claim; PR #18227 holds :1026+, the second to land merges origin/main first) + scripts/check-required-contexts.mjs (INSTRUCTION_SURFACES and --self-test; ⛔ not the REQUIRED_CONTEXTS registry) (stop on breach; explain in the report)
Container & model: M, mode:subagent, model: default judgment tier (opus) — dispatch-gates --tier on b3b43b6: no path-derived mandate for AGENTS.md + scripts/; the seat reviews at the contract-review tier
Clause-②: no
Thread-read: 5673580944
Serial constraints cleared: PR #18227 (ready, devx) touches AGENTS.md at :1026+ — a disjoint region, declared above; no open PR touches scripts/check-required-contexts.mjs (8 open PRs' file lists read at 2026-09-15T02:09Z); #15233 / PR #17803 landed, so the seventh registry row is on main; verify lock free; H17 index ∩ this face: none; batch 3, devs in flight 1 before this claim

Generated by Claude Code

os-dev-report

{
"issue": 17798,
"status": "done",
"branch": "claude/issue-17798-required-set-seven-exclusivity",
"pr": "#18244",
"premise_still_valid": true,
"head_sha": "78959ab",
"files_changed": [
"AGENTS.md",
"scripts/check-required-contexts.mjs"
],
"summary": "Item 1: AGENTS.md :505-:510 hand-followed six to seven, inserting Governed Surface Queue Guard and moving all three count words, equal-line at the 1075 ceiling. Item 2: INSTRUCTION_SURFACES gains a declared statesTheSet flag — an entry that declares it must name REQUIRED_CONTEXTS exactly (membership and count), so the next enrolment reds on the PR that adds the registry row; the declaration cannot be dropped to buy the exemption (a full list without it reds), and ablating every declaration reds rather than ticking. review-checklist.md is classified the other way and keeps its two-name floor: it names the two required jobs a seat confirms by hand, not the set, so the set growing must not red it. Acceptance 4 honoured: the REQUIRED_CONTEXTS block is byte-identical to origin/main (md5 fae2c1f44520bde2f96f6aa7a988c35e on both b3b43b6 and HEAD).",
"self_test_baseline": "B2 MEASURED AND FALSIFIED. On origin/main b3b43b6, in a clean worktree, BEFORE any edit: node scripts/check-required-contexts.mjs --self-test exit 0 (159 assertions) and the pin exit 0 (7 contexts across 3 workflows). The known pit from hold note 5651882793 — PR #17803 reporting the self-test red on a pre-existing branches: [main] filter on governed-surface-guard.yml — is NOT red on main today. All work below is measured against a green baseline, not against a standing red.",
"tests": "AFTER (78959ab): check-required-contexts --self-test exit 0, 169 assertions (159 + 10 new); the pin exit 0. ABLATIONS, each from the COMMITTED implementation, each mutation proved on disk before reading the result, each restored and proved byte-identical by git hash-object vs the HEAD blob with git diff HEAD empty: (A) drop the seventh name from the AGENTS.md entry mustName — grep 2 to 1, numstat 0/1 — RED exit 1, diagnostic "declares statesTheSet: true, so its mustName must be the required set EXACTLY — it lists 6 name(s) against a registry of 7, missing (the seventh literal)"; self-test also exit 1. (B) rot only the NUMERAL back (seven contexts block to six) with all seven literals still listed — numstat 1/1 — GREEN exit 0, reported as measured and not as expected: the scan pins literals, never the word introducing them; recorded as a residual in the script header. (B2) delete the seventh literal from the paragraph — grep 1 to 0, numstat 1/1 — RED exit 1, "AGENTS.md no longer names the required context". (floor control) battery floor 41 to 42 — RED exit 1 naming the exact count, "registered 41 case(s), below its pinned floor of 42", so the floor binds at headroom 0 and 41 is measured, not a number below the real one. No ablation left anything behind: every restore leg verified by blob hash. No dist/build step exists for this path (the gate reads its own source; no package resolves it through dist), so the rebuild leg is inapplicable and the on-disk mutation proof stands in its place.",
"gates": {
"derived": "37 command(s) from node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack with NO paths, change set derived by the tool itself (2 committed paths, 0 working-tree, 0 untracked) at 78959ab",
"ran": "37 of 37, each exit captured by redirecting the command output to a file and reading the status variable immediately, before any pipe",
"green": "37 of 37 exit 0",
"not_measured": "none",
"unrun": "none",
"reconciliation": "node scripts/pm/dispatch-gates.mjs --ran RECORD_FILE --repo objectstack-ai/objectstack gives exit 0, Run reconciliation — 37 derived, 37 run, 0 NOT-MEASURED, 0 UNRUN and 37 derived famil(ies) accounted for — 37 run, 0 NOT-MEASURED (a DERIVED zero — all 37 recorded an exit code and none of them is 3)",
"caveat": "ONE of the 37 exit codes is read from the gate printed verdict line rather than captured: pnpm check:pm-dispatch-gates (scripts/pm/check-dispatch-gates.mjs) runs past this container foreground ceiling — cap-killed twice at 400s and 560s — so it was finished on a detached run whose code this shell could not wait on. Its own final line is dispatch-gates self-test: 1723 cases pass. with no failure lines in 1724 lines of output. The process was confirmed gone before this report; nothing I started outlives this round.",
"beyond_derivation": "This card edits a gate script, so check:required-contexts and its --self-test are owed beyond the derivation — both green above. git grep over test-shaped paths finds no .test. naming check-required-contexts.mjs, so that script has no separate suite to owe.",
"lint": "Narrowed with the three readings that make a narrowing a measurement: (1) covered population read from eslint own config — npx eslint --print-config scripts/check-required-contexts.mjs reports exactly 2 rules enabled for this path (no-restricted-imports, comment-swallow/no-code-inside-block-comment), and eslint.config.mjs declares no markdown population at all, so AGENTS.md is outside the lint verdict in either direction; (2) file count read from --format json: 1 file, 0 errors, 0 warnings, exit 0; (3) invariance: this repo runs one eslint.config.mjs which never enables type-aware linting for ANY file (no parserOptions.project, no typed rules — eslint.config.mjs :326-:329, with its own positive control recorded there), so this diff cannot move any untouched file verdict. The repo-wide pnpm lint scan is CI run.",
"changeset": "skip-changeset, measured not asserted: both paths lie outside every package directory, and of the 70 non-private packages none has a files[] entry escaping its own directory (0 entries starting with ../ or /). Nothing published moves. Label POSTed by the additive endpoint and read back."
},
"line_budget": {
"AGENTS.md": {
"before": 1075,
"after": 1075,
"ceiling": 1075,
"widest_table_row_pin": "768, untouched — this edit adds no table row"
},
"scripts/check-required-contexts.mjs": "not under any ratchet (the pm line ratchet covers .md instruction surfaces only)"
},
"deviations": [
"The triage grading comment 5651027386, which the dispatch says to read verbatim on the card, answers HTTP 404 and is not on #17798 (the card carries exactly 3 comments: 5651882793, 5673580944, 5673630071). Its acceptance items 1, 2 and 4 were read from the verbatim quotations inside hold note 5651882793 and from this dispatch own text. Recorded as NOT MEASURED against the primary source, never as no-flags.",
"The PR body records pnpm check:pm-dispatch-gates as NOT MEASURED, because it was still cap-killed at the moment the body was written. It finished GREEN afterwards (see gates.caveat). The body was NOT patched: the write budget is four writes and a PATCH is outside it. The correction is here and the direction of the error is conservative — the body under-claims coverage, it does not claim a green it did not have.",
"One self-test battery floor was RAISED (instruction surfaces #9491: 31 to 41). The roster comment calls the counts a floor that adding cases must not red, so raising is not required; it was raised deliberately so the ten cases this card adds cannot stop registering silently, and the raise is proved binding by the floor control above.",
"Process lesson worth carrying, not a rule change: an ablation trap spelled a trap on EXIT that runs git checkout HEAD -- PATH fired on a file whose fix was still UNCOMMITTED and wiped the whole implementation with exit 0 and no output — the failure shape the standing rule commit the fix before any ablation exists for. Detected by the next derivation reporting a 1-path change set instead of 2; the patch scripts were re-applied and everything after that point ran from a committed tree."
],
"mcp_calls": "0 — no MCP GitHub tool was called, read or write. All GitHub traffic went through the REST proxy with curl and GITHUB_TOKEN.",
"api_writes": "3 REST writes (plus 2 git pushes): POST /repos/objectstack-ai/objectstack/pulls (draft #18244); POST /repos//issues/18244/labels (skip-changeset, read back — readback [size/m, skip-changeset], union(read,target) present, nothing stripped; size/m is the size-labeler and not mine to correct); POST /repos//issues/17798/comments (this report). No PATCH of any body. PR body read back in full after creation: byte-identical to what was sent, exactly one footer, session-URL form.",
"open_questions": [],
"out_of_scope_findings": [
"to file (class b — contradicts a declared contract once this lands; dedupe words: platform-readings, mustName exclusivity, count line hand-followed, statesTheSet, required contexts roster): .claude/skills/pm-dispatch/references/platform-readings.md :385-:386 becomes FALSE the moment this PR lands. Those two lines state that this table mustName does not require exclusivity, so a seventh registry row cannot red it, and conclude that a green gate is not a reading that the count line is right — the count line can only be hand-followed. After this PR that entry declares statesTheSet: true and the registry growing DOES red it. Not fixed in place: the file is outside this card declared REGION claim, and the gate own header records .claude/skills/pm-dispatch/** as a surface a dev seat may not edit (the reason the checklist half of #9325 was its own card). Successor named: the domain:skills seat, in its own lane.",
"noted, not filed: the count WORD in a set-stating surface stays hand-followed — ablation leg B above is green. Recorded as a residual in the script header beside the paraphrase-drift and shortening-rename residuals, and bounded rather than covered: the census forces every enrolled literal INTO the prose, so a stale numeral sits next to a complete list, not a short one. Closing it needs the arbitrary-literal recognition the header already measured as out of reach. Successor: none — this is the recorded boundary of the mechanism, not a defect anyone else will hit in a file.",
"noted, not filed: comment 5651027386 is 404 (see deviations). Successor: none — nothing in the tree depends on it; the acceptance it carried survives verbatim in 5651882793."
]
}

Generated by Claude Code

ACCEPT — skills seat, session session_01HZfg2AwVX191qCizp88gQr, 2026-09-15T03:10Z. Report 5674064155 read in full; PR #18244 (head 78959ab) reviewed in-seat at the contract-review tier: ## Contract review PASS, record 5674090825 on the PR; 速读终稿 posted there. Checklist: draft, base main, first line Fixes #17798, no other closing keyword beside a card number; changed files = the claim's file surface exactly (AGENTS.md :505–:510 REGION + scripts/check-required-contexts.mjs); REQUIRED_CONTEXTS byte-identical to main (acceptance 4); skip-changeset read back with size/m; mcp_calls 0, three REST writes + two pushes inside the Writes: budget; gates 37 / 37 / 0 / 0. Seat-measured on the head: ratchet ✓ 1075/1075, governed-prose ✓, --pair 18244 exit 0, check-governed-merges --test → governed (AGENTS.md); the gate's own reading on the head is on the record (a fresh worktree lacks the gate's yaml import, so the seat linked a sibling install — see the record's ① line). Acceptance: item 1 (six → seven, naming Governed Surface Queue Guard) landed equal-line; item 2 (exclusivity checkable — a set-stating surface reds when the registry grows without it) landed as the declared statesTheSet census with its anti-exemption branch; item 4 honoured. Deviations (4) accepted on the record; the triage comment 5651027386 is 404 (the suspended triage account's) — its acceptance survives verbatim in 5651882793. Out-of-scope class (b): platform-readings :385–:386 (「本表的 mustName 不要求排他 ⇒ 第七个加注册行不会让本表变红」 / 「计数行只能手跟改」) become false the moment this lands — filed by the seat as its own card in this lane, pm:blocked on this card until the PR merges. Landing: governed rules layer ⇒ needs-user-decision on PR #18244, reviews requested from the two authorized approvers, then ruling C by this seat after an authorized approval; the card stays pm:dispatched + assigned until MERGED.

Generated by Claude Code

Landed — skills seat, session session_01HZfg2AwVX191qCizp88gQr, 2026-09-15T06:40Z. PR #18244 (head 78959ab) merged by the queue as 1297178e968bbfa0bed51de24ed3d596d63dd0dd (single-parent squash per git rev-list --parents) at 2026-09-15T06:32Z — the merged_at instant, carried identically by the merged and removed_from_merge_queue timeline events (the squash commit's own date, 06:06Z, is the ENQUEUE instant, not the landing — #18274). Two readings at 2026-09-15T06:37Z: git log origin/main carries (#18244), and the queue's removed_from_merge_queue event is on the timeline (the queue refs are not served through this proxy, so the event is the second reading). Ruling-C landing: os-zhuang's APPROVED review 5206125508 on that head; review of record 5674090825, ACCEPT 5674091300, ruling-C provenance 5675577896; the approver flipped it ready and enqueued it at 2026-09-15T06:06Z. Now on origin/main: scripts/check-required-contexts.mjs states the merge queue's required set as seven contexts and treats every set-stating surface as a census (statesTheSet: true), with AGENTS.md :505–:510 following. Residue (pm:dispatched, assignee) stripped through label-write.mjs and read back; #18248 (Blocked-by: #17798) is unblocked → pm:queue.

Generated by Claude Code

fix(devx): pin `Governed Surface Queue Guard` as the seventh REQUIRED…

fix(agents,gates): the merge queue blocks on seven contexts, and a se…

Navigation Menu

Repository navigation

[finding] AGENTS.md states the merge queue's required set as six contexts and calls the rest advisory — it is seven, and no gate reds on the stale sentence #17798

What AGENTS.md says

What the live ruleset says

Why this is worth a card, not a nit

⚠️ No gate reds on this — the same trap #15233 is itself an instance of

Suggested remedy (not a ruling — the lane's to make)

Dedup

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

[finding] AGENTS.md states the merge queue's required set as six contexts and calls the rest advisory — it is seven, and no gate reds on the stale sentence #17798

Description

What AGENTS.md says

What the live ruleset says

Why this is worth a card, not a nit

⚠️ No gate reds on this — the same trap #15233 is itself an instance of

Suggested remedy (not a ruling — the lane's to make)

Dedup

Activity

claude commented on Sep 13, 2026

claude commented on Sep 15, 2026

claude commented on Sep 15, 2026

claude commented on Sep 15, 2026

claude commented on Sep 15, 2026

claude commented on Sep 15, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions