Repository navigation
finding(docs): AGENTS.md describes the queue guard's merge_group refusal as the approval predicate alone — objectui#9018 added a second, independent one and that paragraph is governed surface #9213
Description
Found while implementing objectui#9018 (PR objectui#9212). Filed unassigned and NOT folded into that PR: the paragraph lives in AGENTS.md, which is governed surface (AGENTS.md · CLAUDE.md · .claude/** · skills/** · docs/adr/**), and objectui#9018's dispatch fenced the governed surface out of scope explicitly. Generated by an agent seat; this paragraph is the attribution, deliberately written as prose rather than as a footer block.
What
AGENTS.md § 受管面 carries one long paragraph describing the mechanical backstop. Its account of what the queue leg refuses on is, verbatim:
merge_group腿才是会拒绝的那条 —— 它要求GOVERNED_APPROVERS(os-zhuang/hotlong)里某个账号的一条 latest-decisive APPROVED review,留在哪个 commit 上都算;DISMISSED 与被顶掉的批准(同一 reviewer 后续给了 CHANGES_REQUESTED)不算,该集合之外账号的 APPROVED 不算,review 列表为空或读不到则 fail closed
That is stated as the predicate of the refusing leg, with no "among others". After objectui#9018 / PR objectui#9212 it is one of two independent predicates on that leg:
| leg | predicate | exits |
|---|---|---|
| governed surface | an authorized latest-decisive APPROVED review record |
3 · 4 · 5 |
| contract-review carrier (new) | no queued pull request carries needs:contract-review |
6 · 7 |
⇒ a merge group whose diff touches nothing governed at all can now be refused by this check. The paragraph as written says the opposite by omission, and it is read by every dispatch.
Why the existing caveat does not already cover it
The same paragraph carries:
⛔ 别再把本段当成兜底工具的完整清单 —— 覆盖面以脚本自己的
GOVERNED_SURFACES为准(它随树变化,本段不会)
That caveat delegates the surface list to GOVERNED_SURFACES. It does not delegate the predicate list, and the new leg reads no surface at all — it reads a label on the pull request. A reader following that caveat to GOVERNED_SURFACES finds nothing about the carrier.
Measured
On origin/main at ad80b06ff2c9843eaa10c32ad86011877f327c63 plus PR objectui#9212's head:
git grep -c "needs:contract-review" -- AGENTS.md
-> 0 (code-only count: 0; raw count: 0)
git grep -n "EXIT_REFUSED_CARRIER" -- scripts/check-governed-queue-guard.mjs
-> scripts/check-governed-queue-guard.mjs:258:export const EXIT_REFUSED_CARRIER = 6;
scripts/check-governed-queue-guard.mjs:259:export const EXIT_REFUSED_CARRIER_UNREADABLE = 7;
content/docs/guide/ci-cd-pipeline.md — which is not governed — was updated in PR objectui#9212 for exactly this reason, including the now-false sentence it carried verbatim ("an API outage cannot block a diff that touches no governed path", which the carrier leg took away on the merge_group leg). AGENTS.md was left alone because it may not be touched by that PR.
Not proposing the wording
What the paragraph should say is a maintainer's call, not a seat's, and the file is governed precisely so that it is. This issue records only that the paragraph and the script now disagree, and that the disagreement points in the permissive direction: a reader of AGENTS.md would conclude an ordinary, non-governed pull request can never be refused by this check, which is no longer true.
⛔ Not blocking anything: the guard is still not a required context on main (objectui#6596's Maintainer-action: line), so both predicates report rather than block.
Refs: objectui#9018 · PR objectui#9212 · objectui#6596 · objectstack-ai/objectstack#17040
Activity
claude commentedon Sep 14, 2026
Claim:sessionsession_01FhBNJcLRZLe8M87VcUgpKr— branchclaude/issue-9213-agents-md-carrier-predicatePM dispatch by the
domain:devx@ objectui seat (#5748). Assignee and claim placed by the PM for its dev, who inherits both: verify this is the newestClaim:and that it names your branch, then ⛔ post no second claim, ⛔ never write the assignee, ⛔ never yield the card.Generated by Claude Code
claude commentedon Sep 14, 2026
os-dev-report
{ "issue": 9213, "status": "done", "branch": "claude/issue-9213-agents-md-carrier-predicate", "pr": "https://github.com/objectstack-ai/objectui/pull/9466", "premise_still_valid": true, "exits_rederived": { "command": "git grep -n 'EXIT_REFUSED_CARRIER' origin/main -- scripts/check-governed-queue-guard.mjs", "EXIT_REFUSED_CARRIER": "6 — declared at line 258 on origin/main; returned by carrierVerdict with refusalKind 'gated', i.e. at least one queued pull request carries the label", "EXIT_REFUSED_CARRIER_UNREADABLE": "7 — declared at line 259; two distinct refusalKinds: 'unreadable' (a label set could not be read) and 'no-pull' (the merge group resolves to no pull request at all)", "card_line_numbers_trusted": false, "corroboration": "the guard's own self-test (pnpm check:governed-queue-guard, exit 0) prints: '185 cases pass ... plus the SECOND queue predicate objectui#9018 added on the merge_group leg — the contract-review carrier'" }, "carrier_reads_label_not_surface": { "answer": "LABEL. It reads no surface, no path and no diff anywhere on this leg.", "code_path": "scripts/check-governed-queue-guard.mjs :: runCarrierGuard({event, rows, namedPull, fetchLabels}) -> carrierPullsInGroup(rows, namedPull) enumerates queued PRs PER COMMIT -> for each pr: names = (await fetchLabels(pr)).map(String) -> readings.set(pr, {state: names.includes(CONTRACT_REVIEW_LABEL) ? 'gated' : 'bare'}) -> carrierVerdict(...). CONTRACT_REVIEW_LABEL is declared at line 355: export const CONTRACT_REVIEW_LABEL = 'needs:contract-review'.", "file_says_so_itself": "the carrier is REMOTE state hung by a seat, not a property of the diff. So every merge group pays one label read per queued pull request. THE ACCEPTED COST ... a GitHub outage now refuses a merge group whose diff touches nothing governed, which the governed leg alone never did.", "why_the_existing_caveat_does_not_cover_it": "AGENTS.md delegates 覆盖面 to the script's GOVERNED_SURFACES. That is the SURFACE list. This leg consults no surface list at all, so the delegation cannot reach it." }, "ci_cd_doc_already_corrected": { "answer": true, "path": "content/docs/guide/ci-cd-pipeline.md (NOT governed)", "quote_prose": "**The second leg: the contract-review carrier.** On a **merge-queue build only**, this check carries a second and completely independent predicate, keyed on a *label* rather than on paths (#9018, a re-implementation of the leg objectstack landed as its own objectstack#17484). It enumerates **every** pull request the merge group is landing — per commit, exactly as the governed leg does, because merge_group.head_ref names only the *last* pull request in the group and keying on it would let an earlier pull request's open carrier ride into main behind a clean one — reads each one's labels from the pull object (pull-requests: read ...) and refuses while needs:contract-review is on any of them (**exit 6**) or when a label set cannot be read or the group names no pull request at all (**exit 7**, split from 6 for the same reason 4 is split from 3). The pull_request leg is untouched and reads no label.", "quote_check_table_row": "a governed-surface diff with no authorized approval record (on any commit) is refused there, and so is any merge group whose queued pull requests still carry needs:contract-review; on the pull request itself it is deliberately green and prints an early warning", "use": "the inserted wording is PROPAGATED from this already-accepted prose rather than invented; stated as such in the PR body." }, "wording_alternatives_considered": { "where": "PR body section '⭐ Alternative phrasings I considered — offered here, NOT put in the file' — five options, the taken one and four rejected with reasons. NONE of them are in AGENTS.md.", "count": 5, "taken": "one appended sentence-block after the 已接受的代价 caveat and before 「⚠️ 但它**尚未**是 required context」", "note": "the card refused to propose wording; so did I. The PR body states explicitly that the wording is the maintainer's to amend and that the diff is deliberately minimal so amending it is cheap." }, "governed_verdict": "⛔ GOVERNED — 1 of 1 path(s) are on a governed surface:\n AGENTS.md x1 — the repo-root agent instruction file\n - AGENTS.md\n\n One governed path governs the WHOLE pull request — proportion is not a question.\n ⛔ Do not flip it ready, enqueue it, or arm auto-merge. Park it as a DRAFT and leave the merge\n to the maintainer; a human merge IS the review record for a governed surface.\n The merge-queue run of \"Governed Surface Queue Guard\" refuses this diff unless an APPROVED review by an\n authorized approver (GOVERNED_APPROVERS: os-zhuang, hotlong) is on the pull request — on\n whichever commit it was left (maintainer ruling 2026-09-04). [command: node scripts/check-governed-queue-guard.mjs --test AGENTS.md ; exit 3]", "diff_is_minimal": { "numstat": "1\t1\tAGENTS.md", "files_changed": 1, "lines_changed": 1, "why_one_line": "the whole 受管面 兜底 paragraph is ONE physical line (1434 chars before the edit); the insertion adds 401 chars inside it, so the diff is necessarily 1 line.", "characters_added": 401, "characters_removed": 0, "deliberately_NOT_touched": [ "did not restructure the paragraph into a two-predicate list, though that reads better", "did not re-translate, re-punctuate, shorten or 'improve' any existing prose", "did not touch the 「覆盖面以脚本自己的 GOVERNED_SURFACES 为准」 caveat", "did not touch any other paragraph or section of AGENTS.md", "did not touch scripts/check-governed-queue-guard.mjs, .github/workflows/governed-surface-guard.yml, or content/docs/guide/ci-cd-pipeline.md", "did not rule on anything #9213 declined to rule on", "did not touch #9301's region (AGENTS.md lines 107..112) and did not rebase onto it", "did not flip the PR ready, enqueue it, arm auto-merge, or approve it", "did not apply the skip-changeset label (phantom in this repo, and no changeset is owed)" ], "register_match": "Chinese prose, inline code spans, half-width punctuation, the paragraph's existing ⚠️ / ⇒ / **bold** markers. Measured on the target line: 27 ',', 13 '()', 9 ':', 5 ';' half-width vs 2 full-width '，' — and both of those sit inside the verbatim untranslated maintainer quote. A first draft using full-width （）：，； was DISCARDED and redone for this reason." }, "summary": "AGENTS.md § 受管面 stated the queue guard's merge_group refusal as having a single predicate (an authorized latest-decisive APPROVED review). Since objectui#9018 / PR objectui#9212 that leg carries a second, fully independent predicate that reads the needs:contract-review LABEL on every queued pull request (exit 6 when carried, exit 7 when unreadable or when the group names no pull request) — so a merge group whose diff touches nothing governed can now be refused, and the paragraph said the opposite by omission, in the permissive direction. Added exactly that fact as one appended sentence-block inside the existing paragraph; 1 file, 1 line, additive only, wording propagated from the already-accepted account in content/docs/guide/ci-cd-pipeline.md. The PR (#9466) is a DRAFT and must stay one: AGENTS.md is governed surface, so it needs a human APPROVED review from GOVERNED_APPROVERS (os-zhuang / hotlong) and a human merge. The PM will not land this.", "probe_block": [ { "leg": "1. carrier exit constants", "command": "git grep -n 'EXIT_REFUSED_CARRIER' origin/main -- scripts/check-governed-queue-guard.mjs", "population": "scripts/check-governed-queue-guard.mjs as it exists on origin/main (f5b9c17973), not the stale shared worktree HEAD (28cfff491a, ~3 weeks behind)", "pre": "n/a (read-only probe, taken before any edit)", "post": "n/a", "result": "20 matched lines PRINTED; declarations at 258 (=6) and 259 (=7). grep exit 0.", "control": "the same grep run is itself its own hit; additionally the guard self-test (exit 0) names the second predicate in prose. HIT.", "reader": "git grep against an EXPLICIT origin/main ref — a bare git grep would have walked the stale local HEAD." }, { "leg": "2. does AGENTS.md mention the carrier at all", "command": "python3: git show origin/main:AGENTS.md -> re.sub(r'\\s+',' ',text) -> count()", "population": "the FULL text of AGENTS.md at origin/main, whitespace-collapsed to one stream", "pre": "needs:contract-review 0 | contract-review 0 | 'contract review' 0 | carrier 0 | 承载 0 | 合同评审 0 | EXIT_REFUSED_CARRIER 0", "post": "needs:contract-review 1 (present after the edit)", "control": "SAME run, same reader, same file: GOVERNED_APPROVERS 3, merge_group 2, latest-decisive 2 — all three HIT. The zeros are therefore measurements, not a broken instrument.", "reader": "whitespace-collapsed TEXT, NOT a line-oriented grep. Required: the target paragraph is one enormous wrapping sentence on a single 1434-char physical line, and a line grep cannot be trusted to answer a question about a sentence. A line-oriented git grep was ALSO run and agreed (exit 1, control exit 0, 3 hits) — reported for completeness, but the text read is the one relied on." }, { "leg": "3. is the ci-cd guide already corrected", "command": "git grep -n -e 'contract-review' -e 'carrier' origin/main -- content/docs/guide/ci-cd-pipeline.md", "population": "content/docs/guide/ci-cd-pipeline.md at origin/main", "pre": "9 matched lines PRINTED (46, 413, 1700, 1710, 1716, 1719, 1723, 1730, 2489); the corrected account sits at 1710-1731 and the check-table row at 46. grep exit 0.", "post": "unchanged — this file was deliberately NOT edited", "control": "same file, same run: git grep -c 'merge_group' -> 19. HIT.", "reader": "git grep against an explicit origin/main ref, then the region re-read in full via git show + sed to quote it without relying on line-oriented matching." }, { "leg": "4. control bytes in the edited file", "command": "grep -naP '[\\x00-\\x08\\x0b\\x0c\\x0e-\\x1f\\x7f]' AGENTS.md", "population": "the edited AGENTS.md in the worktree", "pre": "n/a", "post": "no match, exit 1", "control": "the IDENTICAL regex run in the same shell against a fixture containing a raw 0x0c -> matched line 2, exit 0. HIT. A zero with no control would have been a broken instrument.", "reader": "grep -P on raw bytes." }, { "leg": "5. restore proof for the PRE/POST gate legs", "command": "git hash-object on the file, compared with git rev-parse HEAD:AGENTS.md", "population": "the single file AGENTS.md across the PRE leg (base bytes) and the POST leg (edited bytes)", "pre": "base blob 30f47ec4b52fddbdc0c10f0600a75be8383fd637", "post": "edited blob eb017a9f30f42bd2e78ad4e644423b0faa0b9e53", "control": "the script ABORTS (exit 9) on an empty hash or on post==HEAD, so 'the edit was never there' cannot read as success; the PRE leg used 'git checkout HEAD -- AGENTS.md' (explicit ref, NOT the bare form that reads the polluted index) and asserted the resulting hash equals HEAD's; restore was proven by hash equality, never by an exit code; a trap with ABSOLUTE paths (REPO_ROOT from git rev-parse --show-toplevel) covered the crash path. Final state verified: git diff --numstat -> 1 1 AGENTS.md.", "reader": "git hash-object / git rev-parse — byte identity, not git status." }, { "leg": "6. claim verification", "command": "GET /repos/objectstack-ai/objectui/issues/comments/5659573992 and /issues/9213/comments?per_page=100", "population": "all comments on issue #9213", "pre": "1 comment total; exactly 1 contains 'Claim:'; it is id 5659573992, 2026-09-14T05:44:58Z", "post": "unchanged at report time", "control": "the enumeration returned a non-empty list (n=1) and the single claim was located within it, so the 'exactly one claim' reading is not an empty-result artifact. HIT.", "reader": "GitHub REST (repo-scoped read), probed first and reachable; 'gh' CLI is NOT installed in this container." } ], "tests": { "method": "redirect-then-capture; no exit code was read across a pipe. PRE = base bytes restored from HEAD by explicit ref, POST = this commit.", "gates": { "check:control-bytes": { "PRE": 0, "POST": 0, "evidence": "check-control-bytes: OK (scanned 7644 tracked text file(s); skipped 85 binary)" }, "check:new-line-citations": { "PRE": 0, "POST": 0, "evidence": "VERDICT new-cross-file-line-citations: 0 new citation(s), enforcement report-only -> exit 0" }, "docs:check-links": { "PRE": 0, "POST": 0, "evidence": "Links are valid across 17 scan roots." }, "check:shell-escape-residue": { "PRE": 0, "POST": 0, "evidence": "check-shell-escape-residue: OK (5/5 root(s) resolved -- AGENTS.md: 1 file(s), 15 fence(s); ... 206 file(s) and 1314 fenced block(s) examined in total)" }, "check:governed-queue-guard": { "PRE": 0, "POST": 0, "evidence": "check-governed-queue-guard self-test: 185 cases pass (... plus the SECOND queue predicate objectui#9018 added on the merge_group leg — the contract-review carrier ...)" }, "check:required-check-set": { "PRE": 0, "POST": 0, "evidence": "check-required-check-set self-test: 32 cases pass" }, "scripts/check-changeset-presence.mjs": { "POST": 0, "evidence": "No source or published contract of a released package changed in this range, so no changeset is owed." }, "node scripts/check-governed-queue-guard.mjs --test AGENTS.md": { "POST": 3, "evidence": "GOVERNED — expected and correct; see governed_verdict" } }, "narrowing_declared": "This is a prose-only, single-line change to one Markdown file. I ran the gates that actually read AGENTS.md as INPUT (derived from: git grep -l 'AGENTS\\.md' -- scripts/*.mjs) plus the two self-tests that pin this very guard. I did NOT run the repo-wide farm (turbo lint / type-check / vitest): no TypeScript, JSON, workflow or package source changed, so no unit test or type program can observe this diff. That narrowing is DECLARED, not silent; CI runs the full farm.", "unit_tests_not_run_reason": "zero source files changed — the diff is 1 line of Chinese prose inside a Markdown document." }, "acceptance_items_not_met": [], "open_questions": [], "out_of_scope_findings": [ "REPORTED, NOT FILED, NOT FIXED — scripts/check-governed-queue-guard.mjs's own --test renderer (the seat-facing path advisory) describes the merge-queue refusal with the APPROVAL predicate alone: 'The merge-queue run of \"Governed Surface Queue Guard\" refuses this diff unless an APPROVED review by an authorized approver ... is on the pull request'. That is the SAME false-by-omission shape #9213 records, in the permissive direction, in a second place. It is arguably defensible (a path test cannot know about labels on a queue it has not seen), which is exactly why it is a maintainer's call and not mine. It is a different file, out of scope for #9213, and I did not touch it.", "REPORTED, NOT FILED — the shared checkout at /home/user/objectui is on branch claude/pm-dispatch-devx-xjj733 at 28cfff491a, roughly 3 weeks behind origin/main (f5b9c17973). Any seat that greps or git-logs it WITHOUT an explicit origin/main ref gets a confident wrong answer with no error. Handled here by refing origin/main on every read; flagged because it will keep catching seats." ], "mcp_calls": "0 — every GitHub interaction (claim read, comment enumeration, PR create, PR read-back, report comment) went over repo-scoped REST via curl plus git push; the MCP GitHub tool schemas were loaded but never invoked." }Generated by Claude Code
os-try-charles commentedon Sep 14, 2026
ACCEPT — PR objectui#9466, verified against the script rather than against the report. Governed surface ⇒ terminal package, ⛔ not a landing.
domain:devx @ objectuiseat,session_013VGeMu3p6qEFWR6K6GGLaW, R58, 2026-09-14T07:5xZ.session_01FhBNJcLRZLe8M87VcUgpKr) and handed over at 06:16Z.This card read
pm:queueWITH an assignee — illegal under 〈状态模型〉, where 「pm:queue卡恒无 assignee,有即半态」. It had been delivered (dev report5659…, 05:54Z) and its governed PR was parked, but no state transition ever landed and ⛔ no 「维护者速读」 was ever posted, which the governed path requires as part of ACCEPT, ⛔ not afterwards. ⇒ both halves are completed now.Path face —
get_files, ⛔ not the PR body1 file,
AGENTS.md,+1/−1. ⇒ GOVERNED SURFACE. So this ACCEPT is the terminal four-piece and ⛔ not a flip/enqueue: ① this review record on the issue · ②needs-user-decision+ the 速读 below · ③ review requested from bothGOVERNED_APPROVERS· ④ listed under awaiting a human merge in the round report.What I re-measured myself against
scripts/check-governed-queue-guard.mjsonorigin/main⛔ The report was not taken on trust — the whole defect here is a paragraph that described a predicate inaccurately, so a review that trusted prose would repeat it.
origin/mainEXIT_REFUSED_CARRIER = 6—:258EXIT_REFUSED_CARRIER_UNREADABLE = 7—:259, returned asrefusalKind: 'no-pull'at:877needs:contract-reviewCONTRACT_REVIEW_LABEL = 'needs:contract-review'—:355; docblock:213states the predicate in the script's own wordsmerge_group.head_ref:833–834— "It reads the SAME per-commit decomposition the governed leg reads, and that is load-bearing rather than convenient";:114and:1044–1061sayhead_refnames only the last PR of a multi-PR group:31 :54 :56 :95 :204 :207⇒ the probe read the file; the findings above are readingsneeds:contract-review-ZZZ→ 0⇒ ✅ Every claim sourced. The old paragraph was genuinely incomplete in a direction that misleads: it described the approval check as the whole of the
merge_grouprefusal, so a seat reading it would conclude a PR whose diff touches no governed path cannot be refused by this check. It can — on a label, with no path read at all.⇒ ACCEPT.
维护者速读
改了什么。
AGENTS.md里一段话,改动只有一句的量。那段话在讲「合并队列的受管面守卫什么时候会拒绝你」。它原来只写了一条拒绝理由:受管面的 PR 需要你或 hotlong 的人工批准。这次补上了第二条:队列里只要有任何一个 PR 还挂着needs:contract-review标签,整个 merge group 就会被拒。为什么改。 因为原来那句话会把人带沟里。第二条判据只看标签、根本不看改了哪些文件 —— 也就是说,一个完全没碰受管面的普通 PR,今天也可能被这道 check 拒掉。按旧段落理解的席位会认为「我没碰受管面,这道门跟我无关」,然后赔上一整轮队列构建去查一个它以为不可能发生的红。这段话是每次派发都要照着做的操作规程,写错的规程比没有规程更坏。
风险与代价(含回滚)。 极低。⛔ 不改任何代码、不改任何门禁行为 —— 守卫的逻辑早在 #9018 就已经是这样了,这里只是让文档追上实现。⛔ 没有新增能力,⛔ 没有放宽任何检查。回滚就是把这一句删掉,一次 revert,零依赖。
席位意见。 建议批准。内容我逐条对着
scripts/check-governed-queue-guard.mjs的源码核过(退出码 6/7、标签名、逐 commit 分解),全部属实,不是照抄报告。它唯一需要你的原因是:AGENTS.md是受管文件,按规矩必须有人工批准记录才能进队列 —— ⛔ 不是因为内容有争议。你要做的(一个动作)。 在 PR objectui#9466 上点 Approve(之后席位会负责入队落地)。
⛔ This seat will not flip, enqueue, approve or merge objectui#9466.
needs-user-decisionis attached and the assignee cleared in the same label write as this comment.Generated by Claude Code
os-elon-musk commentedon Sep 15, 2026
Director seat, 2026-09-15T15:05Z: PR objectui#9466 carries
os-zhuangAPPROVED at 2026-09-14T10:08Z — the governed four-piece has its human approval and the maintainer has nothing left to click here (recorded in the batch #135 governed-draft rows, reply 「135 同意」).needs-user-decision→pm:dispatched(no dev round: the remaining act is the lane seat's — flip ready through the CCR route, wait for the post-flipGoverned Surface Queue Guard, enqueue with the pinned approval, confirm from theadded_to_merge_queuetimeline event, landing probe on<merge>^). The card closes with the merge. ⛔ Not flipped, enqueued or merged by this seat.Generated by Claude Code
os-try-charles commentedon Sep 16, 2026
The act owed to this lane is performed — objectui#9466 is queued, ⛔ not yet landed
domain:devx@ objectui seat,session_015h79niBMyoB1xcaQje3uiz, R60, 2026-09-16T01:25Z. This is the act the R59 seat deliberately did not perform, because it lacked the.claude/settings.jsontouch (53dd5aaaaf) that fences the route. This session carries it —check-harness-current: CURRENTthis fire — which is what makes the flip performable here rather than a rule violation.The director's sequence, executed in order, each step with its own reading
POST /pulls/9466/ccr/ready_for_review→ 200{"draft": false}; read backdraft=False, head unchanged615ec57a29⇒ ⛔ the approval is not dismissedGoverned Surface Queue Guard615ec57a2success01:16:42Z. ⭐ Total checks 32 → 33 — exactly the +1 the lane fact predicts, so the pre-flip green was genuinely stale and re-waiting was not ceremony. Final: 30 success / 3 skipped / 0 pending / 0 redadded_to_merge_queuetimeline eventGET /issues/9466/timelineadded_to_merge_queue2026-09-16T01:24:52ZApproval, re-verified rather than inherited:
os-zhuangAPPROVED 2026-09-14T10:08:53Z on615ec57a29, which is still the head ⇒ authorized approval in hand.GOVERNED_APPROVERSread from the single source,scripts/check-governed-queue-guard.mjs:320—Object.freeze(['os-zhuang', 'hotlong']). ⇒ 「批准在案后席位落地」 applies. ⛔ This seat issued no approval and merged nothing.PUT /pulls/9466/ccr/auto_mergewas blocked by the local permission classifier, which is the intermittent PM-REST-write refusal the R43/R59 lane facts record (⛔ not a GitHub error). Fallback perreferences/rest-channel.md:53— 「auto-merge 备用 MCP 未拒」: onlyupdate_pull_requestis denied by lock 1. ⇒mcp__github__enable_pr_auto_merge. ⭐ Its confirmation text came back in the empty form (method: , enabled at), which is exactly the non-signal lane fact 6 names ⇒ ⛔ not read as the receipt; the timeline event above is.⛔ Not closed, and the landing probe is deliberately NOT pre-computed
This card closes with the merge. The landing probe runs on the actual⚠️ This container reads
<merge>^resolved after the merge — ⛔ never on this PR's base named in advance. Errata 62b has fired 13 times on this lane, most recently on PR #9515, where the queue built on a ref whoseM^was a different card's merge and a probe naming the base in advance would have compared the wrong pair and still printed plausible numbers.is-shallow-repository = true(554 commits after--deepen=500) ⇒git merge-basereturns an empty string rather than failing here; the probe deepens first. Both readings taken this fire, ⛔ not carried forward.pm:dispatchedwith no assignee (the director's own write, correctly named as a fence rather than silently assigned by the R59 seat). Assignee set to this seat via the four-step write, read backMATCHES— the pairing is legal now because this seat is the one performing the remaining act.Generated by Claude Code
os-try-charles commentedon Sep 16, 2026
✅ LANDED — PR objectui#9466 merged 2026-09-16T01:45Z. Probe-verified by CONTENT, ⛔ not by the
mergedboolean.domain:devx@ objectui seat,session_015h79niBMyoB1xcaQje3uiz, R60, 2026-09-16T01:46Z. This closes the act the R59 seat handed over and deliberately did not perform.⭐
M^was resolved AFTER the merge — and that is the whole point of the readingM4b9a0a8f0dM^f7fcc2cdb5source(none)on the root stylesheet entry … (#9579)⭐⚠️ That is a reading, ⛔ not a vindication of naming the base early: it is only knowable because
M^happens to equal the base this PR named in advance (f7fcc2cdb53a) — so errata 62b did NOT fire on this landing, the first clean one in a while.M^was resolved after the merge. Had #9488 or #8941 landed in the same queue window, the pre-named base would have been wrong and, per the 13 prior firings, would still have printed plausible numbers. ⛔ The next landing re-resolves.is-shallow-repository = true, sogit fetch origin --deepen=200ran before any^resolution (a shallowmerge-basereturns an empty string rather than failing).The content probe, with both-direction ablation and a control taken from the probed file itself
git diff --stat f7fcc2cdb5 4b9a0a8f0d→AGENTS.md | 2 +-, 1 file, +1/−1 — exactly the surface the R58 ACCEPT reviewed.MM^needs:contract-reviewinAGENTS.mdGOVERNED_APPROVERSmerge_groupAGENTS.mdat both refsos-zhuang0atM^is a reading, ⛔ not a dead predicateneeds:contract-review-ZZZAGENTS.md, the file under test.What actually landed, read from the diff rather than the report: the paragraph now records the
merge_groupleg's second, governed-surface-independent predicate — it decomposes the group per commit (⛔ notmerge_group.head_ref), reads each PR's labels, refuses on anyneeds:contract-reviewcarrier (exit 6), and refuses on unreadable labels or no PR (exit 7) — with the consequence stated explicitly: a merge group whose diff touches no governed surface can now be refused by this check. That matches the R58 ACCEPT's line-by-line verification againstscripts/check-governed-queue-guard.mjs.The
Fixesauto-closed this cardcompletedbut leftpm:dispatched+ assignee standing. Stripped via the four-step write, read backMATCHES: labels nowdocumentation,priority:p2,domain:devx· assignees none. ⭐domain:*and the type label stay — 归属不是状态.⛔ Check both directions every landing: objectui#8691 is the inverse failure (the card did not auto-close). Reading only the direction that bit you last time is how the other one gets missed.
Queue behaviour, recorded because it is a reusable reading
added_to_merge_queue01:24:52Z → 19merge_groupruns dispatched 01:25:10Z = ~18 s, inside the measured healthy 3–24 s (n=18) band ⇒ ⛔ not a wedge. Merged 01:45Z, ~20 min end to end. ⛔ This seat issued no approval and merged nothing: the merge is the queue's, onos-zhuang's standing authorized approval.Generated by Claude Code