Machine-enforce human review on governed-surface paths — so the no-bypass rule stops resting on seat discipline alone

[os-zhuang]

Provenance: maintainer ruling 2026-08-27 (PM chat, decision-inbox batch 1, adjudication session session_01DKWDdUJ2XNRESVVWUvcpnh), verbatim: 「同意，并继续」 — accepting Option A + C on #6325. Option C is this card: the un-reviewed AGENTS.md merge stands (audit record on #6325), and the governed-surface no-bypass rule becomes machine-enforced so the next incident does not depend on seat discipline.

Why seat discipline is not enough (measured incident)

PR #6183 (governed surface: AGENTS.md) was correctly draft-parked; a GitHub MCP update_pull_request call passing only reviewers silently set draft: false; the PR entered the merge queue and merged as 5b3290fd5 with no human approval. Converting back to draft did not dequeue it. Platform fact: objectstack-ai/objectstack#12200. The mechanism — hidden tool side effect + irreversible queue — will recur.

Scope

Make human review mechanically required for this repo's governed surface (2026-08-18 definition): AGENTS.md, CLAUDE.md, .claude/**, skills/**, docs/adr/**.

Candidate shapes — the implementer measures which is viable, does not assume:

• .github/CODEOWNERS entries scoped to exactly the governed paths, plus the branch-protection "require review from Code Owners" setting. The settings half is a GitHub-UI change only the maintainer can make — if this shape is chosen, the PR half lands the CODEOWNERS file and the card then moves to pm:awaiting-maintainer naming the exact toggle.
• A required status check that fails any PR touching governed paths unless an explicit human approval review is present.

Constraints:

• The door must not block the normal dev-PR merge queue for PRs that do not touch governed paths.
• The enforcement change itself touches governed/config surface — it lands as a draft PR for human merge, per the same rule it enforces.
• Named reader: the objectui execution seat that implements it; central triage routes the domain label.
• Sibling repos carry the same gap; mirroring is a separate triage/maintainer call once this repo's shape is proven — not part of this card.

Refs: #6325 (incident decision + audit record), PR #6183 (timeline), objectstack-ai/objectstack#12200 (tool fact).

[os-litant]

Triage (central triage seat, daily round 2026-08-27, session session_01Czs5tCwMymUvzB9QCA2yhH): routed domain:skills, type Task. Rationale — lane-table boundary rule applied cross-repo: a gate whose SUBJECT is the governed surface itself (.github/CODEOWNERS routing half + governed-path human-review enforcement) belongs to the skills/governance lane; devx owns gates about code/docs quality, and this is neither. The governed-surface definition (2026-08-18) applies to objectui identically. Standing constraints restated from the card: the enforcement PR itself is governed/config surface ⇒ draft PR, human merge; if the CODEOWNERS shape is chosen, the branch-protection toggle is a maintainer-only UI action and the card then moves to pm:awaiting-maintainer naming the exact setting.

---

Generated by Claude Code

[os-litant]

Claim: skills 席派发认领(定级 HIGH——维护者 2026-08-27 裁定 Option C 的执行卡;objectstack 侧同类 guard 已于 08-28 上线运转,有现成参照实现)。
Session: session_01MnijPVVDakqK2J335JoJtq · Branch: claude/issue-6596-governed-queue-guard · Worktree: ../objectui-6596-guard(dev 自建)
File surface: scripts/check-governed-queue-guard.mjs(自 objectstack scripts/pm/check-governed-merges.mjs + queue-guard 族移植适配)+ merge_group workflow + 必需检查接线;不动 skills/**/AGENTS.md 正文。⚠️ 分支保护/必需检查的 GitHub 设置半边只有维护者能拨——若选此形,PR 落 CODEOWNERS/检查文件后卡转 pm:awaiting-maintainer 并点名开关。
Container & model: os-dev 子代理;objectui 无 tier 声明面(objectstack 的 3 declared globs 不跨仓)⇒ PM 判:opus 执行(纯 scripts/workflow 代码)。
Clause-② limb: 本卡就是给治理面加机器门——按裁定本 PR 自身走 draft + 人合。
Serial: 无同文件竞争。落地路径:代码文件 PM 自审,但按卡内约束整 PR 仍 draft 交人合。

---

Generated by Claude Code

[os-litant]

os-dev-report

{
  "issue": 6596,
  "status": "done",
  "branch": "claude/issue-6596-governed-queue-guard",
  "worktree": "/home/user/objectui-6596-guard",
  "pr": "https://github.com/objectstack-ai/objectui/pull/6739",
  "pr_number": 6739,
  "draft": true,
  "premise_still_valid": true,
  "shape_chosen": "2 — required status check (merge_group refusal + pull_request early warning). MEASURED against shape 1: both have their enforcement half in branch protection, so the tiebreak was what each lands as repo files today. CODEOWNERS lands an INERT file (no signal, no self-test, and no way to bind a clearance to a sha); the check lands a workflow that runs from the moment it merges, carries 132 self-test cases, and pins each approval to the PR's CURRENT head. The last point is decisive for THIS incident: #6183's mechanism was a PR whose state changed under a review nobody had given, and a clearance not bound to specific bytes cannot see that class at all. Secondary: objectui already runs a merge_group gate family with a written-down required-context register, so shape 2 slots into existing machinery.",
  "summary": "Landed the objectui half of the governed-surface no-bypass rule as a new gate + workflow. The governed set is the card's 2026-08-18 definition verbatim (AGENTS.md, CLAUDE.md, .claude/**, skills/**, docs/adr/**), pinned in BOTH directions by the self-test and again against the real tree by the vitest pin. Split by event: merge_group REFUSES a governed diff with no authorized APPROVED review pinned to the PR's current head (exit 3 unapproved / 4 unreadable / 5 unattributed); pull_request is an early warning that always exits 0, because a governed PR parked as a draft is the healthy end state and a check red on the healthy case is a permanently red check. Path test runs before any request is constructed, so an ordinary diff costs ZERO API calls, measured with a spy that throws. types: names ready_for_review (not in GitHub's default set) because flipping a governed draft to ready is the first move of the #6183 sequence. Also ships a seat-side --test PATH mode so a refusal hands the reader a runnable command. TWO WIDENINGS DELIBERATELY NOT TAKEN and pinned: .github/workflows/** is not governed (this guard does not govern its own workflow), and the two root rows are EXACT so examples/AGENTS.md stays ordinary source. NOTE: run against its own 7 files the new gate reports NOT GOVERNED (exit 0) — the PR is draft because the CARD requires it, not because the guard demands it of itself. One line flagged for the maintainer: GOVERNED_APPROVERS (os-zhuang, hotlong) is INHERITED from the sibling's 2026-08-27 roster ruling, not ruled for objectui; objectui's card rules the mechanism. Nothing is blocked either way — the remedy printed FIRST is back-to-draft plus human merge, which needs no approver.",
  "files": [
    "scripts/check-governed-queue-guard.mjs (new)",
    ".github/workflows/governed-surface-guard.yml (new)",
    "scripts/__tests__/check-governed-queue-guard.test.ts (new)",
    "scripts/dependabot-merge-gate.mjs (REQUIRED_CONTEXTS registration)",
    "scripts/__tests__/merge-queue-reporting.test.ts (MUST_SUBSCRIBE_MERGE_GROUP reason row)",
    "content/docs/guide/ci-cd-pipeline.md (inventory row + section — both mechanically required by ci-cd-pipeline-doc.test.ts)",
    "package.json (check:governed-queue-guard, governed)"
  ],
  "parity_decision": "objectui-NATIVE, not registered in scripts/upstream-port-pin.json. Two measurements: (1) upstream splits the mechanism over check-governed-queue-guard.mjs (1,381 lines) + check-governed-merges.mjs (2,614 lines), and the register half is mostly a multi-repo post-merge audit plus a provenance-recompute engine for GENERATED artifacts inside governed paths — objectui has ZERO applicable rows (.claude/workflows/ absent; skills/ carries no generator output: no references/_index.md, no react-blocks contract), so a port would carry ~1,500 lines that can never fire. (2) The pin CANNOT express the deletion that implies: validatePin in scripts/check-upstream-port-parity.mjs was driven directly with a deletion divergence and returned the single problem 'files[0].divergences[0].ported is empty' (exit 2, pin unusable). So a pinned port is STRUCTURALLY impossible here, not merely undesirable. The divergence is declared in the file header prose instead, together with the obligation the pin would have carried (a sibling predicate change is a hand re-read, not an auto re-sync). upstream-port-pin.json untouched; check:upstream-port-parity still green on its two pinned files.",
  "required_toggle_for_maintainer": "Settings -> Rules -> Rulesets -> the `main` ruleset -> 'Require status checks to pass' -> add the context named EXACTLY: Governed Surface Queue Guard. It must ALSO be added to the MERGE QUEUE's required set on the same ruleset — that is the set the merge_group build is judged by, and the refusal lives only on that leg; adding it to the PR-side required set alone changes nothing because the PR leg is green by design. Until the toggle is flipped the queue leg REPORTS without stopping anything. The repo-side half is already landed: the context is registered in REQUIRED_CONTEXTS (scripts/dependabot-merge-gate.mjs), which is what merge-queue-reporting.test.ts derives the merge_group subscription floor from. Suggest moving the card to pm:awaiting-maintainer naming this toggle.",
  "gates": {
    "run_at_commit": "c4885f3 (the commit the PR carries; union re-run after the final commit)",
    "method": "redirect-then-capture, never piped to head/tail before capturing the status; heavy runs serialised through scripts/pm/os-verify-lock.sh",
    "check-governed-queue-guard.mjs --self-test": "exit 0 — 'OK check-governed-queue-guard self-test: 132 cases pass'",
    "npx vitest run scripts/__tests__ (under the shared lock)": "exit 0 — 'Test Files  86 passed (86)' / 'Tests  2404 passed (2404)'; lock VERDICT command-exit 0, held 55s, waited 0s",
    "pnpm type-check:scripts": "exit 0 — and NOT vacuous: tsc --listFiles confirms scripts/__tests__/check-governed-queue-guard.test.ts is in the program (1 hit)",
    "pnpm lint:root (under the shared lock)": "exit 0 — '28 problems (0 errors, 28 warnings)', all pre-existing; the CI Lint gate ratchets ERRORS. eslint.config.js has no parserOptions.project / projectService, so no type-aware linting: this diff cannot move any untouched file's verdict",
    "check-lint-coverage.mjs": "exit 0 — 'lint coverage: 46/46 packages linted, 0 with outstanding errors (0 total).'",
    "check-control-bytes.mjs": "exit 0 — 'check-control-bytes: OK (scanned 5596 tracked text file(s); skipped 85 binary).'",
    "check-entry-guard.mjs (+ --self-test)": "exit 0 / exit 0 — '52 scripts/ file(s) — no entry guard outside the baseline; 0 file(s) still hand-type one'; self-test 63 cases pass",
    "check-upstream-port-parity.mjs (+ --self-test)": "exit 0 / exit 0 — '2 ported file(s) match objectstack-ai/objectstack@bf10debd5 modulo their declared divergences.'; self-test 37 cases pass",
    "check-pre-install-import-graph.mjs": "exit 0 — '20 pre-install step(s) in 17 job(s) run 19 scripts/ gate(s); 21 module(s) walked, every non-relative leaf a node builtin.' (the new workflow's pre-install self-test step is inside that count)",
    "check-changeset-presence.mjs": "exit 0 — '7 file(s) changed, 0 of them published source of a package the release covers' / 'No source of a released package changed in this range, so no changeset is owed.'",
    "check-doc-links.mjs": "exit 0 — 'Links are valid across 17 scan roots.'",
    "check-skills-paths.mjs": "exit 0 — 'check-skills-paths: OK (93/94 stated path(s) resolve across 18 guide file(s); 1 baselined).'",
    "check-shell-escape-residue.mjs": "exit 0 — 'check-shell-escape-residue: OK (4/4 root(s) resolved ...)'",
    "check-doc-fence-languages.mjs": "exit 0 — 'every TypeScript block in 223 document(s) is fenced ts/tsx/typescript ...'",
    "check-doc-component-types.mjs": "exit 0 — 'Every documented component type is registered.'",
    "one_red_found_and_fixed_by_me": "pnpm type-check:scripts exit 2, TS2339 x2: the pin test read `surface.prefix ?? surface.exact` over a union of frozen literals, and tsconfig.scripts.json infers those types from the .mjs (allowJs, checkJs off). Rewritten as an `in` narrowing with a comment saying why. Re-run green at c4885f3.",
    "not_run_and_why": "pnpm lint (turbo run lint) and pnpm test (full vitest) — this diff touches ZERO package source, and check-lint-coverage.mjs independently reports 46/46 packages linted with 0 outstanding errors. The affected family is scripts/__tests__ (all 86 files run, not narrowed) plus the doc gates the doc edit reaches. Declared narrowing, per the dispatch's rule; CI runs the full farm either way."
  },
  "changeset": "none — and measured, not recalled: check-changeset-presence.mjs reports 0 of 7 files are published source of a covered package, so none is owed. Mirrors precedent #6672 (parity-gate PR: scripts + workflows + package.json + tests, no changeset). NO skip-changeset label applied — it is a phantom in objectui and ci-cd-pipeline-doc.test.ts pins that it must never be wired.",
  "mcp_calls": "7 — issue_read get, issue_read get_comments, create_pull_request, pull_request_read x2 (body read-back, twice: the first found sanitizer damage), update_pull_request (body fix, with draft:true passed EXPLICITLY — the exact defence against the #6183 mechanism this card is about), add_issue_comment (this one). Bulk reads went through git and the local checkouts; one repo-scoped REST probe returned 403 (session GitHub access not enabled), so MCP was the only write channel.",
  "stopped": false,
  "open_questions": [
    {
      "question": "Is GOVERNED_APPROVERS = [os-zhuang, hotlong] the right roster for objectui, or should it differ from the sibling's?",
      "options": [
        "A - keep the inherited pair (shipped): same maintainer, same governed surface, one-line edit if wrong",
        "B - maintainer names an objectui-specific roster",
        "C - drop the approval path entirely and make back-to-draft + human merge the ONLY remedy"
      ],
      "recommendation": "A as shipped, flagged in the PR body for confirmation. It blocks nothing: the remedy the refusal prints FIRST is back-to-draft plus a human merge, which needs no approver at all. C is worth a thought later — it would remove the shared-identity attack surface completely — but it also removes the re-validation a queue merge gives you, and no ruling covers that trade."
    },
    {
      "question": "Does a FAILING non-required check actually block an objectui merge-queue build?",
      "options": [
        "A - it does not: GitHub judges the queue by the ruleset's required set only, so the toggle is load-bearing",
        "B - it does"
      ],
      "recommendation": "A, and the code is written on that assumption — I could not measure it without the settings, which is precisely why the toggle is called out as the remaining half rather than described as a nice-to-have. Until it is flipped, this guard REPORTS on a queue build; that is stated in the workflow, the script header, the docs page and the PR body rather than left to be discovered."
    }
  ],
  "out_of_scope_findings": [],
  "notes": "Sibling-repo mirroring is explicitly not this card and was not attempted. The shared checkout /home/user/objectui was never edited; all work in the dedicated worktree, now cleaned up. No CI wait: report delivered at draft-PR time per the 2026-08-10 ruling."
}

Generated by Claude Code

---

Generated by Claude Code

[os-litant]

判决:ACCEPT。PM 收件复审:治理面五路径与 2026-08-18 定义逐字一致(两根文件 exact、三树 prefix);形状 ② 系实测选出(CODEOWNERS 形落的是惰性文件、无法钉 sha;pin 移植对删除类分歧结构上不可表达——validatePin 亲测拒收),守卫 objectui 原生实现;workflow 纪律到位:⛔ 无 paths 过滤(skip=绿的坑)、ready_for_review 入 types(正是 ui#6183 事故的第一步)、fetch-depth 0、fail-closed、零插值、read-only 权限;132 自测 + 2404 vitest + 门全绿。开放题裁定:审批人名册取 A(承袭 os-zhuang/hotlong——同一维护者同一治理面,错了是一行改动;维护者审 PR 时可否);题 2 按 A 假设正确(队列只判 required 集,开关承重)。
PR #6739 保持 draft、双推 os-zhuang + hotlong(卡自身约束:本 PR 走人合)。合并后本卡不关:转 pm:awaiting-maintainer,待维护者在 main ruleset 的必需检查与 merge queue required 集里加 context Governed Surface Queue Guard(两处都要加;只加 PR 侧无效)。

---

Generated by Claude Code