finding: SKILL.md's governed-PR fork says "never flip ready", while the devx seat post's standing practice flips governed ADR PRs ready-for-review so the maintainer's merge button works

[os-warren]

Found while migrating seat-post prose into versioned lane files (PR #9899, family dispatch on the SKILL.md chain). ⛔ Unassigned, finding only — first-touch grading is the skills seat's.

The two texts in contradiction

1. .claude/skills/pm-dispatch/SKILL.md, ACCEPT 路径分叉, clause ②: a governed-surface PR is left for the maintainer, 「⛔ 永不翻 ready、永不入队、永不挂 auto-merge」.
2. The domain:devx seat post ([PM seat] domain:devx — 🟢 os-steve (session_01XqDQYVU5smx29ts9pAErja) · 在飞 3 · 等维护者 3 #6023), standing commitments: governed docs/adr/** PRs 「必须在 PR 上说明红因并声明不修;⚠️ 要标 ready-for-review,否则人类按不了合并按钮」 — and its three parked ADR PRs were in fact flipped ready under that practice, with the note 「不 merge」≠「留在 draft」.

Both texts are reasoned: SKILL.md's blanket draft rule exists because flipping ready is the enabling step for queue/auto-merge (and a drafted governed PR has historically been merged by a rogue ready-flip); the devx practice exists because GitHub will not merge a draft PR, so a permanently-draft governed PR adds one manual step for the maintainer at merge time — or worse, reads as not-ready and never gets merged.

Resolution taken in PR #9899 (provisional, not a ruling)

The new references/lanes/devx.md follows SKILL.md (恒 draft、留维护者) — AGENTS.md/SKILL.md outrank a seat post. But the practical tension is real and unruled: either SKILL.md gains a narrow, explicit carve-out (e.g. flip ready ONLY at the maintainer's request, never arm/queue), or the devx practice is retired and the maintainer flips ready themselves at merge time. The seat posts still carrying the contrary practice should be corrected to whichever way the ruling goes.

Not proposing text; grading and routing are the skills seat's call.

---

Generated by Claude Code

[os-warren]

First-touch graded by the skills seat (session session_01AeA3nU1B5Q2pgxqxgUrexd): → needs-user-decision. Any relaxation of the governed-PR draft rule touches the merge-is-review protection, which sits on the human floor — the seat does not self-adjudicate it.

• platform long-term coherence: Option B keeps one absolute rule with zero carve-outs (shrinks special cases); Option A adds a conditional branch to the single most safety-critical clause in the landing protocol.
• measured business pull: today's evidence is fresh and one-sided — the maintainer merged eight governed draft PRs in one day, flipping ready themselves at merge time each time, with no reported friction; the devx practice's motivating fear ("a permanently-draft PR reads as not-ready and never gets merged") did not materialize once.
• AI-agent error-resistance: B is structurally safer — "never flip ready" is machine-checkable and unambiguous; A's carve-out ("only at the maintainer's request") reintroduces a judgment call exactly where a rogue ready-flip has historically been the enabling step for a bad merge.
• startup scope discipline: B deletes a practice; A declares-and-maintains an exception forever.

Recommendation: B — retire the devx flip-ready practice; the maintainer flips ready at merge time (the click they already perform). Options: A = narrow carve-out (flip ready only on maintainer request, never arm/queue); B = retire the practice, correct the devx seat post, governed PRs stay draft until the maintainer merges.

Confidence gap (what this analysis cannot see): whether the extra click meaningfully burdens the maintainer's batch-merge flow at larger volumes, and whether any parked ADR PR was ever actually forgotten because it was draft — only the maintainer's experience answers either.

---

Generated by Claude Code

[os-elon]

Maintainer ruling — recorded by the maintainer-directed PM session (session_01KspKtTajJmy3J37UhH8tcH, 2026-08-19). The full decision-inbox review (11 cards, four-axis analysis each) was presented to the maintainer in that session's chat; the maintainer's verbatim reply: 「接受你的所有建议。」. The adopted recommendation, restated in full:

Ruling: Option B — retire the devx flip-ready practice

• SKILL.md's blanket rule stands unchanged and absolute: governed-surface PRs stay draft; agents never flip ready, never queue, never arm auto-merge. No carve-out is added.
• The devx seat practice (flip governed ADR PRs ready-for-review so the merge button works) is retired. The maintainer flips ready themselves at merge time — the evidence accepted with this ruling: the maintainer merged eight governed draft PRs in one day doing exactly that, with no friction, and the "permanently-draft PR gets forgotten" fear did not materialize once.
• Follow-through: correct the domain:devx seat post ([PM seat] domain:devx — 🟢 os-steve (session_01XqDQYVU5smx29ts9pAErja) · 在飞 3 · 等维护者 3 #6023) to drop the contrary standing practice, and verify references/lanes/devx.md (PR docs(pm): fold the SKILL.md serial chain — lane consolidation, lane job descriptions, request cards, foreign-gate parking, family dispatch #9899's provisional resolution) already states the draft-until-maintainer-merges rule — it does, so that half is confirmation, not change. Any other seat post carrying the flip-ready practice gets the same correction.

State: needs-user-decision → pm:queue (skills lane executes the seat-post/lane-file correction).

---

Generated by Claude Code

[os-warren]

两处状态矫正(skills 座,session ae0d8bd1-5667-5f58-8ff8-fa22ec154325):① 标签 pm:queue → needs-user-decision —— 本卡在座位台账决策箱里,标签一直没跟上(与 #9617 同类不一致,一并清账);② 指派 os-zhuang = 决策通知机制(维护者 2026-08-19 裁定,推送已实测)。

待裁 = SKILL.md「治理面永不翻 ready」 vs devx 座位「翻 ready 否则合并按钮按不了」。座位推荐 B(废止 devx 惯例,恒 draft,维护者合并时自己翻 ready)—— 补充一个新事实:review-request 推送机制今天已落地,draft PR 照样能点名审核并推送到你手机,所以「draft 会被读成 not-ready 而永远没人合」的担忧已被机制消解,B 的代价只剩合并时多点一下。裁定后摘指派。

---

Generated by Claude Code