pm-dispatch SKILL: require a four-lens block on every needs-user-decision card, Blocked-by-aware batch selection, round-brief decision flags

[huangyiirene]

Fixes #7498

Implements both maintainer rulings of 2026-08-11 recorded on the card (body + the 03:11Z scope-extension comment), in one PR, inside the card's file surface (.claude/skills/pm-dispatch/SKILL.md + references/** only).

Ruling 1 — required four-lens block on every needs-user-decision card

• Step 8: after the frame's binding sentence, a new block requires every needs-user-decision card to carry a four-line analysis block — one line per lens — at step-0 filing and step-8 escalation time: ① platform long-term coherence, ② measured business pull, ③ AI-agent error-resistance, ④ startup scope discipline. The evidence line (2026-08-11 re-review reversed 4 of ~30 ledger recommendations: publicPicker.sort is read by the public lookup route but stays undeclared after #7467 — one key of the enforced-never-declarable gap remains #7485, [finding][services] datasource pool 还有两处静默丢弃:turso 整块、mongodb 的两个 timeout 键(#5714/#5931 拒绝集合覆盖不到的剩余面) #7243, objectui#4165, objectui#3804) is recorded with it.
• Step 0: the "Maintainer confirm" classification now requires the block from filing time (previously: "The label alone is the inbox entry; the deep three-axis analysis is written when the card is actually taken up").
• Step 5 dispatch clause template: the self-drive clause's wording updated to 「按裁决与三轴/四棱自裁」, matching how dispatch prompts already quote it.

How the merge respects check:skill-frame-sync. The escalation frame exists as four gate-synchronized copies across three files (.claude/skills/pm-dispatch/SKILL.md, .claude/agents/os-dev.md, and two copies in the published skills/objectstack-pm-dispatch/SKILL.md), with the axis count, order, and binding sentence compared structurally by the gate — and its AXIS_MAP lives in scripts/. Changing the frame's axis count 3→4 is therefore a four-copy + gate-map edit, all outside this card's file surface. So the merge is expressed as: one frame, whose card-face serialization is the four-lens block — ②↔axis 1 (实际业务需求), ①↔axis 2 (项目长远合理性), ③↔axis 3 (防 AI 犯错), and ④ promotes the startup-focus principle already embedded in axis 1 to its own required line. The block text states explicitly it is NOT a second frame. The gate-anchored declaring/binding sentences and the three axis entries are structurally untouched; check:skill-frame-sync stays green. If the maintainer wants the frame itself renumbered to four axes across all four copies + the published mirror, that is a follow-up card with a wider file surface (noted in the report's open_questions).

Ruling 2 — scope extension (03:11Z comment)

• Step 3 selection rule: among pm:queue candidates, a card named by any open card's Blocked-by: line now sorts ahead of plain queue items (fan-out ≥ 1), computed at selection time from the greppable Blocked-by: lines — the same reverse index the unlock sweep already reads; no stored label. Placed after the existing 扇出 ≥ 2 and target: tiers so the standing ruling-backed order is preserved and "named sorts ahead" lands against ordinary queue items.
• Step 9 round brief: the decision-inbox health metric now requires the triage seat's brief to flag decision cards with open downstream dependents (first instance: spec: rule the submitBehavior.url semantics — objectui#4190 dead-ends on it #7496 ← objectui#4190), noting it went live same-day via a standing instruction on seat post [PM seat] triage (objectstack-wide) — 🟢 Routine #6015 and this change makes it durable.
• Considered-and-declined record: the step-3 「不要为此发明新标签」 paragraph now records the 2026-08-11 rejection of a blanket priority:*/pm:unblocks label on depended-upon cards (derived decaying property — the gradient-rot shape the target:v17 ruling refused; no-reader labels banned; release-critical chains covered by target:v17 propagation), with the revisit condition (population well past today's ~6 cards).

Line ratchet

SKILL.md 3027 → 3043 lines (ceiling 3050, headroom 7). Additions paid for by extracting four step-8 narrative passages verbatim to references/incidents.md (append-only, each with an 出处 line per that file's convention): the 第三档 #5373 example, the 两条元判据 precedent narratives (both bullets), and the business-need axis precedent block (#5021/#4988/#4834, #5021 vs #4936). Main-file rules keep a one-line anchor + pointer in each spot.

Same-day churn guard

#7492 (Operational note 1 merge-queue membership criterion) and #7493 (发版板 queries / cloud contradiction) are open undecided cards on this file — this diff touches neither Operational note 1 nor the 发版板 section nor step 9's release-blockers metric; edits are orthogonal to both questions.

Gates run locally (all green)

• pnpm check:pm-skill-ratchet — 3043/3050, self-test 5 cases pass
• pnpm check:skill-frame-sync — 4 copies structurally isomorphic, 3 axes, 4 count mentions agree, self-test 12 cases pass
• pnpm check:doc-authoring — 375 files clean
• pnpm --filter @objectstack/lint run check:doc-formula-expressions — green after building the dependency closure (pnpm --filter '@objectstack/lint^...' build)
• pnpm check:nul-bytes — 7000 files clean

Changeset

None — internal agent tooling (.claude/** only), docs-only for the agent protocol; releases nothing. skip-changeset applies (label applied separately; the PM applies/confirms at acceptance).

---

Generated by Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment

Project	Deployment	Actions	Updated (UTC)
objectstack	Ignored		Aug 11, 2026 5:11am

[os-help]

Closing on the maintainer's direct instruction — provenance, per the execution-record discipline:

• Who: the maintainer, in the skills seat's 专题讨论 (this seat's chat session session_01ATJiqhFNL82YwV8udPkoqX, 2026-08-11).
• Verbatim: asked how to handle this in-flight PR, the maintainer chose 「关闭 pm-dispatch SKILL: require a four-lens block on every needs-user-decision card, Blocked-by-aware batch selection, round-brief decision flags #7552,内容并入批次」, followed by 「包括 7498」 — i.e. fold this PR's content and card pm-dispatch SKILL: require a four-lens analysis block on every needs-user-decision card #7498 into the consolidated skills batch ([batch] pm-dispatch SKILL consolidated amendments + compression — eight ruled cards, two serialized PRs (skills seat, round 1) #7631), so the whole SKILL amendment set lands as one PR with one human merge.
• Not a rejection of the content: this PR was reviewed, ACCEPTed and fully green; its diff (branch claude/issue-7498-decision-four-lens, head 24934cf8b) becomes the reference implementation for the pm-dispatch SKILL: require a four-lens analysis block on every needs-user-decision card #7498 member of batch [batch] pm-dispatch SKILL consolidated amendments + compression — eight ruled cards, two serialized PRs (skills seat, round 1) #7631 — the batch dev reuses it against post-batch anchors rather than re-deriving.
• pm-dispatch SKILL: require a four-lens analysis block on every needs-user-decision card #7498 returns to the queue as a batch member (Fixes #7498 rides the batch PR).

---

Generated by Claude Code