[finding] The decision-card frame has no "governing text" step — two director recommendations on 2026-09-05 contradicted rulings already codified in a repo's AGENTS.md and in a spec docblock, because the existing-ruling search is scoped to card comments and the director ledger

Filed by the director seat (#12708, summon #15, session_01TezFG8ZMrNH6n5VTNpPpdH) for the domain:skills lane's own triage, as the maintainer asked 「问题出在哪里，是 skills 的问题吗」 after overturning two recommendations in one batch. domain:* is not applied here — skills-lane findings are self-triaged by that seat. Class: 原则错/缺.

The two misses, measured

card	seat's recommendation	what already decided it	where that text lives
hotcrm#1543	E — give the hotcrm field-consumer guard a synthetic fixture (kept the guard in hotcrm)	hotcrm's 2026-08-31 ruling: "Lint, validation, gates and diagnostics belong to the platform, uniformly … Tests in this repo pin this repo's own business facts and nothing else … ⛔ Do not grow a gate farm."	`hotcrm/AGENTS.md` § "Scope — a pure metadata application", rule 3 — a governed file, not a card comment
objectstack#15828	C — widen `AnalyticsQuerySchema.where` to accept `FilterArray` (analytics only)	#5158 ruling C: "`where` on a query is a `FilterCondition` and stays one … rather than widening the wire contract (rejected option A)"	the `FilterArray` docblock in `packages/spec/src/data/filter.zod.ts` — a spec docblock citing the ruling, not a card comment

In both cases the filing seat had also missed the text (hotcrm#1543 offered options A–D that all keep the guard; #15828's triage routed to domain:spec on option A), so the director inherited an option space already framed against the standing rule and reasoned inside it. The maintainer overturned both from memory of the rulings.

What the skill text requires today, and the gap

SKILL.md 〈升级与决策〉: 鲜度门 = re-read comments newer than the body; 出决策箱须引一个同趟取回过的裁决 id; the director's own body discipline (summon feat: Comprehensive CRM example demonstrating all ObjectStack protocol features #14) says 「呈报前先在评论区找已有维护者裁决」. All of these search card comments and the director ledger ([PM seat] director(项目总监) — ⏳ vacant · 第 47 场(os-bill)已收班 · 第 43 场续场 5(os-zhuang)、第 44 场(os-project-manager)与第 46 场(huangyiirene)无收班简报 · 台账在评论 #12708).
references/decision-analysis.md 落卡分析模板: 「带 re-check 命令的前提」 — premises are the card's facts (what is on the tree), not the governing rule for the option space.
The 四轴 frame itself worked: once the governing text was read, both recommendations flipped without the maintainer supplying new facts. The failure is retrieval, not weighing.
Rulings are codified into three other carriers the frame never names: a repo's AGENTS.md sections (hotcrm's scope rulings, the objectstack Prime Directives), ADRs, and spec docblocks that quote a ruling by number (#5158's ruling C …). None of the three is a comment, so a comment-scoped search cannot find them, and search_issues cannot see them either.

Proposed change (for the skills seat to shape; a suggestion, not a ruling)

One mandatory line in both the decision-card template and the director's per-card presentation: Governing text: — the AGENTS.md section / ADR / spec docblock / lint rule that already constrains this option space, quoted, with the git grep that found it; or none found with the searches run (git grep -n -iE '<subject terms>' origin/main -- AGENTS.md docs/adr packages/spec/src in each repo the option space touches). A card whose options contradict the governing text is not a decision card — it is a pm:queue card executing the text, or a card asking the maintainer to change the text. Filing seats supply the line; the director re-runs the grep before presenting.

Cost: one grep per card per repo. It would have caught both of today's misses at filing time (hotcrm#1543 → git grep -n 'gate farm' origin/main -- AGENTS.md; #15828 → git grep -n 'FilterArray' origin/main -- packages/spec/src/data/filter.zod.ts).

Not claimed

That the frame is otherwise sound is asserted from two cases only; the skills seat may want the summon #14 batch ledger (#25–#49, 25 batches) sampled for the same failure shape before sizing the change.

Refs: hotcrm#1543 (ruling F, 5552508432) · #15828 (ruling recorded this summon) · #12708 (director ledger) · objectstack#13848 (hotcrm 2026-08-31 ruling ledger) · #5158.

Addendum — a second rule the skills text must carry, ruled by the maintainer on the #15828 case (director seat, summon #15, 2026-09-05T15:2xZ)

Maintainer, live director chat, verbatim: 「objectstack#15828 决裁的问题skills还需要加一点，本项目以协议为基准。所以开发应该对其协议，协议有问题应该立卡修改协议」.

Reading, for the skills seat to encode (the maintainer's words are the rule; this is the seat's paraphrase of what the frame lacks):

The protocol (packages/spec) is the baseline. When a card measures "the spec declares X, the code does Y", the default direction is code → spec: the consumer or producer is corrected to what the protocol declares. That is not one option among several — it is the frame the options are judged in.
A protocol change is its own card, never a side option in a defect card. If the protocol itself is judged wrong, the remedy is a card whose subject is changing the protocol (spec/ADR card, contract-review tier, maintainer floor), filed and ruled on its own. A bug card that lists "widen the spec to admit what the code does" as option A alongside "fix the code" as option B has already mis-framed the decision — runtime: POST /analytics/query refuses the array where the objectui adapter now sends for every array-form filter — AnalyticsQueryRequestSchema.where is FilterConditionSchema, while lowerAnalyticsWhere (the gate ui#6302 measured) accepts filter AST #15828 did exactly that, and the director's first recommendation followed it.
Combined with the first finding above (governing text): the presenter's first two lines on any spec-vs-code card are what the protocol declares (quoted, with the file) and whether the card asks to change it. If it does, it is re-filed as a protocol card; if it does not, the option space is how the code conforms.

Precedents this makes explicit: #5158 ruling C (the FilterArray docblock: "rather than widening the wire contract (rejected option A)"); the hotcrm AGENTS.md 2026-08-31 rule 2 ("A platform defect means you WAIT for the platform fix — do not route around it"); today's #15828 ruling (frontend defect; the protocol was right).

The seat's earlier suggestion of a Governing text: line stands; this addendum adds the direction rule that line feeds. Both are the skills seat's to shape and land.

Generated by Claude Code

Claim: PM loop round 7 — the maintainer's 插队 (15:1xZ, verbatim 「#15929 插队」): the decision-card frame gains a mandatory Governing-text step — the AGENTS.md section / ADR / spec docblock / lint rule that already constrains an option space, quoted with the git grep that found it, or none found with the searches run; a card whose options contradict it is not a decision card. Rules only: the two misses stay on this card, not in the skill files.
Session: session_019RfFHiRCSs3JXLK4cwcfox
Branch: claude/issue-15929-governing-text-step
Worktree: objectstack-issue-15929
Domain: domain:skills, graded p1 on the maintainer's word (finding removed at grading).
File surface: .claude/skills/pm-dispatch/SKILL.md (升级与决策, one rule line, line-neutral at 811/811), references/decision-analysis.md (the 落卡分析模板 skeleton gains the item, line-neutral at 50/50), references/lanes/director.md (the per-card presentation re-runs the grep; at most +1 inside its headroom of 5).
Container & model: S, mode:subagent, contract tier (fable) — dispatch-gates.mjs --tier: SKILL.md hits clause ① (MANDATORY).
Clause-②: no (protocol prose; no contract accept/reject change).
Serial: SKILL.md is also touched by PR #15908 (#15894, at the governed terminal awaiting a human) at lines 93–97 / 351–359 / 592–620 / 755–760; this flight's SKILL.md hunk stays inside 723–732 and must prove a conflict-free git merge-tree --write-tree against that PR's head before opening its PR — the maintainer's 插队 takes precedence over the same-file wait, and the proof stands in for it. GOVERNED (.claude/skills/**) ⇒ draft PR, in-seat review at contract tier, os-zhuang + hotlong, human merge, skip-changeset; at ACCEPT the seat adds needs-user-decision + the 维护者速读.

Generated by Claude Code

os-dev-report

{
  "issue": 15929,
  "status": "done",
  "branch": "claude/issue-15929-governing-text-step",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/15946",
  "premise_still_valid": true,
  "summary": "Two rules now sit in the decision-card frame, written once per carrier in its register: Rule A (Governing text) — SKILL.md 升级与决策 gains one line (落卡与呈报必带 `Governing text:` with the four carriers enumerated; a card whose options contradict it is not a decision card) and the 出决策箱 line now cites the ruling id AND the item; decision-analysis.md's 段落骨架 gains `Governing text(引文 + grep,或 none found + 检索式)` and `协议声明、是否改协议`; director.md 职责二 gains the per-card, per-repo `git grep -n -iE 'TERMS' origin/main -- AGENTS.md docs/adr packages/spec/src` re-run before presenting. Rule B (协议为基准) — SKILL.md gains one line (spec 与代码不一致默认改代码对齐;改协议单独立卡,⛔ 不作缺陷卡的选项) and director.md gains the presenter's first-two-lines rule for spec-vs-code cards (协议原文带文件、是否改协议;改 ⇒ 重立协议卡,否则只议对齐). Line-neutral everywhere by in-place compression: SKILL.md 811/811 (folds 711+712 → 117 B, 727+728 → 116 B), decision-analysis.md 50/50 (the 8-segment skeleton cannot fit one 120-byte line — measured 125–156 B — so it spans two lines and the pay-fold is 适用面 8+9 → 113 B, since no 正文六项 pair fits 120 B without dropping a clause), director.md 72/72 on main's rewritten file (folds 61+62 → 101 B with nothing dropped, 69+70 → 110 B dropping the fillers 的工作/的勤务). Widest line written 119 B (the SKILL.md Governing-text line); the four-axis block 733–751 is byte-identical at the same line numbers; core-rules.md untouched as ruled. Per the seat update, origin/main (a2051fa4e9, carrying PR #15908 at 97afcb5cc and the #15641 rules-only lane rewrite) was merged into the branch as merge commit 3ba4752b31 (SKILL.md auto-merged; director.md conflicted on the rewrite and was resolved on main's content by re-applying the two lines plus the two folds, because the rewrite moved that file's ceiling from 77 to 72 — the brief's +2 headroom no longer existed). Draft PR #15946 opened with `Fixes #15929` first, a `## 维护者速读(草稿)` section (五段,席位意见留空), `skip-changeset` added via the additive labels endpoint with a comparison read-back (read-back == union(pre, target)); body read back byte-identical on create and, after the footer-less PATCH, equal to what was sent plus exactly one platform-appended bare footer. Platform observation: the PR creation at ~15:40Z produced no GitHub Actions runs for ~20 min (only four queued third-party suites; a sibling PR #15949 created at 15:48Z got runs immediately); the 16:01Z push of the merge commit fired the full set. No model identifier in commits, body, comment or files; no assignee/label/ready/queue action taken beyond the ruled skip-changeset add.",
  "tests": "On the final pushed head 3ba4752b31 (also on 2a149b3165 before the merge): `node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack` derived 16 commands (3 paths vs merge base a2051fa4e; `--repo` asserted and holds) — all 16 run with exit captured before any pipe, all exit 0, `--ran` reconcile prints `✓ dispatch-gates --ran: 16 derived famil(ies) accounted for — 16 run, 0 NOT-MEASURED.`; on the pre-merge run `pnpm --filter @objectstack/lint run check:doc-formula-expressions` first exited 3 `PREREQUISITE NOT MET` (formula/lint unbuilt) — read as NOT MEASURED, built via `os-verify-lock.sh -c 'pnpm exec turbo run build --filter=@objectstack/formula --filter=@objectstack/lint'` (VERDICT command-exit 0, 181s held), re-run: `✓ check:doc-formula-expressions self-test: 58 cases passed` + three sweeps clean. `node scripts/pm/check-skill-line-ratchet.mjs --self-test` → `✓ 155 cases pass`; live before: SKILL.md 811/811, decision-analysis.md 50/50, director.md 72/77; live after the merge (script as now on main): `SKILL.md is 811 lines (ceiling 811; headroom 0)`, `decision-analysis.md is 50 lines (ceiling 50; headroom 0)`, `director.md is 72 lines (ceiling 72; headroom 0)`, widest table row pin 342 unchanged. `node scripts/pm/check-skill-id-lint.mjs` → `✓ 24 file(s) clean`. `node scripts/check-skill-frame-sync.mjs` → `✓ 2 copies of the decision frame are structurally isomorphic across 2 files`. `node scripts/pm/check-governed-prose.mjs` → `✓ 2 instruction surface(s) name all 5 registered governed surfaces`. Whole-repo `OS_VERIFY_LOCK_SLOT=issue-15929 bash scripts/pm/os-verify-lock.sh -c 'pnpm lint'` → `VERDICT command-exit 0 · held the lock 91s · waited 0s` on 3ba4752b31 (86s on 2a149b3165; shared-box seconds). Control-character scan over the three files: none. Merge proofs: pre-merge `git merge-tree --write-tree --name-only origin/claude/issue-15894-maintainer-brief-on-governed-prs HEAD` on 2a149b3165 → exit 0, tree 3f8ea1a66d, no conflicting path; post-merge `git merge-tree --write-tree --name-only origin/main HEAD` on 3ba4752b31 → exit 0, tree 3fc609b6cd, no conflicting path. CI on 3ba4752b31, polled in the foreground until every run completed (16:18Z): 28 check runs — 17 success, 9 skipped (paths-gated matrix jobs: Build Core, Build Docs, Check Changeset, Console Pin Gate, Dogfood shards, Temporal Conformance, Test Core shards, Packed-tarball smoke), 2 cancelled concurrency duplicates whose same-named twins are success; required contexts: Lint & Repo Gates success, TypeScript Type Check success, Test Core success, Dogfood Regression Gate success, Build Core skipped, Temporal Conformance skipped. Governed Surface Queue Guard success. No ablation applies (prose-only diff).",
  "mcp_calls": "3 — issue_read get, issue_read get_comments, create_pull_request; everything else (probe, labels + read-back, body PATCH + read-back, check-run polling, report comment) went over repo-scoped REST, which this seat measured reachable (HTTP 200; core 15000/h)",
  "open_questions": [
    {
      "question": "director.md: the brief allowed +2 inside a 77-line ceiling, but the #15641 rules-only rewrite that landed on main during this flight moved the file to 72 lines with a 72 ceiling, so the two presenter lines were paid by two in-place folds in the rewritten file (61+62 → 101 B, nothing dropped; 69+70 → 110 B, dropping only 的工作/的勤务). Is that the intended resolution?",
      "options": ["A keep the folds — line-neutral, every line ≤120 B, no rule lost", "B drop the two director.md lines and carry the presenter duty only through SKILL.md + the template", "C raise the ceiling (maintainer floor, quoted ruling required)"],
      "recommendation": "A, because the ratchet discipline names compressing in place as the payment for a protocol change, the folds lose no rule, and Rule A's director-side duty (re-run the grep before presenting) is the half that catches the second miss class."
    },
    {
      "question": "decision-analysis.md: the ruled fold location (among the 正文六项) was measured infeasible under 120 B without dropping a clause (27–33 are 108/108/82/103/83/93/88 B); the pay-fold went to 适用面 8+9 instead. Accept the deviation?",
      "options": ["A accept — same line count, zero content loss", "B fold 30+31 instead, dropping the 论据是客户、产品或事故后果 examples or the 响亮拒绝/静默泄露 pair"],
      "recommendation": "A, because it keeps every 正文 clause and the skeleton reads as one paragraph."
    }
  ],
  "out_of_scope_findings": [
    "not filed (reported per the brief, for the skills seat): `references/core-rules.md` 升级与决策 digest (its 落卡/出决策箱 lines mirror SKILL.md) now owes one line for the Governing-text item and the protocol baseline — file or fold as the seat prefers",
    "not filed (platform observation, for platform-readings if reproduced): a draft PR opened via MCP at ~15:40Z on an already-pushed head produced no GitHub Actions runs for ~20 min (only four queued third-party check suites), while a sibling PR opened at 15:48Z got runs at once; the next push (synchronize) fired the full set — poll the head after a push, never read an empty check-runs list on a fresh PR as \"nothing runs here\""
  ]
}

Generated by Claude Code

os-closed-card-sweep — machine-findable marker for this generated comment.

Removed the pm-loop state label(s) this closed card no longer claims: pm:dispatched.

Closing pull request: docs(pm-dispatch): governing-text step and protocol-baseline rule in the decision frame #15946, merged.
Closing commit 31403453dd, merged into main.
Left untouched: priority:p1, domain:skills — ownership, priority and outcome are not state claims.
The label set was read back after the write and matched.

A state label claims work is in flight. This card is closed on a merged delivery, so the claim
is stale; every other label is left exactly as it was found. Nothing here is a judgement about
the card, and no verdict-bearing label is ever touched by this sweep.

posted by half-state-patrol run 34005012908 · trigger schedule

Generated by Claude Code

Navigation Menu

Repository navigation

[finding] The decision-card frame has no "governing text" step — two director recommendations on 2026-09-05 contradicted rulings already codified in a repo's AGENTS.md and in a spec docblock, because the existing-ruling search is scoped to card comments and the director ledger #15929

The two misses, measured

What the skill text requires today, and the gap

Proposed change (for the skills seat to shape; a suggestion, not a ruling)

Not claimed

Addendum — a second rule the skills text must carry, ruled by the maintainer on the #15828 case (director seat, summon #15, 2026-09-05T15:2xZ)

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

[finding] The decision-card frame has no "governing text" step — two director recommendations on 2026-09-05 contradicted rulings already codified in a repo's AGENTS.md and in a spec docblock, because the existing-ruling search is scoped to card comments and the director ledger #15929

Description

The two misses, measured

What the skill text requires today, and the gap

Proposed change (for the skills seat to shape; a suggestion, not a ruling)

Not claimed

Activity

os-zhuang commented on Sep 5, 2026

Addendum — a second rule the skills text must carry, ruled by the maintainer on the #15828 case (director seat, summon #15, 2026-09-05T15:2xZ)

claude commented on Sep 5, 2026

claude commented on Sep 5, 2026

github-actions commented on Sep 6, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions