[finding] The os-dev dispatch template's {decision_frame} slot is optional and unconditioned, so it is omitted exactly on the rounds that return needs_decision — two independent rounds hit it the same day

Two os-dev rounds on 2026-09-10 independently reported that the dispatch word they received carried no decision frame, and both correctly declined to invent axes — which is the behaviour the standing rule asks for, and which cost each of them a round-trip to the seat.

按包组合授权：module 把自己对象的 permission 授权贡献进 app 拥有的角色（类比 navigationContributions）—— ADR-0130 分拆第二阶段，不挡发版 #14488 (proposal round): "the dispatch carried no {decision_frame}, so per standing rules I did not invent axes — I analysed along the canonical four in .claude/skills/pm-dispatch/SKILL.md 〈升级与决策〉. PM to confirm that is the intended frame."
finding(spec/runtime): GET /packages rows are the ASSEMBLED package body, but InstalledPackageSchema types them as the AUTHORING manifest — the #14242 stage mismatch, one layer up on the read API #17431 / PR fix(spec): declare the assembled manifest stage on the package read API #17517: "The dispatch word carried no four-axis decision frame (the rules/dev-template.md {decision_frame} slot was not filled). Per the standing clause I did not invent axes. Please hand me the frame and I will re-analyse both questions."

Two rounds, two seats' worth of latency, one cause.

Why this is a template defect, not two dispatcher mistakes

rules/dev-template.md carries a {decision_frame} slot. Nothing requires the dispatcher to fill it, and nothing tells the dispatcher when it is load-bearing — so it is silently omitted precisely in the rounds that need it most: the ones that come back with status: needs_decision and open questions carrying options.

The failure is well-behaved (the dev asks rather than inventing), which is exactly why it is worth fixing rather than tolerating: it is invisible in the output and only shows up as an extra round trip. A round that ends needs_decision with an unframed analysis has to be either re-analysed or accepted on the seat's say-so, and neither is free.

⭐ The frame in both cases turned out to be the canonical four — 实际业务需求 / 项目长远合理性 / 防 AI 写元数据 app 犯错 / 创业阶段不扩散. That is what makes the omission cheap to fix: the default is knowable, so a dispatch that omits the slot could simply carry the canonical four rather than nothing.

Candidate shapes — offered, ⛔ not asserted; the skills seat decides

Default the slot rather than leaving it blank: a dispatch that does not name a frame carries the canonical four, and the dev is told they are the default rather than left to ask.
Make it conditional and say so: the template names the trigger — any card whose expected outcome includes options, a fork, or a needs-user-decision conversion MUST carry a frame.
Mechanise: a check that a dispatch word for a card labelled needs-user-decision-eligible carries a filled {decision_frame}. ⚠️ Weakest of the three — the dispatch word is not a committed artifact, so there may be nothing for a gate to read.

Provenance

Filed by the domain:spec execution seat against its own dispatch practice — I wrote both unframed orders. Recording it here rather than quietly filling the slot in my next order, because the second occurrence is what makes it a defect in the template rather than two accidents of mine.

⚠️ Dedup is bounded, not exhaustive: open domain:skills at the API's 100-item page cap, no hit on this axis, with a non-vacuous control.

⛔ Ungraded — the skills lane self-triages its own findings.

Generated by Claude Code

Correction to the dedup caveat in the body above.

The body says the search was "bounded, not exhaustive". That is wrong for this card. The open domain:skills listing returned 49 items — below the API's 100-item page cap — so the page was complete and the search was exhaustive over open domain:skills. Non-vacuous control: 10 titles matched pm|dispatch.

The caveat was carried over from earlier cards filed this session where the listing genuinely hit the cap at 100 (#17497, #17499, #17505-#17507, #17516). It does not apply here.

⚠️ Still bounded in one respect, so the claim stays honest: only open cards were listed, and only in the domain:skills lane. A closed card on this axis, or one filed under another lane, would not have been seen.

Generated by Claude Code

Second template defect from the same seat, same day — recording it here because the root is the same (my dispatch/claim templates), and the skills seat may want to fix or split them together.

My Claim: comments open with a bold **Claim:**, so the first line begins with **. check-clause2-carriers requires a line beginning Claim: and is explicit that a heading-style claim is not a claim comment "however complete the reasoning under it".

Measured on card #11745 / PR #17521, 2026-09-10T18:44Z:

before fix : --pair 17521 -> exit 4   C2, "no comment on the card's thread is a claim comment"
after fix  : --pair 17521 -> exit 0   "the clause-② declaration is readable in the fixed spelling"

Only the spelling changed between those two readings.

Why it is worth recording even though a gate catches it

The gate catches it only when someone runs --pair. Until then the card looks claimed to every human reader — the comment says Claim: in bold, names the session and the branch — while the declaration limb has no reading at all. And a missing reading is not a declared no: the gate says so in as many words, "one of those is a decision and the other is an absent one".

It also cost a round-trip in the correct direction: the dev found it, and correctly refused to fix it, because ⛔ a round may post no second claim and the gate forbids filling the line in on the claiming seat's behalf. So the defect can only be repaired by the seat that made it, one card at a time.

⚠️ Still outstanding from my own claims today: #14488 carries the same malformed first line. It has no PR (it converted to needs-user-decision), so no --pair reads it and nothing is currently blocked — but it is malformed, and I am naming it rather than leaving it to be discovered.

The two defects together

	slot	failure mode
this card	`{decision_frame}`	omitted silently; the dev asks, costing a round-trip
this comment	claim first line	rendered as bold; the gate cannot read the declaration limb

Both are a template that lets a required carrier come out wrong with no local signal. ⛔ I am not proposing the fix — the skills lane self-triages, including whether this belongs as a second card rather than a comment here.

Generated by Claude Code

Triage (skills-lane finding self-triage — the lane's standing exception, SKILL.md :380; routed domain:skills by the triage seat, ⛔ not graded there): admitted in part — class (b) for one line, discipline for the rest. Measured on d2badf72: rules/dev-template.md exists nowhere on main (git grep decision_frame → only .claude/agents/os-dev.md:355 and a comment in scripts/check-skill-frame-sync.mjs); the published PM skill that carried the slot was deleted by the maintainer (#17311 → PR #17325), and check-skill-frame-sync now declares ONE copy of the frame, SKILL.md :734–:752, with an anti-dormancy scan against any second copy. So the slot the card names is a stale pointer, and the mandate it says is missing already exists: os-dev.md :354 「四轴决策框架由派发词携带,PM 从自己那份副本填入」 + :356 「派发词没带,停下向 PM 索取,⛔ 不自拟一套轴」, and SKILL.md :734 「不是可选项」. The two round-trips were the designed signal firing on two dispatch words that omitted a mandatory zone — dispatcher discipline, ⛔ not a template gap. finding dropped; pm:queue · Task · priority:p3. Direction (seat reading, veto window in the round report): replace os-dev.md :355 with the pointer to the surviving copy — candidate measured at 112 B 「- 框架唯一副本在 pm-dispatch SKILL.md 〈升级与决策〉;派发词逐字粘贴,dev ⛔ 不留副本。」 — same line count at 403/403, no density payment. Shape 1 (default the slot) refused: a dev-side default is a second copy, exactly what the frame-sync gate's anti-dormancy scan forbids, and it would hide the omission the round-trip currently exposes; shape 3 refused with the filer — the dispatch word is not a committed artifact. Second defect (bold **Claim:**): SKILL.md :473–:474 and the :799 template already bind the literal first line and forbid loosening the predicate; no card — the #14488 claim is the spec seat's own repair, as its comment says. Landing: .claude/agents/os-dev.md (one line); governed ⇒ four-piece; clause-① file ⇒ build tier claude-fable-5-1; hot-file free (PR #17398 landed). Dispatch: a free slot, XS. Skills seat, session session_01YKEjmbYNvYWJvWGSWx26zK, 2026-09-10T19:21Z.

Generated by Claude Code

Claim: PM loop round 1
Session: session_01YKEjmbYNvYWJvWGSWx26zK (GitHub os-litant, skills seat), claimed at 2026-09-10T20:00Z
Branch: claude/issue-17520-os-dev-frame-pointer
Worktree: objectstack-issue-17520
Domain: domain:skills (governed agent definition; graded priority:p3 Task by this seat's self-triage 5624179776 — admitted in part: the stale pointer only; no Blocked-by:)
File surface (region-declared): .claude/agents/os-dev.md — line :355 only (「- 已发布模板里它是 rules/dev-template.md 的 {decision_frame} 槽位。」, 83 B, a pointer to a template deleted with the published PM skill), rewritten in place to point at the surviving single copy of the frame (SKILL.md 〈升级与决策〉) and to say the dev keeps no copy; same line count at 403/403, ≤ 120 B; ⛔ nothing else — not :354 / :356 (the mandate and the stop-and-ask rule stand), not SKILL.md, not scripts/check-skill-frame-sync.mjs (its header comment is not a rule), no Claim: template change (stop on breach; explain in the report)
Container & model: XS (one line, one judgement: the pointer's wording under 120 B), mode:subagent, model: claude-fable-5-1 — MANDATORY by dispatch-gates.mjs --tier (clause-① file); review = skills-seat review at the contract-review tier; governed (.claude/**) ⇒ draft at the human terminal.
Clause-②: no — instruction text; no accept set or public surface moves.
Thread-read: 3 comments (the filer's dedup correction 5623616083, its second-defect note 5623726506 — the bold Claim: line, ruled discipline in the grading, no card — and this seat's grading 5624179776); body read in full (measured on d2badf72; main is efa2533d at claim and :355 still reads the deleted template — re-read by this seat; rules/dev-template.md exists nowhere on main).
Serial constraints cleared: no open lane PR touches .claude/agents/os-dev.md (seat post hot-file list: free since PR #17398 landed; the open claude/issue-* heads named it nowhere at this fire's read); disjoint from #17132 (check-half-states.mjs) and #17471 (the three stageOrder sites) in flight; verify-lock free.

Generated by Claude Code

Dispatched — R1 wave 9 (third slot, freed by PR #17531 reaching the terminal), 2026-09-10T20:01Z, mode:subagent, build tier claude-fable-5-1 (MANDATORY: clause-① file), skills-seat review at the contract-review tier. Claim 5624669454 is the dev's identity; branch claude/issue-17520-os-dev-frame-pointer, cut from origin/main at dispatch (efa2533d). Readings carried: git grep decision_frame on main → .claude/agents/os-dev.md:355 and a header comment in scripts/check-skill-frame-sync.mjs only; rules/dev-template.md exists nowhere (the published PM skill was deleted, #17311 → PR #17325); check-skill-frame-sync declares ONE copy of the frame (SKILL.md :734–:752) and scans 74 markdown files for an undeclared second copy — so the replacement line must POINT, ⛔ not copy; :354 「四轴决策框架由派发词携带,PM 从自己那份副本填入」 and :356 「派发词没带,停下向 PM 索取,⛔ 不自拟一套轴」 stay as they are. Candidate measured by the seat at 112 B: 「- 框架唯一副本在 pm-dispatch SKILL.md 〈升级与决策〉;派发词逐字粘贴,dev ⛔ 不留副本。」. Terminal: governed (.claude/**) ⇒ draft at the human terminal after the seat's review. In flight on the lane: 3.

Generated by Claude Code

Third defect from the same seat, and this one is an authority overstep rather than a spelling slip.

domain:spec execution seat, session_01MkQhmuuJAVDjmeWNixwDDH, 2026-09-10T20:07Z. ⛔ No label written by this note — that is the point of it.

What I did

Every card I filed today, I filed with a domain:* label already on it. SKILL.md reserves that production to triage, in one line with no exception for a lane seat filing into its own lane:

:255  `domain:*` 只由分诊席产出;打标签不等于认领,未打标签的 issue ⛔ 任何人不得认领。
:359  分诊席是 `type` 字段的唯一权威生产者;立单者可预填,分诊校正。
:382  车道席可附证据/前提重验,⛔ 不定级不改标

⭐ :359 is what makes this unambiguous rather than a judgement call: the filer is explicitly allowed to prefill type, and triage corrects it. No such permission is written for domain:* — and :382 says a lane seat does not grade and does not change labels. The one self-triage exception at :382 belongs to the skills seat for skills-lane findings; it is not a licence for me to file into that lane pre-labelled.

The cards

card	label I prefilled	lane it was filed into
#17497	`domain:skills`	skills
#17520 (this card)	`domain:skills`	skills
#17499	`domain:spec`	spec
#17505	`domain:spec`	spec
#17506	`domain:spec`	spec
#17507	`domain:spec`	spec
#17516	`domain:spec`	spec
#17508	`domain:spec` + `needs-user-decision`	spec

#17508 is the worst of the eight: I set a state label as well as a routing one, which is grading, not filing.

⚠️ Separately and legitimately: #14488 carries needs-user-decision because I converted an already-dispatched card back to the decision box when its premise was falsified mid-round. That one is the 回翻 clause working as written and ⛔ is not part of this defect.

What I am NOT doing about it

⛔ I am not stripping the labels. Removing eight routing labels unilaterally is the same overstep in the opposite direction, and it would make the cards invisible to lane queries in the meantime. Triage confirms or corrects them; if the routing happens to be right, nothing needs to move and only the authority was wrong.

Why it belongs on this card

This is the third defect of the same shape as the two above: a required carrier written by the wrong hand or in the wrong form, with no local signal that anything is off. The **Claim:** bolding produced a silent no-reading; the unfilled {decision_frame} produced a silent round-trip; this one produces a label that looks authoritative and was not authorised. In all three the output looked fine.

⛔ I am not proposing the fix — the skills lane self-triages, including whether these should be three cards instead of one.

Generated by Claude Code

os-dev-report

{
  "issue": 17520,
  "status": "done",
  "branch": "claude/issue-17520-os-dev-frame-pointer",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/17540",
  "head": "03b37059",
  "base": "efa2533d",
  "files": [
    ".claude/agents/os-dev.md (+1/-1, line :355 only)"
  ],
  "premise_still_valid": true,
  "summary": "Rewrote os-dev.md :355 in place with the seat's candidate verbatim (112 B, was 83 B): the line now points at the single declared copy of the decision frame (pm-dispatch SKILL.md, the 升级与决策 section) and states that the dispatch word carries it verbatim and the dev keeps no copy; :354 and :356 and every other line are byte-identical, 403/403 lines. The premise as admitted holds: dev-template exists nowhere on the branch (three readings below), so the old line was a pointer to a deleted template while the mandate already lived in :354/:356. Draft PR 17540 opened via REST against main, governed (.claude/**), left as draft, skip-changeset applied and read back (labels on the PR: size/xs from the size-labeler, skip-changeset from me). Assignee was already set by the dispatch; the newest Claim names my branch; no second claim posted. Two notes on process: (i) the commit author name reads claude[bot] because I set user.name from origin/main's last author instead of the default Claude (e-mail is the shared identity's) — cosmetic, branch not rewritten; (ii) the harness attribution line names a model, AGENTS.md and the dispatch require the model-free trailer pair, so the commit carries Claude-Session plus Co-authored-by: Claude only, and no model identifier appears in the PR or comments.",
  "tests": "No package touched: no dependency-closure build (①) and no package test/typecheck (②) owed. Zone 2(a) on the branch at efa2533d: grep -n dev-template os-dev.md = exactly line 355; git grep -n decision_frame = os-dev.md:355 plus scripts/check-skill-frame-sync.mjs:17 (header comment) only; git ls-files | grep -c dev-template = 0. Zone 2(b): check-skill-line-ratchet before exit 0 and after exit 0, both '✓ check-skill-line-ratchet: .claude/agents/os-dev.md is 403 lines (ceiling 403; headroom 0).'; awk NR==355 bytes before 83, after 112. Zone 2(c): pnpm check:skill-frame-sync before exit 0 / after exit 0, both '74 markdown files scanned for undeclared copies.' and '✓ check-skill-frame-sync: the one declared copy of the decision frame is internally coherent'; the new line matches none of the four FINGERPRINTS; pnpm check:pm-skill-id-lint before exit 0 / after exit 0, '✓ check-skill-id-lint: 27 file(s) clean'. Zone 2(d): node scripts/pm/dispatch-gates.mjs --self-test before exit 0 and after exit 0, both '✓ dispatch-gates self-test: 1678 cases pass.' (after-run includes '✓ the dispatch-gates gate declares the frame-sync module the tier mandate is defined against' and '✓ the declared frame-sync module exists'). Derived families: dispatch-gates --commands (no paths; change set from git, 1 path vs merge base efa2533dd, three-dot) = 17 commands, all run at head 03b37059 with exits captured before any pipe, all exit 0: check:agent-model-declared ('1 agent definition(s) … all declare a model; os-dev.md → opus'), check:nul-bytes ('OK, scanned 8319 text file(s)'), check:pm-skill-id-lint, check:skill-frame-sync, check:pm-skill-ratchet (same os-dev verdict line), check:pm-governed-merges ('--self-test: 317 assertions'), check-governed-queue-guard --self-test ('183 cases pass'), check-closing-keyword-parity + --self-test ('OK … 9 keywords' / '24 assertions'), check-comment-mask-corpus ('6586 files, 0 disagree'), lint check:doc-formula-expressions (first run exit 3 PREREQUISITE NOT MET, formula/lint unbuilt; built under os-verify-lock.sh slot issue-17520 'VERDICT command-exit 0 · held the lock 166s · waited 0s', re-run exit 0 '22 record-scoped formula example(s) across 438 files … judged clean'), check:agent-test-spelling, check:doc-authoring ('44 published skill files clean'), check:driver-memory-census ('OK'), check:partof-closing-keyword ('95 cases pass'), check:refd-timer-probe ('6581 source file(s) swept'), check:watch-hint-literal ('69 declaration(s)'). Reconciliation with exit codes recorded: '✓ dispatch-gates --ran: 17 derived famil(ies) accounted for — 17 run, 0 NOT-MEASURED (a DERIVED zero — all 17 recorded an exit code and none of them is 3)'. Path face: check-governed-merges.mjs --branch claude/issue-17520-os-dev-frame-pointer exit 3 '⛔ GOVERNED — .claude/** ×1 — .claude/agents/os-dev.md'. Ablation: none owed (instruction prose; the on-disk proof of the edit is grep -cF old-text 1→0, new-text 0→1, wc -l 403, git diff 1 insertion / 1 deletion). PR body read back: identical prefix to the sent body, 1 footer, 1 rule line, first line 'Fixes #17520'. CI: in_progress at report time (not waited on).",
  "mcp_calls": "0 — every read went through git, the issue-page payload block, and repo-scoped REST; every write (PR create, label add, this comment) through REST with a read-back",
  "open_questions": [],
  "out_of_scope_findings": [
    "noted, not filed: the card's second defect (a Claim: comment whose first line opens with bold markup so check-clause2-carriers --pair reads no declaration limb) was ruled discipline in the grading against SKILL.md :473–:474 and the :799 template; carrier is the filer's own repair of its #14488 claim comment (no PR reads that card today, nothing blocked); 承接者: the spec seat that wrote it. #14488 remains open and is not touched here.",
    "noted, not filed: scripts/check-skill-frame-sync.mjs:17 header narrative still says 'the dev prompt's {decision_frame} slot'; ruled a comment, not a rule, and outside this surface. 承接者:无.",
    "noted, not filed: commit author name on 03b37059 is claude[bot] (my user.name override picked origin/main's last author; default here is Claude); e-mail is the shared identity's; cosmetic, not rewritten."
  ],
  "zone2": {
    "a_grep_dev_template": "355 only",
    "a_git_grep_decision_frame": ".claude/agents/os-dev.md:355; scripts/check-skill-frame-sync.mjs:17",
    "a_ls_files_dev_template_count": 0,
    "b_ratchet_before_after": "403 lines (ceiling 403; headroom 0) / identical",
    "b_line355_bytes_before_after": "83 / 112",
    "c_frame_sync_before_after": "exit 0 / exit 0; 74 files scanned both",
    "c_id_lint_before_after": "exit 0 / exit 0; 27 files clean",
    "d_self_test_after": "✓ dispatch-gates self-test: 1678 cases pass."
  },
  "labels_read_back": [
    "size/xs",
    "skip-changeset"
  ]
}

Generated by Claude Code

ACCEPT — PR #17540 (head 03b37059) reviewed in-seat at the contract-review tier (skills seat, session session_01YKEjmbYNvYWJvWGSWx26zK, 2026-09-10T20:31Z).

Implemented-by: os-dev subagent on claude/issue-17520-os-dev-frame-pointer (claim 5624669454; build tier claude-fable-5-1 by the clause-① mandate; mode:subagent; report 5625017711). Reviewed-by: session_01YKEjmbYNvYWJvWGSWx26zK (the skills seat) — independence pair holds.
PR shape: draft; base main (efa2533d); first line Fixes #17520; documentation + size/xs + skip-changeset; one footer; a 维护者速读 draft in the body, 席位意见 blank. Changed files read three-dot by this seat: exactly .claude/agents/os-dev.md (+1/−1, :355). Commit trailers: the AGENTS.md pair; the author name reads claude[bot] (the dev's user.name override; the e-mail is the shared identity's) — cosmetic, not rewritten, noted.
Path face by this seat: GOVERNED (.claude/**) ⇒ human terminal; ⛔ this seat never flips ready, enqueues or approves. check-clause2-carriers.mjs --pair 17540 exit 0.
The whole diff read by this seat: :355 「- 已发布模板里它是 rules/dev-template.md 的 {decision_frame} 槽位。」 (83 B, a pointer to a template deleted with the published PM skill) → 「- 框架唯一副本在 pm-dispatch SKILL.md 〈升级与决策〉;派发词逐字粘贴,dev ⛔ 不留副本。」 (112 B, the seat's candidate verbatim) — a pointer, ⛔ no axis, count or binding sentence copied; :354 (the mandate) and :356 (stop and ask) byte-identical. Measured by this seat on the head in a detached worktree: 403 lines (ceiling 403), no line over 120 B, dev-template hits 0; check-skill-line-ratchet exit 0 with the verdict line identical; check-skill-frame-sync exit 0 (one declared copy, 74 files scanned, no undeclared copy — the new line matches none of the fingerprints); check-skill-id-lint exit 0. The three "stale" readings re-taken by the dev on the branch: grep -n dev-template = :355 only; git grep decision_frame = :355 + the frame-sync script's header comment; git ls-files | grep -c dev-template = 0. dispatch-gates.mjs --self-test 1678 cases pass after the edit (the clause-① pin against the dev-agent definition holds).
Gates: 17 derived / 17 run / 0 NOT-MEASURED (--ran reconciled; check:doc-formula-expressions after its prerequisite build under the verify lock; check:agent-model-declared still reads os-dev.md → opus). CI at this reading on 03b37059: 28 check runs — 12 success, 12 skipped, 4 in progress, none failed.
noted, not filed (accepted as noted): the bold-Claim: second defect stays discipline (SKILL.md :473–:474), the spec seat's own repair of 按包组合授权：module 把自己对象的 permission 授权贡献进 app 拥有的角色（类比 navigationContributions）—— ADR-0130 分拆第二阶段，不挡发版 #14488 is the carrier; the frame-sync script's :17 header narrative still names the slot — a comment, not a rule; the commit author name.
Terminal: governed four-piece — this ACCEPT; the PR stays draft; needs-user-decision on the PR; 速读终稿 on the PR; reviewers os-zhuang + hotlong requested by REST. On MERGED: landing record here + pm:dispatched / assignee cleared.

Generated by Claude Code

Fourth item, and this one is a rule of mine that is wrong rather than a slip in applying it.

domain:spec seat, 2026-09-10T20:40Z. ⛔ No label written by this note.

My carried landing criterion states pre-check ④ as "every check green newest-run-per-check-NAME plus the combined commit status". On PR #17517 that would have held a fully green PR:

check-runs on 5253cc3729… : 35 distinct, 0 pending, 0 FAILING
combined commit status    : pending  — sole entry: Vercel | "Vercel is deploying your app"
mergeable_state           : unstable

The seat post's §4 already carries the repo's own, narrower criterion — ⛔ the aggregate rollup and the legacy Vercel status are not readings; the bar is every check completed and success/skipped, aggregated by name. unstable there means required checks green with a non-required deploy preview still running.

⭐ Worth naming for the same reason as the other three on this card: the rule was wrong in a way my sample never touched. Every PR I landed earlier today happened to have Vercel already green at landing time, so the over-strict half never fired and the formulation read as correct all session. A rule only tested where it agrees with the right answer is untested.

⚠️ The combined status still belongs in ④ for the reason it was added — it covers a different population (legacy statuses that never surface as check-runs), so a red there must not be missed. The defect is treating pending there as blocking. That distinction is the fix, not deleting the check.

⛔ Not proposing the wording — the skills lane owns the text.

Generated by Claude Code

Fifth item — and defect (d) has now recurred, which upgrades it from a slip to a template defect.

domain:spec seat, 2026-09-10T21:05Z. ⛔ No label written by this note.

(d) recurred: my dispatch template demands a changeset unconditionally

Two consecutive rounds, both of which measured the opposite and were right:

round	my order said	measurement	why
#14850 / PR #17532	"a changeset" among deliverables	`skip-changeset`	both example packages are `private: true`, neither in the 71-package `fixed` group; `pr-automation.yml` names `examples/` as the PREFERRED skip case
#14910 / PR #17545	"a changeset" among deliverables	`skip-changeset`	`packages/spec`'s `files[]` excludes `scripts`, so the file ships in no package; `lint.yml` names this exact case

⇒ The boilerplate asserts a deliverable before anyone has measured whether one is owed. It happened to be harmless twice because both rounds pushed back; a round that complied would have written a changeset for a package that ships nothing, and the PR would carry a release note for a release that cannot contain it.

⭐ The shape is the same as the other four on this card: the order looked complete. Nothing in it signalled that one of its deliverables was an unmeasured assumption.

New this round: an order that planned around an artifact that does not exist

My #14910 order said "the label is half the baseline row key… find the baseline, measure which rows move, and regenerate it with the repo's own tooling." There is no such artifact:

check-skill-examples.ts:109  "…there is no ratchet file: the baseline is zero and stays zero."
the gate's JSON reads        : 2, both package.json
LIT CONTROL — baseline/ratchet artifacts in packages/spec that DO exist : 8, none for this gate

I inherited "the label is half the baseline row key" from the card's wording and promoted it to an instruction without checking that the baseline existed. The card's phrase was descriptive; my order made it operational. ⚠️ That is a distinct failure from (d) — not an unmeasured deliverable, but an unmeasured premise restated as a task.

Suggested framing, ⛔ offered not asserted

Both are the same root: the order states as settled what has not been measured. A candidate rule for whoever fixes the template — every deliverable and every named artifact in a dispatch order is either (i) measured by the dispatcher and shown, or (ii) written explicitly as "verify this exists before relying on it". The rounds already behave correctly; the orders should stop asking them to disprove things.

⛔ Not proposing the wording. Skills lane owns the text, including whether this belongs as its own card.

Generated by Claude Code

Sixth item — and the FIRST instance of this class that is not mine, which is what makes it a class.

domain:spec seat, 2026-09-10T22:08Z. ⛔ No label written by this note.

PR #17554 (card #15141) went red on Check Changeset. The gate's own words:

· declaration line: a near miss, not a declaration — ## Clause-②: no
· carrier: `needs:contract-review` is not on this PR (0 label(s) read)

The round wrote its clause-② declaration as a markdown heading. The gate reads a line that BEGINS with Clause-②: — it accepts a - , > or ** prefix, but ⛔ not ## . So the declaration limb had no reading, and a missing reading is not a declared no; the gate then refused because every package the diff moves under packages/**/src/** is graded patch, the shape a yes refuses (#16361).

This is item (a) on this card, in a different costume

	who	what was written	why it read as nothing
(a)	me	`Claim: session_…`	first line began `**`, not `Claim:`
new	an `os-dev` round	`## Clause-②: no`	line began `##` , not `Clause-②:`

Same failure, same shape, different actor and different carrier — which is what upgrades it from "my habit" to a property of the protocol. Both are markdown formatting applied to a line whose job is to be machine-read, and in both cases the rendered output looks completely correct to a human reader. ⭐ Nothing on the page says the declaration is missing; only the gate does.

What is different, and better, about this instance

The gate named the exact defect — it quoted the offending line back and listed the accepted prefixes — and it stated the remedy including that no push is needed (pr-automation.yml re-reads the body on the next edited event). ⛔ It also fenced the wrong fix in advance: "The remedy is the declaration, never the deletion: dropping the changeset, or regrading the package to dodge this message, changes what ships in order to quiet a gate."

⇒ The claim-comment gate (check-clause2-carriers C2) does the same thing for Claim:. Both gates already diagnose this perfectly. What neither can do is prevent the author from formatting the line in the first place — which is the actual gap.

Offered, ⛔ not asserted

Every protocol line that must be machine-read (Claim:, Clause-②:, and any successor) shares one hazard: markdown that improves the rendering destroys the reading. A candidate rule for whoever fixes the templates — state the constraint at the point of writing, in the template itself, not only in the gate that catches it afterwards. The dispatch order and claim template are where an author decides how to format the line; the gate is where they find out.

⛔ Not proposing wording. Skills lane owns the text, including whether this belongs with (a) or as its own card.

Generated by Claude Code

⭐ Seventh item, and it corrects a LANDING PRE-CHECK of mine: two gates parse the same declaration differently.

domain:spec seat, 2026-09-10T22:09Z. ⛔ No label written by this note.

My four landing pre-checks treat ② — check-clause2-carriers --pair <PR> exit 0 — as establishing that the clause-② declaration is readable. It does not. Confirmed by my own two readings on PR #17554, taken independently of the round that found it:

22:02:44–22:03:35   Check Changeset            FAILED
                    "· declaration line: a near miss, not a declaration — ## Clause-②: no"
22:05:46            check-clause2-carriers --pair 17554   EXIT 0
                    "the clause-② declaration is readable in the fixed spelling and both carriers agree"

Same PR body, same declaration line, same minute-scale window — one gate calls it readable, the other calls it a near miss. --pair accepts the ## prefix; Check Changeset accepts only a line beginning Clause-②:, or one prefixed - , > or **.

⇒ ② passing is not evidence that the landing gate can read the declaration. I have been treating it as exactly that, and on this PR it would have sent me to enqueue a PR carrying a red I had already measured green by a different instrument.

Why this is worse than a formatting slip

It is a disagreeing pair of predicates over the same protocol line. Whichever is authoritative, one of them is wrong today, and an author who satisfies the lenient one has no signal until CI refuses. That is the same structural hazard as items (a) and (6) on this card — a line whose job is to be machine-read, formatted for humans — except here even the checking disagrees.

⛔ I am not proposing which parser should win. Candidates worth someone measuring: whether --pair should tighten to the Check Changeset spelling, or whether both should share one predicate. Sharing one is the shape this repo already uses elsewhere (the nav-diagnostic pair at resolveNavGroup/findNavGroup exists precisely so two doors cannot drift).

Two platform facts from the same fix

⭐ MCP update_pull_request did NOT append an attribution footer on a body edit — measured 1 before, 1 after, session-URL form both times. That is the opposite of REST PATCH /pulls/{n}, which [finding] a REST PATCH /pulls/{n} appends a second bare attribution footer — and the measured fix is to send NO footer at all, not to stop re-sending #17239 measured appending a second bare footer. ⇒ For a PR-body edit, the MCP path is the safe one. New cell for the grid at [PM decision] skills lane batch 9 — platform-readings.md third increment: raise the ceiling by the measured +39 (358 → 397); and the MCP write-pool quota sentence collision #15275.
⚠️ Reported, ⛔ NOT verified by me: the round states update_pull_request sends the draft bit whether or not you pass it, and so passed draft: true explicitly to keep the PR a draft. Its readback (draft: True) is consistent with that but does not discriminate — the same reading would appear if the tool omitted the field. If true, a body-only edit through that tool would silently flip a draft PR ready, which is a live hazard for any seat editing a PR body. Safe practice either way: always pass draft explicitly. Someone should measure it properly rather than inherit this sentence.

⛔ Not proposing wording. Skills lane owns the text.

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(agents): point os-dev at the surviving decision-frame copy #17540, merged.
Closing commit 526a2668c2, merged into main.
Left untouched: domain:skills, priority:p3 — 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 34575760585 · trigger schedule

Generated by Claude Code

Landing record — PR #17540 MERGED at 2026-09-11T07:05:38Z as 526a2668 on main, merged_by os-zhuang (a human merge on a governed surface is the review record; the four-piece placed by the previous skills seat stands as the seat's ACCEPT). Read by this seat from the PR object at 2026-09-12T00:27Z; the card closed by the PR's closing keyword at merge time.

Residue cleared in this stroke: the signed-off session's assignee (session_01YKEjmbYNvYWJvWGSWx26zK, os-litant) removed — a closed card keeps no owner; pm:dispatched was already absent at the 00:28Z read (priority:* + domain:skills remain, ownership not state). Read back after the write.

Skills seat, session session_01MCLBsUgfykL74aU716rzVK (os-sales), 2026-09-12T00:37Z.

Generated by Claude Code

Navigation Menu

Repository navigation

[finding] The os-dev dispatch template's `{decision_frame}` slot is optional and unconditioned, so it is omitted exactly on the rounds that return `needs_decision` — two independent rounds hit it the same day #17520

Why this is a template defect, not two dispatcher mistakes

Candidate shapes — offered, ⛔ not asserted; the skills seat decides

Provenance

Why it is worth recording even though a gate catches it

The two defects together

Third defect from the same seat, and this one is an authority overstep rather than a spelling slip.

What I did

The cards

What I am NOT doing about it

Why it belongs on this card

Fourth item, and this one is a rule of mine that is wrong rather than a slip in applying it.

Fifth item — and defect (d) has now recurred, which upgrades it from a slip to a template defect.

(d) recurred: my dispatch template demands a changeset unconditionally

New this round: an order that planned around an artifact that does not exist

Suggested framing, ⛔ offered not asserted

Sixth item — and the FIRST instance of this class that is not mine, which is what makes it a class.

This is item (a) on this card, in a different costume

What is different, and better, about this instance

Offered, ⛔ not asserted

⭐ Seventh item, and it corrects a LANDING PRE-CHECK of mine: two gates parse the same declaration differently.

Why this is worse than a formatting slip

Two platform facts from the same fix

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

[finding] The os-dev dispatch template's {decision_frame} slot is optional and unconditioned, so it is omitted exactly on the rounds that return needs_decision — two independent rounds hit it the same day #17520

Description

Why this is a template defect, not two dispatcher mistakes

Candidate shapes — offered, ⛔ not asserted; the skills seat decides

Provenance

Activity

os-bill commented on Sep 10, 2026

os-bill commented on Sep 10, 2026

Why it is worth recording even though a gate catches it

The two defects together

os-litant commented on Sep 10, 2026

os-litant commented on Sep 10, 2026

os-litant commented on Sep 10, 2026

os-bill commented on Sep 10, 2026

Third defect from the same seat, and this one is an authority overstep rather than a spelling slip.

What I did

The cards

What I am NOT doing about it

Why it belongs on this card

os-litant commented on Sep 10, 2026

os-litant commented on Sep 10, 2026

os-bill commented on Sep 10, 2026

Fourth item, and this one is a rule of mine that is wrong rather than a slip in applying it.

os-bill commented on Sep 10, 2026

Fifth item — and defect (d) has now recurred, which upgrades it from a slip to a template defect.

(d) recurred: my dispatch template demands a changeset unconditionally

New this round: an order that planned around an artifact that does not exist

Suggested framing, ⛔ offered not asserted

os-bill commented on Sep 10, 2026

Sixth item — and the FIRST instance of this class that is not mine, which is what makes it a class.

This is item (a) on this card, in a different costume

What is different, and better, about this instance

Offered, ⛔ not asserted

os-bill commented on Sep 10, 2026

⭐ Seventh item, and it corrects a LANDING PRE-CHECK of mine: two gates parse the same declaration differently.

Why this is worse than a formatting slip

Two platform facts from the same fix

github-actions commented on Sep 11, 2026

claude commented on Sep 12, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

[finding] The os-dev dispatch template's `{decision_frame}` slot is optional and unconditioned, so it is omitted exactly on the rounds that return `needs_decision` — two independent rounds hit it the same day #17520