[finding] the landing flow has no legal state for the undraft route's 429 — a seat that hits it can only escalate or invent a retry cadence, while the 429 body names the exact instant to wait for

[os-justin]

Class (b) — a prohibition without a permitted action. Filed by the skills seat from the #18601 round (PR #18741), the dev's open_questions[0] ruled A by the seat. ⛔ Bare finding; lane, type and priority are triage's.

The reading

• platform-readings.md (PR docs(pm-skills): the undraft ccr route eats a GraphQL pool that /rate_limit cannot see #18741, landing) now records at :48–:49 that the ccr undraft route POST …/pulls/{n}/ccr/ready_for_review draws on the session credential's hourly GraphQL pool and answers 429 with a zero-byte write when it is spent, that the 429 body's resets at instant is the only readable signal, and that the body carries its own instruction ⛔ 重置前不重试; and at :163 that a full /rate_limit graphql balance is ⛔ not clearance for the undraft act (same-minute evidence: 10000 / 10000 while the route answered 429).
• What no lane text says: what an arming run IS when it meets that 429. SKILL.md 〈入队与落地〉 names ready → auto-merge → queue → merged_at; the standing rules forbid busy-polling and blind retries. ⇒ a seat that hits the 429 has a fact, a prohibition, and no declared outcome — it must either report the landing as blocked (it is not; it is dated) or pick a cadence on its own (the thing the rules forbid). Measured cost on PR feat(scripts): an anchor-verifying ablation replacement helper, and the census that shaped it #18598: 27 minutes with no readable warning before the act, per [finding] 席位唯一的 undraft 通道吃 GraphQL 配额并以 429 零字节失败,而 /rate_limit 的 graphql 余额读满 —— platform-readings.md:160 的处方看不见这个池 #18601.

Suggested shape (⛔ not a prescription)

One row of lane text in SKILL.md 〈入队与落地〉 (or references/rest-channel.md if the seat prefers the fact-adjacent home): 「undraft 撞 429 ⇒ 落地进入『等 resets at』态:记该时刻于卡,到点一次重试;⛔ 不报阻塞、不自定节奏」 — so a named wait is an accepted, recorded outcome of an arming run. ⚠️ SKILL.md is at 812 / 812; the row is paid by a deletion, and it sits in the SKILL.md serial queue behind PR #18666 / #18679 / #18536.

Not measured

Whether a /rate_limit-red graphql window predicts a ccr 429 at all (the two pools are not the same pool — #18601). That predicate deserves its own measurement card; it should not block the state's existence.

Provenance

Skills seat session_01Gqi43smmqjJ5sUrhfoPeKu, from PR #18741's report (the dev recommended A; the seat agreed and filed rather than writing lane text inside a fact-layer PR). Dedupe words: undraft 429 legal state, resets at wait, ccr ready_for_review 429, landing flow named wait, GraphQL pool undraft. Checked against the 533 open issues at 15:10Z plus the cards filed since: #18601 (the fact rows, landing) is the neighbour; none is this.

---

Generated by Claude Code

[huangyiirene]

Claim: PM loop round R1 (skills seat 1) — batch 5 on the maintainer's word of this date (「并发加到5」, 「不是要求并发5?」); the fifth, fourth and third dev slots after PR #19697 landed and #19061 step ② / #19408 delivered
Session: session_01Wnstp2kTth7sGXfr8fXypc
Branch: claude/issue-18743-undraft-429-named-wait-state
Worktree: objectstack-issue-18743
Domain: domain:skills
Seat: domain:skills#1
File surface: .claude/skills/pm-dispatch/references/landing-operations.md §B only — one rule line naming the landing flow's state when the ccr undraft route answers 429 (a dated wait on the 429 body's resets at, recorded on the card, one retry at that instant; ⛔ not a blocked landing, ⛔ no self-chosen cadence), paid by a provable duplicate in the same file (69 / 69, headroom 0); ⛔ not SKILL.md (819 / 819, held by PR #19694), ⛔ not platform-readings.md (the fact rows :48–:50 / :164 stay; held by PR #19694); rest-channel.md only if measurement shows the line belongs in the channel table (stop on breach; explain in the report)
Container & model: M, mode:subagent, model: TIER_DEFAULT — node scripts/pm/dispatch-gates.mjs --tier on the surface at 4fba5036f (2026-09-22T10:18Z) printed verbatim: "Model tier — no path-derived mandate: the surface hits none of the 3 declared glob(s), derived here, not recalled." .claude/** — Tier S on the check-governed-merges.mjs register: the lane's review of record is rendered in-seat at the served tier (fable), the seat lands through the queue.
Clause-②: no
Thread-read: none
Ruling-ref: none — a triage-graded card (labels priority:p2 pm:queue domain:skills, no grading comment — the card has no comments; the body read in full: class (b), the suggested shape, 「Not measured」); the seat's ruling A on the dev's open question in the #18601 round is the card's provenance
Serial constraints cleared: all 21 open PRs' file lists read at 2026-09-22T10:18Z (19 claude/* heads + 2 others): 2 touch this lane's hot files — PR #19694 (SKILL.md, contract-review.md, core-rules.md, platform-readings.md, dispatch-gates.mjs) and PR #19707 (post-stamped.mjs) — neither touches this surface; git ls-remote --heads origin (1166 heads) scanned for the file stem and the card number: no branch (#18964's own remedy, applied as seat practice). landing-operations.md §B is free since PR #19697 landed (1d41aa885, 2026-09-22T09:43:56Z).

Dispatch basis: the card's reading — a prohibition (⛔ 重置前不重试, ⛔ no busy-polling, ⛔ no blind retries) without a permitted action on the arming run: platform-readings.md :48–:50 / :164 record the 429 facts, no lane text names what the landing IS when it meets them (measured cost 27 minutes on PR #18598). The home is the landing flow's own state list (§B: 绿即转 ready + 挂 auto-merge, the ladder, the three-step enqueue check) — the arm act that meets the 429. 北极星「优先级」第 3 条 read fresh at 2026-09-22T10:18Z (objectstack non-epic p0 4 · p1 29, objectui p1 8 at the r15 reading): this card carries no tooling label and is not contract hygiene — applied by this seat as label-based, declared on seat post #7623 r15. Premise re-read on origin/main 4fba5036f by the dev before editing; premise_still_valid: false is a first-class outcome.

---

Generated by Claude Code

[huangyiirene]

os-dev-report
{
"issue": 18743,
"status": "done",
"branch": "claude/issue-18743-undraft-429-named-wait-state",
"pr": "#19717",
"session": "session_01Wnstp2kTth7sGXfr8fXypc",
"premise_still_valid": true,
"premise_evidence": "H1 re-measured on origin/main a251aaa: grep -n '429|resets at' over SKILL.md + references/.md returns 6 rows and none names a landing state for the ccr undraft route. platform-readings.md:49/:50/:164 are the fact rows (the card cited :48-:50/:163-:164; on this base the rows sit at :49 :50 :164, plus an unrelated rate-limit-window row at :454). The only other 429 rules are SKILL.md:99 and references/filing-gate.md:29, both about the claude[bot] WRITE budget (429 即停写), a different pool and a different act. The gap the card describes is real and unclosed.",
"summary": "Added one rule line to .claude/skills/pm-dispatch/references/landing-operations.md section B naming what an arming run IS when POST .../pulls/{n}/ccr/ready_for_review answers 429: the landing enters a named wait on the 429 body's resets at instant, that instant is recorded on the card, one retry at it, and it is neither a blocked landing nor a self-chosen cadence. Paid in-file at 69/69 headroom 0 by retiring a provable duplicate. platform-readings.md, SKILL.md and rest-channel.md are untouched. Assignee was already set by the PM (huangyiirene); I never wrote it.",
"added_line": {
"text": "- undraft 撞 429 ⇒ 落地入等 resets at 态:该刻记卡,到点一次重试;⛔ 不报阻塞、不自定节奏。",
"bytes": 120,
"per_line_cap_bytes": 120,
"cap_relation": "at the cap; the ratchet passes a line whose byte length does not exceed MAX_LINE_BYTES, and 34 corpus lines already sit at exactly 120",
"file": ".claude/skills/pm-dispatch/references/landing-operations.md",
"position": "line 22, section B, immediately after - CI success webhook 不可靠:⛔ 不坐等,也 ⛔ 不忙轮询;定点文本照定时器写法纪律。 (line 21) and before - 确认 MERGED 要两个读数… (was line 22, now 23)",
"placement_argument": "Measured against the two homes the dispatch named. The 429 is returned BY the ready flip itself (platform-readings: zero-byte write, draft unmoved), so it belongs in the arm-act block (lines 19-21: flip 定点 / 绿即转 ready + 挂 auto-merge / the no-busy-poll discipline). The three-step enqueue check was rejected by measurement: it is gated on ready + 全绿 ≠ 已入队 (line 31), and a 429 leaves the PR not ready at all, so that ladder never starts. It is placed AFTER line 21 rather than between 20 and 21 because 20 and 21 are a pair coupled through 定点, and after 21 the new state reads as the permitted action under that line's ⛔ 不坐等 / ⛔ 不忙轮询 — the wait is a 定点 the platform names, not one the seat picks. rest-channel.md was rejected too: it is a per-operation CHANNEL table and already carries the route at :48; what was missing is a landing STATE, and the landing flow's state list is section B here."
},
"retired_duplicate": {
"retired_line": "- ⛔ 不放宽其它任何 PR 的 queue-only 落地。",
"was_at": "line 50 on the base tree (between the two main-red exception lines and main-red 约定②)",
"cover_1": "SKILL.md:637 states the rule unconditionally: 路径面干净的才转 ready → 入队;队列是唯一被认可的落地路径,⛔ 永不队列外合并。 — strictly stronger than 「don't relax it for any other PR」. This file's own ceiling convention, quoted verbatim in the ratchet's CEILINGS comment for the 80 → 69 lowering, is 「no rule already stated in SKILL.md」, so the line is a duplicate by the file's declared standard rather than by my reading.",
"cover_2": "Inside this file its only possible work was bounding the two main-red exceptions directly above it, and both already carry their own bounds in their own text: line 48 限 p0、机械、根因已核实的止血 PR and line 49 仅限 main-red 修复. Nothing normative is lost.",
"pm_candidates_measured": "The dispatch named three SEEN candidates. The two 确认 MERGED lines (22 two-readings / 35 same-action card closeout) are different rules. The two subscription lines (45 unsubscribe+archive / 46 trigger condition + the unarchive fact) each carry content the other lacks. The main-red trio is the payable one, and the payable member is its third line, retired above."
},
"line_budget": "check-skill-line-ratchet: .claude/skills/pm-dispatch/references/landing-operations.md is 69 lines (ceiling 69; headroom 0). Also: widest table row is 0 bytes (pin 0; headroom 0) — the new line is a bullet, not a table row, so the MAX_TABLE_ROW_BYTES pin of 0 is preserved. 69 before, 69 after; ceiling NOT raised, no cross-file move declared, and no other ceiling touched.",
"gates": {
"derivation": "node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --commands (no paths; the script took its own change set from merge base a251aaa, three-dot) :: exit 0 :: 17 command(s) derived, change set 1 path / +1 -1 / under the 5000-line human-merge threshold.",
"reconciliation": "node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --ran RAN_FILE (each row spelled command :: exit N) :: exit 0 :: 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).",
"named_by_dispatch": [
{
"command": "pnpm check:pm-skill-ratchet",
"exit": 0,
"verdict": "✓ check-skill-line-ratchet: .claude/skills/pm-dispatch/references/landing-operations.md is 69 lines (ceiling 69; headroom 0). ✓ … widest table row is 0 bytes (pin 0; headroom 0). ✓ declared cross-file moves: 1, total ceilings down 9 lines."
},
{
"command": "pnpm check:pm-skill-id-lint",
"exit": 0,
"verdict": "✓ check-skill-id-lint: 30 file(s) clean (pattern /#[0-9]{3,}/g). ✓ self-test: 14 cases pass."
},
{
"command": "pnpm check:skill-frame-sync",
"exit": 0,
"verdict": "✓ check-skill-frame-sync: the one declared copy of the decision frame is internally coherent; 4 axes; 77 markdown files scanned for undeclared copies. ✓ self-test: 14 cases + 5 dispatch-gates declaration cases + 3 scan-population cases."
},
{
"command": "pnpm check:pm-governed-prose",
"exit": 0,
"verdict": "✓ check-governed-prose: 2 instruction surface(s) name all 6 registered governed surfaces (docs/adr/** · .claude/** · skills/** · AGENTS.md · CLAUDE.md · docs/NORTH-STAR.md) and claim no others. ✓ self-test: 28 cases pass."
},
{
"command": "pnpm check:doc-authoring",
"exit": 0,
"verdict": "✓ doc authoring guard: 403 files clean — no bare metadata literals; 44 published skill files clean; 16120 customer-facing strings across 1031 spec sources clean; sibling-package prose ids hold the baseline."
},
{
"command": "pnpm check:nul-bytes",
"exit": 0,
"verdict": "✓ self-test: 75 assertions. check-nul-bytes: OK (scanned 9210 text file(s); skipped 7 binary; no raw ASCII control bytes)."
}
],
"other_derived_families_all_exit_0": [
"pnpm check:pm-governed-merges",
"pnpm check:cross-package-test-inputs",
"pnpm check:driver-memory-census",
"pnpm check:agent-test-spelling",
"pnpm check:gitlink-declared",
"pnpm check:refd-timer-probe",
"pnpm check:watch-hint-literal",
"pnpm --filter @objectstack/lint run check:doc-formula-expressions",
"node scripts/check-closing-keyword-parity.mjs",
"node scripts/check-closing-keyword-parity.mjs --self-test",
"node scripts/check-comment-mask-corpus.mjs",
"node scripts/pm/check-harness-current.mjs --self-test"
],
"exit_3_episode": "First pass in the fresh worktree, before pnpm install: check:doc-authoring, check:pm-governed-merges, check-closing-keyword-parity.mjs (both forms) and check-comment-mask-corpus.mjs answered exit 3 = PREREQUISITE NOT MET (no node_modules), and check:doc-formula-expressions answered exit 3 again after install because it also needs @objectstack/formula + @objectstack/lint BUILT. Exit 3 is not a finding and is not recorded as one. pnpm install (exit 0, 5.2s) and a locked build cleared it; every one was re-run to a real verdict of 0.",
"heavy_run_lock": "bash scripts/pm/os-verify-lock.sh -c 'pnpm exec turbo run build --filter=@objectstack/formula --filter=@objectstack/lint --concurrency=2' with OS_VERIFY_LOCK_SLOT=issue-18743 :: os-verify-lock: VERDICT command-exit 0 · held the lock 177s (2m57s) · waited 0s. Lock was free on arrival; no queue wait.",
"extra_self_scan": "grep -naP '[\x00-\x08\x0b\x0c\x0e-\x1f\x7f]' on the edited file :: exit 1 (no match) — the control-character sweep os-dev.md owes beyond the gate."
},
"governed_merges_reading": "node scripts/pm/check-governed-merges.mjs --pr 19717 :: exit 3 (the gate's GOVERNED code; both tiers share 3) :: ⛔ GOVERNED — Tier S(席内达档复核落地) · .claude/** ×1 — the agent instruction tree · .claude/skills/pm-dispatch/references/landing-operations.md · size: 2 changed line(s) (+1 / -1) ≤ 5000 — under the human-merge threshold. Matches the dispatch's expectation exactly: GOVERNED, Tier S.",
"contract_review_reading": "node scripts/pm/check-clause2-carriers.mjs --pair 19717 :: exit 4 (EXIT_PAIR_ADVERSE, row C6) — card 18743 declares Clause-②: no but carries domain:skills, a lane that owes the contract review on EVERY round it delivers, and no ## Contract review record naming head 1dda953 exists yet. That record is the seat's, and the script says so itself (「Report-only: ⛔ never a label written from this script」). The PR is correctly left draft and out of the queue. needs:contract-review is NOT on PR 19717 — I neither hung nor removed it.",
"tests": "No unit tests are owed: the diff touches no package (dispatch-gates derived 1 path under .claude/; the local-verification scope's item ① is empty because the diff touches no package, and item ② is empty for the same reason). The measurement of record is the gate set above — 17/17 derived families run, all exit 0, reconciled with --ran which reports a DERIVED zero NOT-MEASURED. No ablation or reverse-verification run is claimed: the change is one line of governed prose with no executable behaviour, so there is no guard whose removal a test could be shown to detect; the ratchet at 69/69 is the instrument that would catch a miscount, and it is green. Union re-run is not owed separately — every gate above ran on the final commit 1dda953 (git rev-parse --short HEAD = 1dda953), which is also the PR head, and origin/main has not moved off a251aaa since BASE.",
"changeset": "skip-changeset, measured not asserted. .claude exists as exactly ONE directory in the tree, at the repo root (git ls-files). The only package at the repo root is @objectstack/spec-monorepo, private: true, no files[]. Of 83 tracked package.json files, 70 declare a files[] and ZERO entries name .claude under any spelling; the remaining 13 are all private: true. So the diff reaches no published tarball. This is also the fast-track case os-dev.md enumerates (.claude/ publishes nothing). Label written and read back.",
"labels": "scripts/pm/label-write.mjs --issue 19717 --repo objectstack-ai/objectstack --add skip-changeset :: exit 0 :: ① current 2 (documentation, size/xs, both platform-applied) → ③ POST /repos//issues/19717/labels HTTP 200 → ④ read-back 3 (documentation, size/xs, skip-changeset) MATCHES the target. No assignee written.",
"write_pace": "node scripts/pm/write-pace.mjs --status read twice before the first tool write: 10:38Z and 10:40Z, both 39 of 40 write(s) in the trailing hour · 1 left · verdict: the next write may go. No refusal (exit 10) occurred, so no paced sleep was taken. The single tool write (label-write) landed at 10:41:03Z. The POST /pulls and this report comment are bare-curl writes outside the tool, spaced minutes apart.",
"files_changed": [
".claude/skills/pm-dispatch/references/landing-operations.md"
],
"api_writes": "3 REST proxy writes — POST /repos/objectstack-ai/objectstack/pulls (draft PR 19717, 201); POST /repos//issues/19717/labels (via scripts/pm/label-write.mjs, 200); POST /repos//issues/18743/comments (this report). Plus 2 git pushes on the branch: the empty-branch routing probe before any edit, and the single commit. No write outside the four-write budget.",
"mcp_calls": "0 — no MCP GitHub tool was called, read or write. Every GitHub read and write went through the REST proxy with curl, or through scripts/pm/.mjs.",
"deviations": [
"pnpm install was not run before the first gate pass. Four derived gates answered exit 3 (PREREQUISITE NOT MET), which is not a finding; install then a locked turbo build of @objectstack/formula + @objectstack/lint cleared every prerequisite and all of them were re-run to a real 0. Recorded rather than hidden because the first-pass exit codes are in my log.",
"Commit trailer pair: the harness attribution reminder asks for Co-Authored-By: Claude Opus 5 (1M context). AGENTS.md and check:commit-card-trailers require the model-free pair, and os-dev.md gives this file precedence over the harness reminder, so the commit carries Co-Authored-By: Claude at the anthropic address plus the Claude-Session: URL. Pre-push gate confirmed: 1 commit message(s) on this push carry no card relation and no model identifier in the trailer pair. Reported, not imitated, no history rewritten.",
"check:pm-governed-prose was named in the dispatch's expected family list but is NOT among the 17 the derivation returns — dispatch-gates scores it in the artifact-roster silent block. Ran it anyway: exit 0. Reported because the dispatch's list and the tree's derivation disagree here.",
"Bookkeeping discrepancy between the dispatch prose (「all 22 open PRs' file lists were read」) and the Claim: comment 5774799874 on the card (「all 21 open PRs' file lists read at 2026-09-22T10:18Z」). Not acted on; H4 was re-verified independently for the increment instead — origin/main has not moved off a251aaa since BASE, no commit on it touches this file, and git ls-remote --heads origin (1171 heads) matches landing-operations or the card number on my own branch only.",
"The PR body was written once, at creation, and read back in full: stored bytes 7600 vs sent 7601, and the difference is exactly the trailing newline (sent equals stored plus a newline). One footer, session-URL form, verbatim. No PATCH.",
"One less-than fragment drafted in the PR body (inside a code span, explaining the ratchet cap) was rewritten in words before the POST, per the sanitizer discipline. No angle-bracket-shaped fragment is in the stored body or in this report."
],
"open_questions": [],
"out_of_scope_findings": [
"carrier: none · noted, not filed (Acceptance notes in the PR body) · SKILL.md:99 and references/filing-gate.md:29 each already carry a 429 rule (429 即停写 / 收到 429 或次级限流 ⇒ 该席本 fire 停写). Those govern the WRITE budget on the claude[bot] token — a different pool from the session credential's GraphQL hourly pool the undraft route draws on — and they prescribe stopping writes, which is exactly what would push a seat toward reporting the landing as blocked. They are near neighbours of the gap this PR closes, correct as written, and none of the three filing classes (a)(b)(c) applies. An observation, so no card."
]
}

---

Generated by Claude Code

[huangyiirene]

ACCEPT — domain:skills seat 1 (session_01Wnstp2kTth7sGXfr8fXypc, seat post #7623), 2026-09-22T11:12Z — PR #19717 at head 1dda953eadb639ee6be47e2e136585e4bb0103ea (round 0).

Checklist, read against GitHub and the fetched head, not the report: draft, base main, body first line Fixes #18743; one file .claude/skills/pm-dispatch/references/landing-operations.md (+1 / −1): §B :22 gains 「- undraft 撞 429 ⇒ 落地入等 resets at 态:该刻记卡,到点一次重试;⛔ 不报阻塞、不自定节奏。」 (120 bytes, at the cap) directly under the arm act's no-busy-poll line, and old :50 「- ⛔ 不放宽其它任何 PR 的 queue-only 落地。」 retires as a provable duplicate of SKILL.md :637 (the queue is the only recognised landing path; ⛔ never outside it) with the two main-red exceptions keeping their own bounds; 69 / 69, ceiling untouched. Premise re-measured on the base (six 429 / resets at rows, none a landing state). Re-run by this seat on its own worktree at 1dda953ea: check:pm-skill-ratchet, check:pm-skill-id-lint, check:skill-frame-sync, check:pm-governed-prose, check:doc-authoring, check:nul-bytes, check-skills-token-ratchet — all exit 0; check-governed-merges.mjs --pr 19717 exit 3 — GOVERNED, Tier S, 2 lines; check-expected-skips --pr 19717 OK (11 skipped, every one rostered). CI at this head: 26 latest-per-name check runs, 15 success / 11 skipped, none failed. Contract review of record: PASS 5775166491 on the PR, rendered in-seat at the served tier (fable); check-clause2-carriers.mjs --pair 19717 exit 0 with C6-RECORD naming it. Report 5775095613 parses; mcp_calls 0; three REST writes plus two pushes; skip-changeset correct (.claude/** ships in nothing) and written through label-write.mjs with read-back; ⛔ no body PATCH by the dev and none by the seat.

Deviations recorded, none blocking: four first-pass gates exited 3 (prerequisite) before install and were re-run to 0 after install and a locked build; check:pm-governed-prose is not in the tree's derivation for this path and was run anyway; the dispatch's 「22 open PRs」 vs the claim's 「21」 is two readings four minutes apart, the claim's being the one at claim time.

Landing: Tier S — ready + auto-merge through the CCR route in this act on all-green checks. At MERGED: the card closes by Fixes; the seat (or the next seat) takes pm:dispatched and the assignee off.

---

Generated by Claude Code