Repository navigation
ci(skills-ratchet): exclude the whole release-bot version: stamp line from the token count and re-pin the ceilings so headroom stays 0 and 1 (#23177 ruling B, item A2′) #23192
Description
Filing-gate category ③, a maintainer-ruled duty: #23177, ruling B (6139629628, the maintainer's 「同意」), item 2 (A2′, which replaces A2).
Reader who acts: the domain:devx execution seat. It lands in scripts/check-skills-token-ratchet.mjs, which is not a governed path, so it goes through the ordinary queue.
What the ruling binds (quoted from 6139629628, item 2)
A2′ (replaces A2): no tolerance band.
scripts/check-skills-token-ratchet.mjscounts authored content only: the frontmatterversion:line that the release bot rewrites is excluded from the count (or normalized to a fixed-width placeholder before counting); ceilings stay shrink-only and the current headroom (0 and 1 tokens) stays as it is. This is also #23176's structural fix; whatever one-token trim is in flight there is the stop-gap, not the route.
Where main stands (so the dev starts from the landed state)
- release(next): #21988 reds on the skills token ratchet at 18.0.0-next.0 (the prerelease version stamp adds 7 bytes to two SKILL.md files with 0 and 1 token of headroom) #23176 did not land a one-token trim. It landed route A (
6138951701) as PR fix(skills-ratchet): price the release stamp's prerelease segment at stable width #23187 →8814211948. The gate prices the stamp's prerelease segment (-next.N) at zero, throughpricedTokens()/stampPrerelease(), reading the stamp withextractMetadataVersionfromscripts/check-skill-compatibility-version.mjs. - What A2′ adds beyond that: the stable part of the stamp still counts today. For example,
18.10.0is one byte longer than18.9.0, so a zero-headroom file can still go red on a version number no author wrote.
What done looks like
- For the files the version script stamps (
stampedSkillPaths()), the whole frontmattermetadata.versionline is excluded from the count, or normalized to one fixed-width placeholder. The count no longer moves with any version value. - Every
CEILINGSrow for a stamped file is re-pinned by exactly the bytes taken out, so each file's headroom onmainis unchanged (objectstack-data0,objectstack-automation1, and the others as they are). It is shrink-only: no ceiling minus its count grows. - The header's counting-convention section and the hand-audit sentence say what now holds. The route-A prerelease-only code is replaced, not kept alongside.
- The self-test pins: a stamp change of any width (
17.7.0→18.10.0-next.12) moves no count; one authored byte over a zero-headroom ceiling is red; a version-shaped string outside the stamp is priced; and the per-file headroom before equals after on the real bundle.
Activity
objectstack-fleet commentedon Oct 13, 2026
Claim: PM loop round 7 ·
domain:devxseat 1 · 2026-10-13T07:02Z · maintainer-ruled: #23177 ruling B, item A2′ (6139629628, 「同意」)Session:
session_012zwsLMxQBwHUYJgPyrpEvxAccount:
zhuangjianguo(the seat's linked user asGET /useranswers it; the card's assignee)Branch:
claude/issue-23192-ratchet-exclude-stamp-lineWorktree:
objectstack-issue-23192Domain:
domain:devxSeat:
domain:devx#1File surface:
scripts/check-skills-token-ratchet.mjsonly (its count, itsCEILINGSrows for stamped files re-pinned downward by exactly the excluded bytes, its header and its self-test). ⛔ No edit toskills/**,scripts/check-skill-compatibility-version.mjsor any governed surface. ⛔ No ceiling's headroom growsContainer & model:
M,mode:subagent,model: opus(the default judgment tier;dispatch-gates --tieron this surface: "no path-derived mandate")Clause-②: no (an internal repo gate)
Responsibility: the ratchet still prices the stable part of the release bot's version stamp, so a zero-headroom file can go red on a version number no author wrote (
18.9.0→18.10.0) | no platform path covers it: route A (PR #23187) excluded only the prerelease segment | the next stable-width change of the stamp reds the gate for no authored reasonThread-read: 6139629628
Serial constraints cleared: no open PR touches
scripts/check-skills-token-ratchet.mjsorscripts/check-skill-compatibility-version.mjs(all open PRs' file lists read at this act).objectstack-fleet commentedon Oct 13, 2026
os-dev-report
{
"issue": 23192,
"status": "done",
"branch": "claude/issue-23192-ratchet-exclude-stamp-line",
"pr": "#23199",
"session": "session_012zwsLMxQBwHUYJgPyrpEvx",
"premise_still_valid": true,
"summary": "scripts/check-skills-token-ratchet.mjs now leaves the whole release-stamp line (frontmatter metadata.version, plus its newline) out of every stamped SKILL.md count at every stamp width; route A (stampPrerelease / STAMP_PRERELEASE_RE) is deleted, not kept alongside. The line is located by the stamp WRITER itself: stampLine() runs deriveFrontmatter with a non-numeric probe and takes the one line it rewrote (a second rewritten line throws), so no second parser and no edit to check-skill-compatibility-version.mjs. Exclusion was chosen over a fixed-width placeholder because on the basis tree d4aaeaf every stamp line is 20 bytes = exactly 5 units, so ceil((B-20)/4) = ceil(B/4)-5 for every file: all ten stamped CEILINGS rows dropped by exactly 5 (ai 6806->6801, api 6319->6314, automation 5785->5780, data 6128->6123, formula 6002->5997, i18n 6338->6333, platform 5833->5828, query 5552->5547, ui 3856->3851, upgrade 6193->6188) and per-file headroom is identical on all 54 priced files (data 0, automation 1; 0 diffs, 44 rows untouched). The hand audit gains one grep term. Judgment call to review: a width-6 placeholder would have needed zero re-pins, which is simpler in a literal sense; I rejected it because it builds today's stamp width into the convention as a constant and keeps pricing bytes no author wrote. The reasons are in the PR body. Premise check: main at d4aaeaf carried route A exactly as the card says, and ablation 2 below shows the stable part still moved the count (18.9.0 -> 18.10.0: 32 -> 33). Self-test roster 76 -> 79 (11 route-A batteries replaced by 14), SELF_TEST_BATTERY_FLOOR 75 -> 79. Commits use the AGENTS.md model-free trailer pair and the PR body the AGENTS.md session-URL footer, per that file's precedence over the harness attribution reminder. The worktree is removed (node_modules deleted first, then git worktree remove, refused nothing), and the scratch clone of changeset-release/main is deleted.",
"tests": "All on final head 0f5c195. Gate self-test: node scripts/check-skills-token-ratchet.mjs --self-test -> "✓ check-skills-token-ratchet self-test: 79 cases pass." (exit 0). Gate: node scripts/check-skills-token-ratchet.mjs -> "✓ check-skills-token-ratchet: 54 authored bundle file(s) within their ceilings; 10 generator-owned file(s) measured, not ratcheted." (exit 0). pnpm check:ratchet-remedy-authority -> "OK check-ratchet-remedy-authority: 275 scripts swept ...; 14 mark the expanding remedy ⛔ MAINTAINER-ONLY" (exit 0). dispatch-gates --repo objectstack-ai/objectstack derived 31 families for the 1-path change set: the 30 direct commands each exited 0 (incl. check:nul-bytes OK over 11195 files, check:entry-guard, check:scripts-symbol-anchors, check:self-test-wired, bare-root-worklist --self-test, check:cross-package-test-inputs), and pnpm check:pm-dispatch-gates exit 0 with "✓ dispatch-gates self-test (fast tier): 1833 cases pass". --ran reconciliation at 0f5c195: 31 derived, 31 run, 0 NOT-MEASURED, 0 UNRUN. Script-suite duty: no .test. names this script (git grep), so its --self-test is its suite. eslint narrowed (CI owns pnpm lint): npx eslint --no-inline-config --format json on the one file -> 1 file, 0 errors, 0 warnings; population = eslint.config.mjs glob **/*.{...,mjs,...}, resolved by --print-config; invariance = resolved parserOptions {ecmaVersion, sourceType} with no project, so no type-aware rule exists and untouched files cannot change. Before/after on main: per-file table in the PR body; 0 headroom diffs over 54 files. Ablations (one-time, scripts/ablation-replace.mjs; each mutation proven on disk, anchor 1->0 and blob changed; each restore proven, blob == HEAD 5c98de59e and git diff HEAD empty; no build/dist involved, since the gate runs from source): (1) unpriced=0 -> self-test 4 red (width invariance got [32,34,33,35]; zero-headroom green; hand audit; real-bundle re-pin), gate red automation over by 4 and data over by 5; (2) route A restored -> same 4 red, width invariance [32,33]; (3) every indented version: line excluded -> 3 red ("a version-shaped string outside the stamp is priced" 13 vs 18; hand audit and re-pin name objectstack-platform/SKILL.md, which has a second indented version: line); (4) numeric probe 1.0.0 -> self-test and gate both throw "the stamp writer rewrote 2 lines for one probe stamp". changeset-release/main at 0b5641e (stamp 18.0.0-next.0, a one-time shallow sparse clone, deleted afterwards): its route-A gate exit 0; this script copied in: gate exit 0, self-test 79 pass, all 54 headroom values identical to main. With the clone restamped to 18.0.0-next.0, 18.10.0, 18.10.10-next.12, 18.100.10 and 100.100.100, objectstack-data read 6123/6123 at every width, while route A went red at 100.100.100 (6129 over 6128).",
"mcp_calls": "0 — no MCP GitHub tool called",
"api_writes": "3 — each one relay stroke sent as POST /repos/objectstack-ai/objectstack/dispatches (scripts/pm/fleet-write/dispatch.mjs) and executed by fleet-write.yml as objectstack-fleet[bot]: (1) pr_create -> POST /repos/objectstack-ai/objectstack/pulls (draft) + POST /repos//issues/23199/assignees [zhuangjianguo], body read back 11658/11658 bytes identical; (2) labels_add via scripts/pm/label-write.mjs -> POST /repos//issues/23199/labels [skip-changeset], read back size/m + skip-changeset; (3) this report via scripts/pm/post-stamped.mjs -> POST /repos//issues/23192/comments. Plus 3 git pushes (empty-branch probe, 5b7df2d, 0f5c195), which are not REST writes.",
"open_questions": [],
"out_of_scope_findings": [
"carrier: 承接者:无 · noted, not filed (only in the PR Acceptance notes): deriveFrontmatter also rewrites the compatibility: major pins (@objectstack/spec 17.x), and those stay priced because A2′ names only the version: line. Their width changes only when a major gains a digit (next case 99 -> 100), so this is dormant. Dedupe words: skills ratchet, compatibility pin, major width, release stamp"
]
}
objectstack-fleet commentedon Oct 13, 2026
Review: ACCEPT · PR #23199 (draft, head
0f5c195b1) ·domain:devxseat 1 ·session_012zwsLMxQBwHUYJgPyrpEvx· 2026-10-13T07:48ZChecked on GitHub, not on the report's say-so.
PR shape and scope
main. Body line 1 isFixes #23192, line 2 isClause-②: no.zhuangjianguo. Labelssize/mandskip-changeset, which is right for a root gate script.scripts/check-skills-token-ratchet.mjs(+208 / −95).check-governed-merges --pr 23199), 303 lines.skills/**, no change tocheck-skill-compatibility-version.mjs, no workflow.6139629628item 2 (A2′) is quoted verbatim in the body.The change, read on the diff
CEILINGSrows each drop by exactly 5: ai 6806→6801, api 6319→6314, automation 5785→5780, data 6128→6123, formula 6002→5997, i18n 6338→6333, platform 5833→5828, query 5552→5547, ui 3856→3851, upgrade 6193→6188. Each row's comment carries the A2′ step.version: "17.7.0"plus its newline is 20 bytes on every file. Soceil((B−20)/4) = ceil(B/4) − 5, and every file's headroom is unchanged. The report reads 0 diffs over 54 priced files (data 0, automation 1). No ceiling minus its count grows.stampLine()asks the stamp writer itself. It runsderiveFrontmatterwith a non-numeric probe, takes the single line that changed, and throws if more than one line changes (ablation 4: a numeric probe rewrites the compat pin too, and the gate refuses rather than guesses). There is no second parser.stampPrerelease()andSTAMP_PRERELEASE_REare deleted, not kept alongside. They had no consumers outside this file (git grep).version:line excluded: the outside-stamp row fails, namingobjectstack-platform/SKILL.md's secondversion:line.Evidence (report)
main: the gate exits 0 over 54 files;check:ratchet-remedy-authorityis OK; 31 derived gates were run, all exit 0, with 0 NOT-MEASURED.0b5641e55with this script, the gate exits 0 and headroom is identical tomain.18.0.0-next.0,18.10.0,18.10.10-next.12,18.100.10and100.100.100, data reads 6123/6123 at every width. Route A went red at100.100.100.CI on
0f5c195b1at this review: 24 green, 11 skipped, 6 running, none red.Landing held, deliberately: #21988 (
18.0.0-next.0) is in the merge queue now. The queue groups entriesALLGREEN, so this seat arms this PR only after #21988 has merged. The release candidate never shares a group with an unrelated change.Out-of-scope finding (not filed): the
compatibility:major pins stay priced. A2′ names only theversion:line, and they change width only when a major gains a digit. They are noted in the PR's Acceptance notes.objectstack-fleet commentedon Oct 13, 2026
Landed: PR #23199 →
44e2d505c7(Fixes #23192) ·domain:devxseat 1 ·session_012zwsLMxQBwHUYJgPyrpEvx· 2026-10-13T08:21Zmain,scripts/check-skills-token-ratchet.mjsis blobf386e08a56, equal to the PR head0f5c195b1.completedon the merge, andpm:dispatchedis removed with this record.main. The ratchet leaves the whole release-stamp line out of every stamped SKILL.md count. The ten stamped ceilings each moved down by 5, so per-file headroom is unchanged.18.0.0-next.0) merged at 2026-10-13T08:04Z, so the release candidate never shared a merge-queue group with this change.