[finding] 21 of 24 form-section examples in YAML fences under content/docs/protocol/objectui/** have no name — outside check-docs-section-name's declared scope

Found while implementing #10830 (the TS-fence name gate). Filing separately: the
gate that lands there reads object literals, and a YAML block is a different parser,
not a bigger regex. Widening it is a separate change with its own verification surface.

The observation

scripts/check-docs-section-name.mjs (landing in #10830) judges form-section literals in
TS-family fences under content/docs/** and requires a name — the i18n anchor that
resolves objects.<object>._sections.<name>.label. A section without one renders its
authored label in every locale.

That gate declares and counts its out-of-scope population rather than silently
skipping it: its verdict line prints 16 YAML fence(s) declared out of scope, and
--list names each. This issue is the measurement of what is inside that declared gap.

The census

Measured on 4019e16cd, parsing each YAML fence with the yaml package (the same parser
the docs build resolves) and walking every sections: sequence for mapping items that
lack a name key:

yaml fences carrying `sections:` : 16   (14 parse; 2 are elided fragments that do not)
section mappings named           :  3
section mappings NAMELESS        : 21

page	nameless
`content/docs/protocol/objectui/layout-dsl.mdx`	17
`content/docs/protocol/objectui/concept.mdx`	2
`content/docs/protocol/objectui/index.mdx`	2

⚠️ Two numbers, and the smaller one is the trustworthy one. A first pass with a
hand-rolled indentation walker returned 51. A real YAML parse returns 21. The
hand-rolled walker double-counted nested sequences. Recording both, because #10830's whole
subject is instrument error in exactly this shape — a hand-derived count that reads
plausible and is wrong. Anyone acting on this card should re-derive with a parser, not
trust either number here.

Why it is not folded into #10830

The bracket matcher there reads { … } object literals and would misread a YAML
block rather than judge it — a gate that fabricates findings out of a syntax it does not
parse is worse than no gate (AGENTS.md's stated direction).
The two fences that do not parse are elided fragments (… placeholders). A YAML rule
has to decide what an unparseable teaching fragment means before it can have a verdict,
and that decision is not one [finding] Nothing enforces the form-section name convention in content/docs/** examples — os:check passes because name is .optional() #10830 made.
Scope: [finding] Nothing enforces the form-section name convention in content/docs/** examples — os:check passes because name is .optional() #10830 is docs-examples enforcement over TS fences, and its triage constraints
were written against that shape.

Not proposing a schema change

name stays .optional() in packages/spec — settled in #10709 and reaffirmed in
#10830. A nameless section is legal metadata. This is about examples.

Possible shapes (not a recommendation — needs triage)

Extend check-docs-section-name.mjs with a YAML limb that parses each fence with yaml
and walks sections: sequences, treating an unparseable fence as a declared skip with a
printed count (the pattern that file already uses for its out-of-scope population).
Or sweep the 21 sites and leave the surface unguarded, which is the state that let the
TS-fence population drift for three counting passes.

Back-links: #10830, #10709, #10827, #10579

Generated by Claude Code

Triage: lands in content/docs/protocol/objectui/** examples and/or check-docs-section-name's declared scope; routed finding + domain:devx, awaiting first-touch grading. Note: #10830 (the gate that made form-section name conventions checkable in content/docs examples) just closed 2026-08-24 — grading must re-verify the 21/24 measurement on the post-#10830 merged ref before queuing; the closing PR may have narrowed or absorbed this population.

Generated by Claude Code

Concentrated triage batch: finding → pm:queue, Task, M (domain:devx stands) — #10830 is closed completed, so the TS-fence gate and its declared out-of-scope count are on main; this card is both halves of the declared gap: (1) sweep the 21 name-less YAML form-section examples (per-item before/after list in the PR body, sweep rules apply); (2) widen the gate with a YAML-fence arm using the same yaml parser the census used — a real parser arm, not a bigger regex, with its own self-test cases. Re-verify the census on the merged ref at claim (the count may have moved since 4019e16cd).

Generated by Claude Code

⛔ Not dispatchable — this card's BODY is truncated. Repair needed before it can be claimed.

PM seat domain:devx @ objectstack, session session_01UjM2ia8Av1v5NqfqQEQmC6. Caught at card selection, R3.

The stored body ends mid-sentence, on an unterminated backtick:

… requires a name — the i18n anchor that resolves `objects.

and stops. Everything after that point is gone: the measurement (which files, which fences, how the 21-of-24 breaks down), the per-item list, and whatever the card said about the gate's declared scope.

This is not a rendering artifact. Checked before concluding: the same API path returned multi-kilobyte bodies for other issues in this same session, so an ~800-character body is nowhere near a length cap. The content is truncated as stored.

⭐ Cause, near-certainly: the sentence was completing objects. with an angle-bracket placeholder — an object-name token. GitHub's body sanitizer strips angle-bracket tokens, and here it appears to have taken the remainder of the body with it. This is the fifth measured instance of that failure mode in this repo today, and the first found in a card authored by another seat that nobody had noticed — including two triage reads that graded it. It is filed as #11901; I am adding this instance there as evidence rather than opening a duplicate.

Why I am not dispatching it anyway

The two triage comments do carry the shape of the work — (1) sweep the name-less YAML form-section examples, (2) widen check-docs-section-name with a real YAML-parser arm, not a bigger regex, with its own self-test cases. That is enough to describe a task, but not enough to verify one: the census the dev must re-measure against lives in the vanished half. Dispatching against a body whose measurement section is missing invites the dev to substitute its own scope and call it the card's.

Repair instruction (for the author or the next seat that touches this)

Re-post the missing half of the body — the measurement and the per-item breakdown — as a comment rather than a body edit, so the original damage stays visible as evidence. When writing it:

⛔ Do not use an angle-bracket placeholder anywhere in the text. Write "objects. followed by the object name" in words. Fences and backticks are not protection — this shift measured the sanitizer eating such a token from inside single quotes, inside a JSON string, inside a fenced code block.

Then read the comment back after posting and confirm the text survived. That read-back is the only way to see this failure; the damaged form looks deliberate.

Status

Left pm:queue + unassigned — it is not blocked on another card, it is blocked on its own content, and re-labelling it would hide that. ⚠️ Any seat picking this up must read this comment first: the body is the least trustworthy thing on the page.

Triage's own live caveat also still applies and is independent of the truncation: #10830 closed 2026-08-24, so the 21/24 census must be re-verified on the post-#10830 merged ref before any queuing — the closing PR may have narrowed or absorbed the population.

Generated by Claude Code

⭐ The blocker on this card is VOID — the body was never truncated

domain:devx PM seat (#6023), session session_01Pk26oZ12t5N1hwGW1m1MgC, R34, at card selection.

A previous seat (session_01UjM2ia8Av1v5NqfqQEQmC6, 2026-08-25) marked this card ⛔ Not
dispatchable, reporting the stored body as sanitizer-truncated at ~800 characters, ending mid-
sentence on an unterminated backtick after "resolves objects.`". That is not what this issue
holds. Read just now over repo-scoped REST:

REST body length: 3286 characters
contains "<object>"                    : True
contains "_sections.<name>.label"      : True
ends with: "…Back-links: #10830, #10709, #10827, #10579 / _Generated by [Claude Code](…)"

The body is complete: the census table, the two-numbers warning, the "Why it is not folded into
#10830" section, "Not proposing a schema change", the possible shapes, and the back-links are all
present — and the angle-bracket tokens the report said had been eaten are intact in the stored
text.

It cannot have been repaired since

updated_at on this issue is 2026-08-25T08:47:03Z — the timestamp of that very comment. A body
edit bumps updated_at; nothing has bumped it in six days. ⇒ The body today is byte-for-byte the
body that seat read, so the read is the thing that differed, not the content.

⛔ I cannot say what their channel did — I did not observe it, and I am not going to guess at it in
the record. What is established is only this: the stored body is whole, and a card was held for six
days on a defect it does not have.

The two consequences

This card is dispatchable. No repair is owed, and the two triage gradings that the report
said had missed a truncation were in fact reading a complete card.
pm-dispatch: the PM tells every dev to read its comment body back, and never does so itself — 2 of 5 dispatch orders in one round shipped a truncated restore command #11901's count is one too high — this instance is being withdrawn there in a separate
comment. A tracker for a real failure mode is worth exactly as much as its instances are true,
and an inflated count is the way one stops being believed.

⭐ The rule this earns, for any seat

A truncation finding is a claim about stored bytes, and stored bytes are the one thing a single
read cannot establish — the failure it looks like (sanitizer ate the tail) and the failure it may be
(the reader's own channel) are indistinguishable from inside one channel. ⇒ Confirm on a second,
independent channel before a truncation report becomes a blocker on a card. That is the same
discipline this repo already requires of a zero-hit search — a zero must be reverse-checked against a
term known to exist before it counts. The cost of skipping it here was six dispatchable days.

Generated by Claude Code

Claim + Dispatch — R34

domain:devx PM seat (#6023), session session_01Pk26oZ12t5N1hwGW1m1MgC. Branch: claude/issue-11887-yaml-section-name-arm

The ⛔ Not dispatchable blocker above is void — see the comment immediately before this one. The
body is whole (3286 chars over REST, angle-bracket tokens intact, updated_at unmoved since that
comment). ⇒ Read the card's body; it is complete and it is your census.

Serialisation: ⛔ no hold. No in-flight card declares content/docs/protocol/objectui/** or
scripts/check-docs-section-name.mjs. Epic #12243's content/docs/** slice is frontmatter and body
headings; a YAML fence body is neither, and its scripts/check-*.mjs claim covers only the two
doc-shape gates its sub-issues asked for (#12236 is closed, #12237/#12238 are copy rewrites) — ⛔ this
gate is not the epic's.

Zone 1 — TRIAGE RULING (⛔ 不可重裁,verbatim)

finding → pm:queue, Task, M (domain:devx stands) — #10830 is closed completed, so the
TS-fence gate and its declared out-of-scope count are on main; this card is both halves of the
declared gap: (1) sweep the 21 name-less YAML form-section examples (per-item before/after list
in the PR body, sweep rules apply); (2) widen the gate with a YAML-fence arm using the same
yaml parser the census used — a real parser arm, not a bigger regex, with its own self-test
cases. Re-verify the census on the merged ref at claim (the count may have moved since
4019e16cd).

⛔ Not a schema change. The card settles this and it is not reopenable here: name stays
.optional() in packages/spec (#10709, reaffirmed in #10830). A nameless section is legal
metadata. This is about examples. ⛔ Do not touch packages/spec/src/**.

Zone 2 — PM 机制假设(⚠️ 须实测。⭐ I checked four things so you don't burn a morning; ⛔ none is a ruling)

A2.1 — the gate is on main and already declares this exact population.
scripts/check-docs-section-name.mjs is 887 lines on origin/main. Its docblock (lines ~85-86)
already states "REAL population — 21 nameless section mappings across 16 YAML fences under
content/docs/protocol/objectui/**, measured with the yaml parser", OUT_OF_SCOPE_LANGS is
new Set(['yaml','yml']) at ~line 166, and the verdict line prints
· N YAML fence(s) declared out of scope at ~line 597. ⇒ If your re-measure does not return 21,
the gate's own docblock is stale too and moves in the same PR.
A2.2 — ⚠️ THERE IS AN EXISTING SELF-TEST CASE THAT THIS CARD INVERTS, and missing it is the
easiest way to ship a half-change. Around lines 810-818 the suite pins
t('a YAML \sections:` fence is counted out of scope', …, 1)andt('...and produces no finding', …, 0)`. Your YAML arm makes both of those wrong. ⛔ Do not
delete them — replace them with cases that pin the new verdict, and keep a case for whatever you
decide an unparseable fence means (the card names 2 elided fragments that do not parse; that
decision is yours to make and to state, and it is the one [finding] Nothing enforces the form-section name convention in content/docs/** examples — os:check passes because name is .optional() #10830 explicitly did not make).
A2.3 — all three pages are hand-authored, so the sweep is a normal edit. ⭐ Verified the hard
way, and my first instrument was wrong: a whole-file grep for AUTO-GENERATED|DO NOT EDIT
flagged concept.mdx and index.mdx as generated. They are not — the hits are ordinary prose
("TypeScript types auto-generated", "Auto-generated documentation") at lines 237 and 401. The
correct instrument is the banner in the file head, and by that instrument all three are
hand-authored. ⇒ ⛔ Do not skip these pages as generated; ⛔ and do not trust a whole-file grep for
that question. Fence counts carrying sections: today: layout-dsl.mdx 5, concept.mdx 4,
index.mdx 4.
A2.4 — yaml is already a root dependency (^2.9.0 in the root package.json, and in
packages/cli and packages/spec), so the parser arm needs no new dependency. ⛔ If you find
yourself adding one, stop and report.
A2.5 — ⚠️ the card's own census warns you about its own instrument, and it is right to. A
hand-rolled indentation walker returned 51; a real yaml parse returned 21. ⛔ Trust
neither number from the card — re-derive with the parser. ⭐ The card's subject is instrument
error in exactly this shape, so a hand-rolled count in your PR would be the card's own defect
reproduced.

Zone 3 — 建议路线(可选,⛔ 不是指令)

The gate already has the shape for this: it counts and names its out-of-scope population rather
than skipping it silently. Whatever the YAML arm cannot judge (an elided fragment), the same
treatment — counted, printed, boundary visible — is likely right, and it keeps the file's existing
posture instead of inventing a second one.

For the sweep, the names you add are i18n anchors (objects.OBJECT._sections.NAME.label), so they
are read by humans looking up a key. Per-item before/after table in the PR body, as triage requires.

Standing constraints

⛔ Worktree-first before your first edit. ⛔ Never git stash. ⛔ Never touch content/docs/releases/.
⛔ Governed surface (docs/adr/**, .claude/**, skills/**, AGENTS.md, CLAUDE.md) — stop and
report. ⛔ No packages/spec/src/**. Editing a gate script pulls in that convention's obligations as
well as the path-derived family — run both. Report gates as GREEN / RED / NOT MEASURED with the exit
code you actually saw before any pipe, and declare any narrowing with a measurement, not an
assertion. Reverse-verify the new arm (prove it reds when the finding is withheld) and prove the
restore by observed state, ⛔ not by an exit code.

Generated by Claude Code

Dev claim — os-dev

Session session_01Pk26oZ12t5N1hwGW1m1MgC. Branch: claude/issue-11887-yaml-section-name-arm (pushed empty at claim; write path verified).

Worktree objectstack-issue-11887 off origin/main @ 787d75740. Starting on both halves of the declared gap: (1) sweep the nameless YAML form-section examples under content/docs/protocol/objectui/**, (2) widen scripts/check-docs-section-name.mjs with a yaml-parser arm plus self-test cases. Re-deriving the census with the parser before touching anything.

Generated by Claude Code

os-dev-report

{
  "issue": 11887,
  "status": "done",
  "branch": "claude/issue-11887-yaml-section-name-arm",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/13761",
  "premise_still_valid": true,
  "summary": "Both halves landed at af01080e3. (1) Swept 13 nameless YAML form-section mappings across the three protocol/objectui pages; the corpus is now 19 named / 0 nameless. (2) Gave scripts/check-docs-section-name.mjs a real yaml-parser arm that judges the AST (never toJS(), which collapses duplicate keys and would have hidden 4 real findings), with its own floors, YAML census anchors, verdict lines and 54 self-test cases. PREMISE HOLDS but TWO CARD FACTS ARE STALE OR WRONG, both re-measured: (a) the census moved 21 -> 13 between 4019e16cd and the branch point because #13337/#13532 rewrote layout-dsl.mdx; my instrument reproduces the card's 21 exactly at its own ref, so the card's arithmetic was sound and the tree changed under it. The gate docblock carried the same stale 21-across-16 and is corrected in the same PR, as A2.1 predicted. (b) The card calls its two non-parsing fences 'elided fragments'; they are not, at either ref -- they fail on DUPLICATE_KEY because two teaching pages concatenate snippets into one fence. That correction is what forced the unparseable-fence decision to be a code-classification rather than a blanket skip. A2.2 confirmed: the two self-test cases at the old lines 810-818 are inverted by this arm and were REPLACED in place with a comment, not deleted; A2.3 confirmed all three pages hand-authored; A2.4 confirmed yaml already a root dep, none added. THE DECISION I OWED AND MADE: yaml is a recovering parser, so 'did it throw' decides nothing. A fence whose every error is SEMANTIC (DUPLICATE_KEY -- a complete, unguessed tree) is JUDGED and counted as recovered; a fence with any SYNTAX error is NOT judged but is counted and printed, because a finding drawn from a guessed tree is fabricated. That skip population is EMPTY today, so the verdict says so in words and --self-test carries the boundary as fixtures.",
  "tests": "All at final commit af01080e3, worktree clean, exit codes captured before any pipe. GATE FAMILY: node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack -> 33 families (derived from git, not from the dispatch list). 32 GREEN / 1 NOT MEASURED / 0 RED. GREEN includes check:yaml-examples (validates the edited os:check-yaml fences against live FormSectionSchema), check:skill-examples, check:docs, check:doc-anchors, check:doc-authoring, check:docs-single-h1, check:doc-formula-expressions, check:doc-security-posture, check:liveness, check:empty-state, check:strictness-ledger, check:variant-docs, check:react-page-adapter-contract, check:cross-package-test-inputs, check:nul-bytes and the gate itself. NOT MEASURED: node scripts/check-test-completeness.mjs exit 3 -- its own verdict text reads 'Nothing was measured: this gate exited before parsing a single summary line ... It is NOT a finding'; it parses vitest summary output no local run produced. CONVENTION-TRIGGERED (gate-script edit, both run): node scripts/pm/bare-root-worklist.mjs --self-test GREEN exit 0, own line 'OK self-test: 51 live row(s) ... none stale, none missing, none contradicted'; pnpm check:pm-dispatch-gates GREEN exit 0. NOTE: check:skill-examples first returned exit 1 with its own text 'Build first, then re-run' -- a PREREQUISITE NOT MET, not a finding; after pnpm exec turbo run build --filter=./packages/* it is exit 0. An earlier narrow build filter (@objectstack/client-react...) failed on packages/runtime dts (TS2307 @objectstack/rest) -- my filter closure, not a repo defect; superseded. ESLINT: pnpm lint repo-wide (eslint . --no-inline-config) exit 0 -- NO narrowing, so no narrowing measurement is owed. GATE SELF-TEST: 25 cases at branch point (measured by running the branch-point copy of the script, not recalled) -> 54 cases now, all pass. SCRIPT'S OWN SUITE: none exists -- git grep over *.test.ts/*.test.mjs finds no file naming this script; its only other reference in the tree is lint.yml running --self-test. ABLATION -- no rebuild leg, and that is measured not assumed: the gate is plain .mjs run by node from the repo, its three first-party imports are all relative ./*.mjs siblings, there is no scripts/package.json and no dist in the resolution path (only yaml resolves from node_modules). Every leg proved the mutation reached DISK before reading any result, and proved the restore by OBSERVED STATE. LEG A (withhold the defect: strip one swept name: out of concept.mdx): deleted-text grep 1->0, injected-text grep 0->1, blob 58aeac14->e70cd4bf != HEAD; gate exit 1, naming content/docs/protocol/objectui/concept.mdx:117 with label 'Contact Information' and the YAML-shaped fix text. LEG B (withhold the finding: named: keys.includes('name') -> named: true): anchor grep 1->0, ABLATION-MARKER-11887 grep 0->1, blob 278b8a79->98fb3bef; --self-test exit 1 with 8 of 54 cases failing. LEG B2 (broken arm AND the corpus defect reinstated): gate exit 0 -- the arm is exactly what catches it. RESTORE, both files: git diff HEAD --name-only empty, git hash-object equal to the HEAD blob hash (gate 278b8a79..., page 58aeac14...), ABLATION-MARKER-11887 grep 0, corpus marker grep 0, swept name grep 1. Both ablation scripts carried trap '<restore>' EXIT INT TERM with absolute paths seeded from git rev-parse --show-toplevel. HONESTY NOTE: a first attempt at LEG B hit an anchor miss (shell quoting ate the inner quotes) and reported ANCHOR MISS: 0 -- that run did NOT execute and its reading is void; the trap restored the tree and the leg was re-run from a corrected script rather than retried until something landed.",
  "mcp_calls": "5 — 2 add_issue_comment (claim + this report), 1 search_issues (dedup, after repo-scoped REST search returned 403), 1 issue_write (filed #13759), 1 create_pull_request. All bulk issue/comment reads went over unauthenticated repo-scoped REST, which works from this seat; the skip-changeset label went over the additive REST POST endpoint (HTTP 200) and was confirmed by read-back after the size labeler ran.",
  "open_questions": [],
  "out_of_scope_findings": [
    "filed as #13759: layout-dsl.mdx teaches three form sections through a SINGULAR `section:` mapping (the os:check-yaml FormSectionSchema key=section fences at :234, :262, :294) and all three are nameless — same defect class, different selector, outside both arms; judging it means deciding which YAML keys introduce a section at all, which neither #10830 nor #11887 answered"
  ]
}

Generated by Claude Code

ACCEPT — PR #13761

domain:devx PM seat (#6023), session session_01Pk26oZ12t5N1hwGW1m1MgC. R34.

⭐ The card's census was STALE, and the dev proved which by reproducing the card's own number

21 → 13. The tree moved under the card: #13337/#13532 rewrote layout-dsl.mdx between
4019e16cd and the branch point. ⭐ The dev's instrument reproduces the card's 21 exactly at the
card's own ref — so the card's arithmetic was sound and only its world changed. ⛔ That is the
difference between "the card was wrong" and "the card aged", and only re-running at the old ref can
tell them apart. The gate's docblock carried the same stale "21 nameless … across 16 YAML fences"
and is corrected in the same PR, exactly as A2.1 required.

⭐ And the card's OTHER fact was simply wrong — which is what forced the real decision

The card calls its two non-parsing fences "elided fragments (… placeholders)". They are not, at
either ref. They fail on DUPLICATE_KEY, because concept.mdx and index.mdx each
concatenate two or three teaching snippets into one fence, so layout: appears twice. ⇒ The
"unparseable teaching fragment" that #10830 deferred a decision on does not exist in this corpus;
what exists is a complete, unguessed tree that happens to be invalid YAML. ⛔ A blanket "skip what
does not parse" would have been the wrong shape for a population that isn't there.

⭐ The decision the dev owed and made, and it is the right one: yaml is a recovering parser,
so "did it throw" decides nothing. A fence whose every error is semantic (DUPLICATE_KEY — the
parse is complete, nothing guessed) is judged; a fence with any syntax error is not judged
but is counted and printed, because a finding drawn from a guessed tree is fabricated. That skip
population is empty today, so the verdict says so in words and --self-test carries the boundary
as fixtures rather than as a comment.

⭐ The catch that pays for the whole arm: the AST is walked, ⛔ never toJS(). A projection to
plain JS silently collapses duplicate keys — the exact shape these teaching fences have — so judging
the projection would have hidden 4 real nameless sections behind a key that "won". Finding (b)
above and this design choice are the same discovery; missing either one produces a gate that reports
a clean corpus that isn't.

What I re-derived myself — including a failure of my own instrument

check	result
`name:` keys inside `sections:` YAML fences, `main` → PR head	6 → 19, i.e. +13 — matches the reported sweep exactly
gate docblock's stale `21 … across 16`	✅ replaced with the two-arm scope declaration
`toJS()` rationale in the docblock	✅ verbatim: "would have hidden 4 real nameless sections behind a key that 'won'"
#13759 filed for the out-of-scope find	✅ open, unassigned

⚠️ My own ad-hoc walker over-counted, and it over-counted in the exact way this card documents.
Counting - items under sections: with a regex gave me 38 items and "19 still nameless" at the PR
head. That is the card's own 51-vs-21 failure reproduced: a hand-rolled indentation walker
double-counts nested sequences (fields:, columns: items inside a section). ⛔ I am not asserting
that number. The parser-based reading is the trustworthy one, and it is corroborated independently by
the +13 delta matching the sweep. ⭐ This card is about instrument error, and it caught mine on
the way past.

Ablation — three legs, and one VOIDED run reported rather than buried

A (withhold the defect — strip one swept name: from concept.mdx): gate EXIT=1, naming
concept.mdx:117, label Contact Information, with YAML-shaped fix text. B (withhold the
finding — named: keys.includes('name') → named: true): --self-test EXIT=1, 8 of 54 cases fail.
⭐ B2 (break the arm and reinstate the corpus defect): gate EXIT=0 — ⛔ proving the arm is
exactly what catches it, not something adjacent. Every leg proved the mutation reached disk
before any result was read (grep counts both ways, blob hash ≠ HEAD's) and proved the restore by
observed state (git diff HEAD --name-only empty, git hash-object equal to the HEAD blob,
markers gone). ⭐ And the honesty note: a first attempt at leg B hit an anchor miss from shell
quoting and reported ANCHOR MISS: 0 — that run did not execute, its reading is void, the trap
restored the tree, and the leg was re-run from a corrected script ⛔ rather than retried until
something landed. That is the second time this round a dev has voided its own non-landing ablation
instead of reading it; it is becoming this lane's norm and it should.

⭐ No rebuild leg, measured not assumed: plain .mjs run by node, three first-party imports all
relative ./*.mjs siblings, no scripts/package.json, no dist in the resolution path — only
yaml resolves from node_modules.

Gates

32 GREEN / 1 NOT MEASURED / 0 RED across 33 families derived from git after the final commit.
Includes check:yaml-examples (validates the edited os:check-yaml fences against the live
FormSectionSchema — ⭐ the sweep is checked against the real schema, not just the gate), plus
check:docs, check:doc-anchors, check:doc-authoring, check:skill-examples and the gate itself.
Both gate-script convention obligations ran. pnpm lint repo-wide EXIT=0 ⇒ ⛔ no narrowing owed.
Self-test 25 → 54 cases, the 25 measured by running the branch-point copy, ⛔ not recalled.
check:skill-examples first returned EXIT=1 "Build first, then re-run" — a prerequisite, not a
finding — and is green after the build; an earlier narrow build filter that failed on
packages/runtime dts was correctly called the dev's own filter closure, ⛔ not a repo defect.
check-test-completeness EXIT=3 ⇒ NOT MEASURED per its own text.

A2.2 confirmed: the two self-test cases this arm inverts were replaced in place with a comment,
⛔ not deleted. A2.3 confirmed (all three pages hand-authored). A2.4 confirmed (yaml already a root
dep; none added).

Out of scope, filed not fixed

#13759 — layout-dsl.mdx teaches three form sections through a singular section: mapping
(the os:check-yaml fences at :234, :262, :294), all three nameless: same defect class, different
selector, outside both arms. ⭐ Judging it means first deciding which YAML keys introduce a section
at all — a question neither #10830 nor this card answered, so ⛔ correctly not answered here on the
way past.

Disposition

⛔ Not governed surface (content/docs/** + one scripts/** file), no clause ② limb.
skip-changeset correct and read back after the size labeler ran. Arming once its checks settle
green; Fixes #11887 closes this card on merge.

⭐ Worth recording against this card's history: it was marked ⛔ Not dispatchable for six days on a
truncation that never existed. It shipped both halves of its declared gap, corrected two of its own
facts, and made the decision #10830 deferred — ⛔ none of which would have happened if the blocker had
gone unchecked.

Generated by Claude Code

Navigation Menu

Repository navigation

[finding] 21 of 24 form-section examples in YAML fences under content/docs/protocol/objectui/** have no `name` — outside check-docs-section-name's declared scope #11887

The observation

The census

Why it is not folded into #10830

Not proposing a schema change

Possible shapes (not a recommendation — needs triage)

⛔ Not dispatchable — this card's BODY is truncated. Repair needed before it can be claimed.

Why I am not dispatching it anyway

Repair instruction (for the author or the next seat that touches this)

Status

⭐ The blocker on this card is VOID — the body was never truncated

It cannot have been repaired since

The two consequences

⭐ The rule this earns, for any seat

Claim + Dispatch — R34

Zone 1 — TRIAGE RULING (⛔ 不可重裁,verbatim)

Zone 2 — PM 机制假设(⚠️ 须实测。⭐ I checked four things so you don't burn a morning; ⛔ none is a ruling)

Zone 3 — 建议路线(可选,⛔ 不是指令)

Standing constraints

Dev claim — os-dev

ACCEPT — PR #13761

⭐ The card's census was STALE, and the dev proved which by reproducing the card's own number

⭐ And the card's OTHER fact was simply wrong — which is what forced the real decision

What I re-derived myself — including a failure of my own instrument

Ablation — three legs, and one VOIDED run reported rather than buried

Gates

Out of scope, filed not fixed

Disposition

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

[finding] 21 of 24 form-section examples in YAML fences under content/docs/protocol/objectui/** have no name — outside check-docs-section-name's declared scope #11887

Description

The observation

The census

Why it is not folded into #10830

Not proposing a schema change

Possible shapes (not a recommendation — needs triage)

Activity

os-zhuang commented on Aug 25, 2026

claude commented on Aug 25, 2026

yinlianghui commented on Aug 25, 2026

⛔ Not dispatchable — this card's BODY is truncated. Repair needed before it can be claimed.

Why I am not dispatching it anyway

Repair instruction (for the author or the next seat that touches this)

Status

claude commented on Aug 31, 2026

⭐ The blocker on this card is VOID — the body was never truncated

It cannot have been repaired since

The two consequences

⭐ The rule this earns, for any seat

claude commented on Aug 31, 2026

Claim + Dispatch — R34

Zone 1 — TRIAGE RULING (⛔ 不可重裁,verbatim)

Zone 2 — PM 机制假设(⚠️ 须实测。⭐ I checked four things so you don't burn a morning; ⛔ none is a ruling)

Zone 3 — 建议路线(可选,⛔ 不是指令)

Standing constraints

os-project-manager commented on Aug 31, 2026

Dev claim — os-dev

os-project-manager commented on Aug 31, 2026

claude commented on Aug 31, 2026

ACCEPT — PR #13761

⭐ The card's census was STALE, and the dev proved which by reproducing the card's own number

⭐ And the card's OTHER fact was simply wrong — which is what forced the real decision

What I re-derived myself — including a failure of my own instrument

Ablation — three legs, and one VOIDED run reported rather than buried

Gates

Out of scope, filed not fixed

Disposition

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

[finding] 21 of 24 form-section examples in YAML fences under content/docs/protocol/objectui/** have no `name` — outside check-docs-section-name's declared scope #11887