docs(prompts): renumber ui-library prompt sections 1,3,4,5 to 1,2,3,4

[baozhoutao]

Fixes #9319

.github/prompts/ui-library.prompt.md numbered its sections 1, 3, 4, 5. Before renumbering anything, the card's real question had to be answered: was ## 2 deleted, or was it never written? Those have different correct repairs.

Answer: ## 2 was NEVER WRITTEN. Settling commit: 4e7737788

The file has exactly one commit in its entire history, and there it is a pure creation — new file mode, @@ -0,0 +1,66 @@, zero deletion lines. The headings read 1, 3, 4, 5 in that first commit and have never read anything else. So the gap records nothing absent, there is no section 2 to restore, and none is added.

This is the same finding, and the same repair, that objectui#9145 reached for this file family's letter runs: "It was not dropped... there is no C category to restore." The two sibling prompts born in that same commit are numbered contiguously (component.prompt.md 0-4, engine.prompt.md 1-7); only this file skipped — what a hand-numbering slip looks like.

Probe block — every count names its population; every zero carries a control that HITS

All legs read the document as text (the subject is markdown headings; a code-only reader would score this change as a no-op). Ref named per leg.

Leg 1 — was a ## 2 ever added or removed on this path, on any ref?

git log --all --oneline -S '## 2' --pickaxe-regex -- .github/prompts/ui-library.prompt.md   # 0 commits
git log --all --oneline -S '## 3' --pickaxe-regex -- .github/prompts/ui-library.prompt.md   # 1 commit  <- CONTROL HITS

Population: commits on any ref whose diff changes the count of that heading string in that path. ## 2 = 0; control ## 3 = 1 (4e7737788). The search would have found a deletion had one happened.

Leg 2 — is the file a condensation of one of the 17 prompt files that same commit deleted?

git grep -l 'Core Philosophy' 4e7737788b^ -- .github/prompts/   # 0 files
git grep -l 'Shadcn'          4e7737788b^ -- .github/prompts/   # 9 files  <- CONTROL HITS

Population: prompt files existing at the parent commit. No predecessor carried this file's content, so nothing was carried over with a section dropped in transit.

Leg 3 — the headings themselves, pre and post, both COUNTED (ref origin/main = fb4ec65daf, then worktree)

Population: lines matching ^## in .github/prompts/ui-library.prompt.md.

	pre	post
headings (matched lines printed in-run)	1. Core Philosophy / 3. Component Standards / 4. Development Workflow / 5. Visual Standards	1. Core Philosophy / 2. Component Standards / 3. Development Workflow / 4. Visual Standards
^## 2.	0	1
^## 5.	1	0
numbered list lines ^[0-9]+\. (must not move)	7	7

The equal leg (7 = 7) is the guard that the edit touched headings only; git diff is 3 insertions / 3 deletions, all ^## lines.

Leg 4 — does anything CITE these section numbers? Three populations, three controls, all zero:

grep -rn '§' .github/prompts/                    # 5 occurrences, NONE pointing at this file
grep -n '§E of `component.prompt.md`' .github/prompts/engine.prompt.md   # 1  <- CONTROL HITS

All five § cross-references in the prompt surface point at component.prompt.md (§2.A, §B, §E). This control matters because objectui#9145 had to move exactly such a reference (§F to §E) when it renumbered — so cross-refs demonstrably exist here.

git grep -c '3-component-standards' -- .   # 0 files  (and #4-development-workflow, #5-visual-standards, #1-core-philosophy: 0)
git grep -l '#development-workflow' -- .   # 2 files  <- CONTROL HITS

Population: tracked text containing a GitHub heading-anchor slug (slugs embed the number). Anchor links exist in this repo; none point at this file's sections.

git grep -in 'ui-library' -- . | grep -icE 'section|§|#[0-9]'   # 0

Population: tracked lines naming this file alongside a section reference.

Leg 5 — the machine readers. Per the repo's own registry scripts/markdown-test-inputs.mjs, 4 test suites read this file: check-prompt-component-keys, check-installed-spec-pin-claims, ci-cd-pipeline-doc, dollar-dialect-alias-census. Running the gate's scanner directly shows it parses this file's 4 ## N. sections and validates the ### A./B./C. run under Component Standards — but it uses the section number only as an error label, and objectui#9145 scoped it explicitly: "the ## N. numbers themselves are a different claim and are not judged." The gate's test suite cites ui-library 0 times (control: component.prompt 18 times in the same file). So the renumbering moves no assertion.

Checks run before pushing (exit codes captured before any pipe)

check	pre	post
check:prompt-keys	0	0 — 29 ### X. headings across 8 letter runs, all contiguous
check:control-bytes	0	0 — 7585 tracked text files
docs:check-links	—	0 — links valid across 17 scan roots
check:doc-fences	0	0
check-changeset-presence.mjs	0	0 — no changeset owed
the 4 test suites above	—	0 — 4 files, 187 tests passed

node scripts/check-governed-queue-guard.mjs --test .github/prompts/ui-library.prompt.md said: NOT GOVERNED — 1 path checked against 5 governed surfaces; none matched. An ordinary pull request. (Measured, not assumed.)

Notes for the reviewer

• The changeset uses empty frontmatter — the repo's first-class "no release" declaration (472 existing changesets use it), and check-changeset-presence.mjs confirms no changeset is owed. It exists to record the measurement durably.
• Not done, deliberately: the file's stray double blank line before the renumbered Component Standards heading is untouched (whitespace, not numbering), and no ## N. sequence check was added to check:prompt-keys. objectui#9145 pinned the letters so the gap could not reopen; the numbers remain unpinned by its own explicit scoping. Pinning them is a real follow-up but needs a maintainer decision (runs legitimately start at 0 in component.prompt.md and 1 in engine.prompt.md), so it is not a rider on this card.
• check:doc-links as named in the dispatch does not exist; the link gate is spelled docs:check-links. Ran that.

🤖 Generated with Claude Code

https://claude.ai/code/session_01FhBNJcLRZLe8M87VcUgpKr

---

Generated by Claude Code

[claude]

PM pre-merge probe — domain:devx @ objectui seat (#5748)

⚠️ Posted BEFORE the merge, deliberately. A probe published after landing is a report; published before, it is a prediction that can be falsified. Every value below was re-taken by this seat on the ACTUAL current base, ⛔ not copied from the dev's report and ⛔ not from the base this PR records.

⛔ ERRATA 62b — the base MOVED under this PR while it was in review

	sha
base this PR records	cb725e78fb
ACTUAL origin/main at probe time	136c8842d6
what moved it	136c8842d6 — test(app-shell): pin the permission door's draft unwrap against read decorations (#9419)

⭐ Measured, not assumed: git diff cb725e78fb 136c8842d6 -- .github/prompts/ui-library.prompt.md is EMPTY — the base advance does not touch the subject file, so the pre-values below are unchanged by it. git merge-tree against the new base yields 0 conflict markers.

⚠️ The final merge parent will be whatever the queue puts in front of it, which is ⛔ not knowable now. I re-confirm against the actual <merge>^ in the LANDED note.

The legs — every count names its population, every zero carries a control that HITS

Leg 1 — was a ## 2 heading ever added or removed on this path, on ANY ref?
Population: commits on any ref whose diff changes the count of that heading string in .github/prompts/ui-library.prompt.md.

• -S '## 2' → the ONLY commit is 40fcd145ff — this PR's own fix. ⛔ Not a falsification: the search now sees the branch because the branch is pushed. No pre-existing commit touches it.
• CONTROL -S '## 3' → 1 commit, 4e7737788b — HITS, so the search would have found a deletion had one happened.
• NONSENSE control -S '## 97' → 0.

Leg 2 — is the creating commit a pure add? 4e7737788b, the file's only pre-existing commit: new file mode 100644, hunk @@ -0,0 +1,66 @@, and deletion content lines = 0 (counted, printed). ⇒ the headings read 1, 3, 4, 5 at birth. Nothing was removed, so ## 2 was NEVER WRITTEN — it was not deleted. That is the whole question this card turned on, and it is what makes renumbering the correct repair rather than the evidence-destroying one.

Leg 3 — the headings, BOTH SIDES, MATCHED LINES PRINTED (pre = 136c8842d6, post = 40fcd145ff):

	pre	post
^## matched lines	1. Core Philosophy · 3. Component Standards · 4. Development Workflow · 5. Visual Standards	1. Core Philosophy · 2. Component Standards · 3. Development Workflow · 4. Visual Standards
^## 2\.	0	1
^## 5\.	1	0
NONSENSE ^## 9\.	0	—

⭐ The equal-on-both-sides leg carries its printed guard, per this seat's standing rule: numbered LIST items ^[0-9]+\. = 7 pre and 7 post. Equal — which is the point: it proves the edit moved headings only and left the file's numbered content untouched. (Guard is non-empty: 7, ⛔ not the empty-string hash e3b0c44298fc1c14.)

Diff is 3 insertions / 3 deletions, every one a ^## line.

CI on 40fcd145ff

34 check runs, non-green = NONE. All 9 required contexts at success, enumerated rather than summarised: Lint · Type Check · Build & E2E · Test (shard 1/4) · Test (shard 2/4) · Test (shard 3/4) · Test (shard 4/4) · Build Docs · Changeset Declaration.

Governed Surface Queue Guard: success — and it ran while this PR was still a draft. ⭐ That retires a wrong note this seat has been carrying: I had recorded the guard as added by the draft→ready flip. Measured across 4/4 open drafts in this lane (this PR, #9410, #9391, #9389), it is present and green on every one of them before any flip; nonsense control Governed Surface Queue Guardd absent. ⇒ the governed verdict is readable before the flip, not after. The PR is correctly NOT GOVERNED (.github/prompts/** matches none of the 5 governed surfaces).

What I did NOT verify

⛔ I did not re-run the repo's test suites locally; I read CI's verdict. ⛔ I did not independently re-derive the dev's leg-5 claim about which 4 test suites read this file from scripts/markdown-test-inputs.mjs — CI's green on those suites is the evidence I am relying on there, and it is weaker than a direct reading.

⇒ Flipping ready and enqueueing.

---

Generated by Claude Code

[claude]

✅ LANDED — 0a6372e3a9

Merged through the queue at 2026-09-13T19:05:53Z. Timeline, the actual instrument rather than any tool's echo: added_to_merge_queue 18:47:34Z → removed_from_merge_queue 19:05:52Z → merged 19:05:53Z.

⛔ ERRATA 62b — re-confirmed against the ACTUAL <merge>^, and this time it did NOT bite

merge commit : 0a6372e3a9  docs(prompts): renumber ui-library prompt sections 1,3,4,5 -> 1,2,3,4 (#9421)
<merge>^     : 63fb72c421  fix(types)!: retire the published `layout` from RecordDetailsComponentProps (#9416)

⚠️ The base moved THREE times while this PR was open — cb725e78fb (recorded on the PR) → 136c8842d6 (#9419) → 63fb72c421 (#9416, which jumped the queue ahead of this PR). The pre-values in the pre-merge probe #issuecomment-5655316248 were taken on 136c8842d6, and I re-took them again on 63fb72c421 before the merge.

⭐ <merge>^ is 63fb72c421 — the base of that final re-take. So the published pre-values are correct, ⛔ but only because they were re-taken after each advance; both earlier bases would have been the wrong parent to cite. This is the eleventh-and-twelfth firing of this erratum today, and the first time the discipline visibly earned its cost rather than caught an error after the fact.

The probe, re-run once more on <merge>^ = 63fb72c421 and on main as it now stands

	pre (63fb72c421)	post (0a6372e3a9, on main)
^## matched lines	1. Core Philosophy · 3. Component Standards · 4. Development Workflow · 5. Visual Standards	1. Core Philosophy · 2. Component Standards · 3. Development Workflow · 4. Visual Standards
^## 2\.	0	1
^## 5\.	1	0
guard — numbered LIST items ^[0-9]+\.	7	7
NONSENSE control ^## 9\.	—	0

The guard is equal on both sides by design and carries its printed matched lines above: it proves the edit moved headings only. The merge diff is .changeset/9319-… (+34) and the prompt file (+3/−3), every one of the three changed lines a ^## heading.

What this card actually settled

⭐ The visible defect was trivial — headings running 1, 3, 4, 5. The card's real question was 「was ## 2 DELETED, or NEVER WRITTEN?」, and those have different correct repairs: renumbering restores an intended sequence in the first case and destroys the evidence of something absent in the second.

Measured: 4e7737788b is the file's only pre-existing commit and is a pure creation — new file mode, @@ -0,0 +1,66 @@, 0 deletion lines. The headings read 1, 3, 4, 5 at birth. ⇒ NEVER WRITTEN. Nothing is absent, so renumbering erases no evidence and is the correct repair. Controls: pickaxe ## 3 → 1 commit (HITS); nonsense ## 97 → 0.

Nothing cites the numbers — measured three ways, each with a control that hit, and the load-bearing one is that objectui#9145 had to move a §F→§E reference when it renumbered this same file family, so such citations demonstrably exist on this surface and this sweep would have found them.

⭐ A note this seat owes the record, because it corrects ITSELF

Governed Surface Queue Guard ran green on this PR while it was still a DRAFT. I had been carrying a note that the draft→ready flip is what ADDS that check — that note was wrong and is retracted. Measured across 4/4 open drafts in this lane at the time (this PR, objectui#9410, objectui#9391, objectui#9389): present and green on every one, nonsense control Governed Surface Queue Guardd absent. ⇒ the governed verdict is readable before the flip. This PR was correctly NOT GOVERNED — .github/prompts/** matches none of the five governed surfaces.

Deliberately not done, and still open

The dev left two things undone on purpose and said so; ⛔ neither is a defect in this PR:

1. The file's stray double blank line at the old section-2 seam — whitespace, not numbering, and authoring residue rather than a deletion scar (the creating commit added it that way).
2. No ## N. sequence check was added to check:prompt-keys. objectui#9145 pinned the letter runs so that gap could not reopen and scoped the numbers out by name; this card is that unclosed half reopening. ⚠️ A sequence gate is a real candidate but needs a maintainer call, because legitimate runs start at 0 (component.prompt.md) and at 1 (engine.prompt.md), so 「contiguous from 1」 would red a correct file. ⛔ Not a rider on this card.

Also corrected in passing, ⛔ not a repo defect: this seat's dispatch named a gate check:doc-links that does not exist — the link gate is spelled docs:check-links. The dev ran the real one (exit 0) rather than reporting a gate they had not run.

---

Generated by Claude Code