docs(ci): stop asserting the skip-changeset label object exists

[baozhoutao]

Fixes #4912

The skip-changeset repository label has been deleted. This PR repairs the two sentences that the deletion made false.

The label is gone

DELETE /repos/objectstack-ai/objectui/labels/skip-changeset returned HTTP 204; a follow-up GET returns 404, while GET .../labels/domain:devx still returns 200 — so the 404 is a reading, not a broken route.

⭐ This falsifies a premise carried on the card since 2026-08-25. Two earlier seats measured that no agent seat could perform this act (the MCP toolset exposes get_label only, and direct REST returned 403 with "GitHub access is not enabled for this session"), and the card sat in pm:awaiting-maintainer on that basis. The MCP half still holds — there is still no label create/update/delete tool. The REST half has changed: repo-scoped REST is reachable from this container now, and the delete went through. GITHUB_TOKEN is still the literal 14-byte string proxy-injected; the real credential is proxy-injected.

Triage's precondition, re-verified immediately before deleting

Triage required, verbatim: 若是删标签,先确认没有任何 PR 正挂着它。

reading	value	firing control taken on the same call
label exists	yes (ededed, empty description)	—
open PRs carrying it	0 at 17:22:44Z	same endpoint with labels=domain:devx returned 3
workflow files reading it	0 under .github/	same grep for changeset hits 5+ workflow files

⚠️ The GitHub search API is 403 through this proxy and returns a body with no total_count. Read without capturing the HTTP status that is a silent zero — exactly the false negative that would have made this precondition look satisfied for the wrong reason. The counts above come from the repo-scoped /issues?labels= endpoint, which is 200.

Seven closed PRs carried the label (#4700, #4665, #4639, #4630, #4628, #4615, #4129). Nothing read it on any of them.

I also checked for a consumer outside this tree, since the objectstack sibling wires the name in lint.yml, pr-automation.yml and check-empty-changeset.mjs. Every one of those reads its own repo's event payload; none names objectstack-ai/objectui in a labels call. No cross-repo consumer.

Why there is a diff at all

content/docs/guide/ci-cd-pipeline.md asserted a live fact about the label:

That reading has since expired — a skip-changeset label now exists in this repository's label set

and pointed forward to "#4912 tracks deleting the object". The pin's failure message likewise asserted "The label object exists in this repository". All three are false once the object is gone.

⛔ I did not replace them with the mirror-image claim that the label is gone, and the page explains why in its own words: a point-in-time labels-API reading is not a fact this repository can keep true. That lesson is already mechanised — the third assertion in the pin is not.toContain('checked against the labels API'). Absence is even less durable than presence was: GitHub mints a label the first time one is applied by name, and the instruction to apply it still reaches agents from the objectstack sibling (scripts/pm/dispatch-gates.mjs still emits it; root cause tracked at objectstack#17017). One API call re-mints it.

So the minting and the deletion are both written as dated history, and the present-tense guidance is made independent of whether the object exists — consistent with the sentence already on the page, "Whether or not you can still see it in the picker, applying it declares nothing".

A note in the test records that the deletion does not retire the two pins, so a later reader does not remove them as obsolete.

The three pins were never in danger

They assert tree facts, not label-object facts, so none went stale. Both strings they require are still present and the forbidden one is still absent — verified with the test's own normalization (blockquote markers stripped, whitespace collapsed), because these sentences wrap across lines and a raw substring grep reports 0 for all of them. That vacuity trap is documented in the test and I walked into it once before checking properly.

Evidence

All on final commit c88617f79, clean tree, worktree from origin/main = 42ddd7f71. Exit codes captured by redirect before any pipe.

• vitest run scripts/__tests__/ci-cd-pipeline-doc.test.ts --reporter=verbose → exit 0, Tests 58 passed (58). Both A skip-changeset label exists on this repo, but no workflow reads it — the repo's own test records that neither the workflow nor the label was ever real #4912 cases confirmed collected by name, not inferred from a total: ✓ ... never wires the phantom skip-changeset label into a workflow or a gate and ✓ ... keeps the page denying skip-changeset rather than describing it.
• Narrowed union vitest run scripts/ → VERDICT command-exit 0, Test Files 132 passed | 2 skipped (134), Tests 3833 passed (3835). Narrowed to scripts/ because that is where every real reader of the changed page lives — git grep -l ci-cd-pipeline over source returns 20 files and every one that actually reads the file is under scripts/__tests__/.
• type-check:scripts → exit 0, pnpm echoed tsc -p tsconfig.scripts.json as the positive control against a zero-match no-op.
• lint:root unnarrowed → exit 0, 32 problems (0 errors, 32 warnings), none in either changed file.
• check:control-bytes → exit 0, scanned 7066 tracked text file(s), plus an independent grep -naP control-byte self-scan over both changed files (no hits).
• docs:check-links → exit 0, Links are valid across 17 scan roots.
• Heavy runs went through the shared verify lock; verdicts read from its VERDICT line, never a bare $?.

Reverse verification, run from the committed state so restore is a git checkout HEAD --, each leg carrying a trap on EXIT/INT/TERM with absolute paths, mutation confirmed on disk by counting anchored text rather than by an editor's exit status:

1. Broke the pinned denial (reads it → consults it): anchor count 1 → 0, injected 1 → suite exit 1, failing with this PR's rewritten message. Restore: anchor back to 1, git diff HEAD empty, disk blob 9d0447795 identical to the HEAD blob.
2. Put a point-in-time reading back ("The label does not exist (checked against the labels API, 2026-09-09)"): phrase count 0 → 1 → suite exit 1 on do not put a point-in-time labels-API reading back on this page. ⭐ This is the load-bearing one: it shows the obvious fix would have gone red, which is why the correction is shaped the way it is. Restore verified the same way.

No build or dist/ is involved and I am saying so rather than omitting it: vitest transforms these tests from source, so there is no stale artifact for the ablation to hide behind.

Housekeeping

• Changeset: none. node scripts/check-changeset-presence.mjs → exit 0, verdict line: "No source or published contract of a released package changed in this range, so no changeset is owed." A docs page and a root test file; neither is published source of any package.
• Not governed surface: node scripts/check-governed-queue-guard.mjs --test on both paths → NOT GOVERNED — 2 path(s) checked against 5 governed surface(s); none matched, with AGENTS.md as the firing control. Left as a draft per dispatch regardless.
• ⛔ No skip-changeset label on this PR — on this card of all cards, and the object no longer exists.
• ⛔ I removed the label from none of the seven closed PRs before deleting it, and opened none of them.

🤖 Generated with Claude Code

https://claude.ai/code/session_01FhBNJcLRZLe8M87VcUgpKr

---

Generated by Claude Code

[claude]

⭐ LANDED — merged 256c709e2 at 2026-09-09T18:07:32Z

Probe run verbatim from the Armed note (5606302512); full reading on the card: #4912 (comment)

All six legs pass. The label reads 404 with domain:devx → 200 as the firing control; the live-fact sentence is gone (post 0 / pre 1) while skip-changeset still appears 3× in the page, so it was corrected rather than blanket-deleted; :319's .not.toContain('checked against the labels API') — the pin that reddens the obvious mirror-image fix — is intact; and both #4912 cases are still collected by name.

⭐ The replacement text is dated history (:1920, "by 2026-08-25 a skip-changeset label sat on seven pull requests…"), which is what survives a re-mint. ⚠️ And a re-mint is live: the objectstack sibling still instructs agents to apply the label by name, which is how the deleted object came to exist. Root cause is objectstack#17017.

⛔ One of my probe controls was mis-specified again (the pattern omitted the source's backticks) and was repaired before anything was read off it — the fourth such miss today, all the same shape: a pattern written from what the prose says rather than what the file contains.

---

Generated by Claude Code