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

[os-steve]

Observation only, filed unassigned while working #4636 / PR #4639. Nothing is broken today;
this is about a label that reads as a mechanism to the next agent and is not one.

What is there

skip-changeset exists as a repository label, and it is currently applied to PR #4639.

Why that is a problem

The repo's own pipeline-doc test states that this label was never real — it is named in the
list of phantom mechanisms that a deleted second workflow inventory had drifted into
documenting:

scripts/__tests__/ci-cd-pipeline-doc.test.ts:180-186

objectui#3724: this page was not the only workflow inventory. `.github/WORKFLOWS.md`
held a second one — hand-maintained, linked from nowhere, and pinned by nothing —
which had drifted to documenting 5 workflows that did not exist (including a
changeset gate skippable with a `skip-changeset` label; neither the workflow nor the
label was ever real) ...

And no workflow or script reads it:

grep -rn "skip-changeset" .github/ scripts/
  -> scripts/__tests__/ci-cd-pipeline-doc.test.ts:184     (the prose above, only)

So the label is inert. The changeset gates that actually run — check-changeset-presence,
check-changeset-fixed, check-changeset-no-major — have no skip path keyed on it.

The label's own metadata is consistent with it having been minted by an API call that
applied it, rather than deliberately created: default grey ededed, empty description,
unlike the repo's curated labels.

The harm is the ordinary phantom-mechanism one, and it has already half-happened: a label
named after a documented-but-nonexistent gate now exists, so the next agent that greps for
"how do I declare this PR publishes nothing" finds a label, applies it, and believes the
declaration landed. On PR #4639 the label and the body already disagree — that PR's body
explicitly says "The skip-changeset convention was not applied because this diff is not
tooling-only", while the label sits on it.

Options

• A — delete the repository label. Restores the state objectui#3724 established.
  Anything currently carrying it loses nothing, because nothing reads it. Cheapest and it
  matches the ruling that deletion, not a second ratchet, was the resolution for this exact
  phantom.
• B — make it real. Only if someone actually wants a labelled skip path on the changeset
  gate — which .github/WORKFLOWS.md documents 5 workflows that do not exist and omits 9 that do — including a changeset gate and a skip-changeset label neither of which is real #3724 declined once already, and which would weaken a gate that currently has
  no bypass.
• C — leave it. The label keeps reading as a mechanism to every future agent.

A looks right, plus optionally a line in ci-cd-pipeline-doc.test.ts's existing
phantom-mechanism family asserting the label's absence, so it cannot be silently re-minted the
next time someone applies it by name. I did not act on any of this: I did not remove the label
from PR #4639 either, since I did not put it there.

Deliberately not pm:queue — this is an observation, not a queued defect.

---

Generated by Claude Code