docs(scripts): the version-gate roster names the two repositories that carry it

[drewstone]

The header claimed these files are kept byte-identical in three repositories.
Two carry them.

agent-eval has never had check-version-bump.mjs, check-api-surface.mjs,
lib/api-surface.mjs or api-surface.json on main, and no CI step that runs
them. The only copy sits on feat/export-surface-gate, written 2026-08-18,
unmerged, and 59 commits behind main as of 2026-09-01. That branch's own copy of
check-version-bump.mjs had already drifted from this one before the
published-baseline fix existed, so the three-name roster was never true — not
since the day it was written.

The correction names exactly the two repositories that carry the file, and says
plainly why the third is absent, so nobody restores it on the strength of the
sentence alone.

The stale count in check-api-surface.mjs goes with it: "three copies of one
rule end up with three different bugs" was tied to the same roster. It reads
count-free now, which keeps the argument and drops the number.

Adopting the gate in agent-eval is a decision about that repository's release
discipline, not an edit this file is owed. It has an open release train
(0.173.0), and switching a new gate on mid-flight would red it. Tracked as its
own issue: tangle-network/agent-eval#729

Documentation only. No behavior changes.

Byte-identity

Verified by checksum after the edit — the same patch is applied to both
repositories, so check-version-bump.mjs, check-api-surface.mjs,
lib/api-surface.mjs and tests/version-bump-check.test.ts remain identical
across agent-runtime and agent-knowledge. The companion PR is tangle-network/agent-knowledge (opened next).

Gates

• node --check on both edited scripts — clean
• pnpm run lint — clean
• npx vitest run tests/version-bump-check.test.ts — 36 passed
• pnpm run check:version-bump — passes (no consumer surface change)

[tangletools]

✅ Auto-approved drewstone PR — a9b952f0

This PR was opened by the trusted drewstone account.
The full PR reviewer audit still runs separately and will publish findings if it detects issues.

This approval is provisional. It rests on the audit running. If the audit cannot run — for example the CLI bridge rejects it — this approval is dismissed rather than left standing, so an unrun check never reads as a passing one.

_{tangletools · auto-approval · reason: drewstone_author · 2026-09-01T23:57:44Z}