Two stat corrections plus a ref bump, together because release-plugin.mjs left a
half-state that could only be closed with the gate green.
repo-standard v0.5.0 -> v0.9.0 (operator-approved). Approved as "bump to the
latest tag, v0.8.1", but v0.8.1 became unreachable while the session ran:
plugin.json had moved to 0.9.0, and the helper refuses a target its plugin.json
does not match ("asked to release 0.8.1 — bump plugin.json first"). v0.9.0 was
tagged at 22:05 today (1dfd408), plugin.json == README badge == 0.9.0, so it is
a consistent released version and the faithful reading of the approval. Same
breaking v0.8.0 boundary that was already accepted, one minor further.
Ungated axes re-measured at the new ref, per CLAUDE.md: tests 135 -> 170
(`node --test` on a `git archive v0.9.0` extract), checks 14 -> 14 (the README
check table still has 14 rows; v0.6.0-v0.9.0 re-levelled finding CODES under
existing rows). Badges at v0.9.0 are version/platform/skills-1/license, so
both numbers stay ungated and rot silently — they were measured, not carried.
repo-mailbox 406 -> 402. This reverses the refusal sent from here 20 minutes ago
(and d60ad40's 406), and the reversal is legitimate rather than a flip-flop: 402
was wrong then and is right now because the REF MOVED underneath it. When
repo-mailbox asked, v0.21.0 did not exist as a tag and the catalog pinned
v0.20.3, whose badge says 406 — writing 402 then would have produced a hard
ERROR and a dangling ref. A parallel session has since bumped the ref to
v0.21.0 (9d04179), and the badge AT that ref says 402. The catalog documents
what installs, so the number follows the tag.
That parallel commit knowingly landed a red gate for the catalog to clean up;
this commit is that cleanup. Gate now: 12 OK, 0 WARN, 0 ERROR, 0 SKIP.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XPz4FNKSn2QwoMhRFn4WNo
Kjell Tore Guttormsen
22ef8c7e65