REVIEW.md sets the review level of a PR from the files it touches
(routing and auth, write surfaces, image, workstation agents, CI and
guards), the rules of each zone with their source, the verification
commands and the review pitfalls. scripts/review-levels.py replays the
levels on the 31 PRs that reached main and checks, offline, that its
patterns match the Haut table.
CLAUDE.md said "no auto-deploy": since 2026-10-12 a push to main
redeploys through the Forgejo webhook. The Deploy section now says so
and points to REVIEW.md.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
