Skip to content
GitLab

chore: review process, shared claude config and command guard

Merged Khanh Pham requested to merge chore/review-process into develop

What

Review process and shared Claude Code config for Ahad.

  • REVIEW.md: what reviewers look for, finding format BLOCKING / SHOULD-FIX / NIT / PRE-EXISTING, verdict line, three-round cap. Leads review MRs with /review-mr <iid>.
  • .gitlab/CODEOWNERS (risk tiers by path, CE ignores it for approvals) and .gitlab/merge_request_templates/Default.md.
  • Track .claude/ (agents, commands, rules, skills, hooks, settings.json) and every CLAUDE.md. Per-machine files stay ignored (settings.local.json, worktrees/, RESUME.md); defaultMode: bypassPermissions moved to local settings so nobody inherits it.
  • .claude/hooks/guard.mjs (shared, identical across Holatech repos): PreToolUse guard blocking untracked .env/key writes, committed-migration edits, secrets in content, synchronize: true, push to main/master/develop, force push, --no-verify, volume-wiping docker commands, reset --hard/clean -f. Per-repo config in guard.config.json.
  • moanapay-review is now a checklist read by /review-mr; the format lives in REVIEW.md.

Risk

Low: docs, tooling and agent config only. No app code, no migrations.

Test evidence

  • node --test .claude/hooks/*.test.mjs → 5 pass, 0 fail.
  • Guard dry run: git push origin develop → blocked (exit 2); pushing this branch → allowed (exit 0).
  • Secret scan over all newly tracked files: no keys, tokens or passwords (only the guard's own regexes match). Skills that hold real credentials are kept out of git locally.

🤖 Generated with Claude Code