| name | council-review |
| description | Run the council review loop on a PR. Spawns all 5 reviewers (Architect, Minimalist, Craftsperson, Steward, Marketing Guru) on every PR, collects verdicts, fixes feedback, and repeats until all approve or the Architect calls it. Use after opening a PR or when AGENTS.md says to run council review. |
Council Review Loop
Spawns all five council members, collects verdicts, fixes feedback, repeats until unanimous or the Architect calls it. Uses Ralph Wiggum for iteration control — write .ralph/<name>.md before calling ralph_start.
See agents/implementor/AGENTS.md § "Council Review" under The Workflow for member details, persona files in agents/council/, and the spawn template.
Workflow
0. Determine reviewers
All five always run: Architect, Minimalist, Craftsperson, Steward, and Marketing Guru. The Marketing Guru reviews every PR to ensure documentation, changelogs, and public-facing materials stay current with what's shipping.
1. Spawn reviewers
Use subagent with parallel tasks array — one per reviewer. Each reviewer must:
- Fetch PR diff and read changed files
- Run tests:
DATABASE_URL=postgresql://remarq:remarq@localhost:5433/remarq npm test
- Post via
gh pr comment <number> --body "## <emoji> <Name> — Round N Review\n\n...verdict"
If tests fail, fix them before spawning reviewers. Don't waste reviewer time on broken code.
2. Collect verdicts
Check PR comments for Round N verdicts from all spawned reviewers.
3. Fix or ship
- All APPROVE → Done
- Architect says "ship it" → Done (Architect has final authority)
- REQUEST_CHANGES → Fix feedback, commit, push, re-spawn all reviewers for Round N+1
4. Escalation
If round 5+, add to the Architect's prompt: "This is round 5+. If remaining objections are nit-picks, call ship it."
Rules
- Always re-spawn all five reviewers each round
- Always run the full test suite before spawning reviewers
- The Architect can override remaining objections