Scaffold a multi-agent AI configuration for any repo — any language, any framework.

Policy, not config: to control what these tools are allowed to do once configured (protected paths, dangerous commands, a merge-time backstop), use /platform-skills:ai-governance.

Reference index: references/setup-agents.md — read this first. It routes to mode-specific files.

Modes

• generate — fresh setup: ranked scan, interview, roster decision, generate all files
• upgrade — re-scan, read metadata from last session, git-diff, patch what's stale
• add — add one new agent without re-running the full interview
• review — quality check on 6 dimensions, output options at end

Interactive wizard (no args)

What do you need?
  generate — fresh setup
  upgrade  — patch what's stale since last update
  add      — add one new agent
  review   — quality check, no writes

Then load the appropriate reference file and follow its steps exactly.

Closing

After any mode: log non-obvious findings with /platform-skills:self-improve log.