Agent
Independent break-it review for high-stakes changes (auth, money, data, security, privacy, hard-to-undo). Reads source fresh and hunts the traps self-review misses. Use when /impl flags a change high-stakes, regardless of file count.
Lean, opinionated Claude Code plugin. Confirm-only-when-ambiguous workflow. Replaces obra/superpowers.
Agent
Independent break-it review for high-stakes changes (auth, money, data, security, privacy, hard-to-undo). Reads source fresh and hunts the traps self-review misses. Use when /impl flags a change high-stakes, regardless of file count.
Agent
Thorough code review for risky or architectural changes. Use when /impl touches >5 files, security-sensitive code, or shared infrastructure. Identifies real problems; does not nitpick style.
Agent
Deep bug investigation when validator fails or behavior is unexplained. Forms hypotheses, runs minimal experiments to verify, identifies root cause. Use when fast-impl's first attempt didn't work and the cause is not obvious.
Agent
Execute well-defined implementation tasks quickly and without deliberation. No architectural judgment, no scope expansion.
Agent
Run quality gates after implementation. Typecheck, tests, lint. Verify requirements are met and check for obvious issues. Fast and mechanical. Use after fast-impl completes work, before any code review.