What problem does it solve?
Coordinates multiple AI teammates to preserve independent analysis, avoid contamination of findings, and produce higher-confidence reviews or parallel implementations when single-session approaches are insufficient. It prevents anchoring and groupthink by enforcing independent phases, defines task sizing and assignment to avoid merge conflicts, and provides timeout and failure handling so reviews complete even when teammates fail or go idle.
Core Features & Use Cases
- Adversarial review protocol: Independent analysis, share findings, challenge, and synthesize to separate consensus, disputed, and unique findings with confidence levels.
- Implementation team orchestration: Create task lists, assign non-overlapping file sets, monitor progress, and run final verification.
- Teammate health and fallback handling: Timeout, failure, and idle handling with partial-result collection and single-session fallbacks.
- Use Cases: Security reviews of PR diffs with multiple facets, large feature implementations across modules, and debugging workflows with competing hypotheses.
Quick Start
Ask the agent to spawn an adversarial three-person review team to independently analyze the PR diff, collect findings, run challenge rounds, and synthesize a prioritized review summary.