What problem does it solve? Complex multi-feature projects are slow and error-prone when handled by a single agent, and single-model code review has blind spots that let real bugs ship. This Skill provides a structured workflow for decomposing work into parallel agent waves and enforcing rigorous multi-round review before merging. ## Core Features & Use Cases - Wave Execution: Decompose features into independent sub-tasks and run 2-4 agents per wave with worktree isolation, committing after each milestone. - Iterative Review Loop (Ralph Loop): Submit code to an external LLM with rotating review lenses (security, UX, correctness, equivalence, performance) until consecutive approvals are reached. - Adversarial Dual-Review (Santa Method): Require two independent reviewers, ideally different models, to both approve before merging code that ships without human review. - Use Case: When building an auth module, spawn three parallel agents for the implementation, then run security and correctness review rounds via repomix piped to GPT or Gemini until two consecutive approvals gate the merge. ## Quick Start Ask the agent to decompose the current feature into independent sub-tasks and run them as a parallel agent swarm with an adversarial review loop after each wave.
