An adversarial autonomous coding harness (GAN + anvil) — plan → generate ↔ evaluate, sprint by sprint. A Claude Code plugin.
Agent
Backend QA evaluator. Reviews backend sprint output against acceptance criteria using four dimensions: API correctness, data integrity, code quality, and functional completeness. Runs tests and exercises APIs directly. Use after a backend generator sprint completes.
Agent
Backend coding agent. Implements backend sprints from the product spec, one feature at a time. Focuses on API correctness, data integrity, test coverage, and robust error handling. Use for backend implementation sprints.
Agent
Frontend QA evaluator. Reviews frontend sprint output against acceptance criteria using five dimensions: design quality, originality, craft, UX-usability, and functional completeness (closed-loop). Interacts with the running app via browser automation to test real user flows and verifies feature loops close…
Agent
Frontend coding agent. Implements frontend sprints from the product spec, one feature at a time. Produces polished, distinctive UI code that avoids generic AI aesthetics. Use for frontend implementation sprints.
Agent
Product specification planner. Takes a brief user description (1-4 sentences) and expands it into a complete product spec with Sprint decomposition. Classifies the project as frontend/backend/fullstack, tags features by priority (P0/P1/P2), and emits a sprint dependency DAG. Use when starting a new ganvil build or…
Agent
A planning agent for rebuilding interfaces from screenshots and written specifications.