Spec-driven development toolkit for AI coding agents. Works with any AI agent that supports skills or custom instructions — opencode, Cursor, Claude Code, Copilot, Windsurf, Aider, and others.
Agent Claude Code
TDD Green Phase agent. Implements the minimum code needed to make failing tests pass. Follows language-idiomatic patterns and standard error handling. Does NOT modify tests.
Agent Claude Code
TDD Red Phase agent. Reads feature specs from the specs/ directory and generates failing tests that capture the expected behavior before any implementation exists. One test at a time — verify it fails — then hand off to the Green Phase agent.
Agent Claude Code
TDD Refactor Phase agent. Improves code quality, applies language idioms, enforces security best practices, and removes duplication — all while keeping every test green. Runs linters and verifies tests after each change.