kaizen

What problem does it solve? Coding agents repeat the same mistakes across sessions because session failures, corrections, and errors are never systematically captured or fed back into project rules, hooks, and documentation. ## Core Features & Use Cases - Session Learning Extraction: Scans Claude Code, Codex, and Copilot session transcripts for user corrections, repeated tool errors, and repeated edits, then performs root-cause analysis (minimum 3 levels of why, KEDB matching against existing notes, cross-scope checks) before saving learnings to .kaizen/. - Learning Application & Lifecycle: Groups pending learnings by root cause and applies them to deterministic mechanisms (linters, pre-commit hooks, scripts) or rules, skills, docs, and upstream issue reports, with status tracking validated by a lifecycle checker. - Automated Hook Pipeline: A Stop hook records unextracted-activity sentinels, a PreToolUse gate blocks git commit until extraction is verified, and a SessionStart hook injects pending learnings into agent context. - Use Case: After an agent repeatedly fails a command in a session, run the extract flow to record the root cause in .kaizen/, then apply it as a pre-commit check so the failure cannot recur. ## Quick Start Ask the agent to reflect on this session and extract the most important learning into the .kaizen directory.