continuous-learning

To extract reusable patterns from Claude Code sessions, run the shell-based stop-hook evaluator to analyze session transcripts and save detected patterns to the local learned-skills directory.

Continuous learning in Claude Code works by evaluating completed session transcripts to detect actionable patterns like error handling and debugging techniques, then storing them as learned skills for future reuse.

To save learned skills locally, configure the shell-based evaluator as a stop-hook with a minimal config.json file, and it will automatically direct extracted patterns to the ~/.claude/skills/learned/ directory.

No external dependencies are required to run the session evaluation stop-hook; it only needs a minimal config.json file and a shell-based evaluator to analyze transcripts and store patterns.

Pattern extraction triggers during post-session evaluation, where the evaluator analyzes transcript length to determine when to identify and extract actionable workflow improvements and debugging techniques.

This session evaluation approach is specifically designed for Claude Code transcripts and stop-hooks, meaning it may not directly support other AI coding assistants without significant modification.