continuous-learning

You can extract reusable patterns from Claude Code session transcripts by configuring a Stop hook to run an evaluation script at the end of each session, automatically identifying error resolutions and debugging techniques to save as learned skills.

Continuous learning automates pattern detection by evaluating your session transcript via CLAUDE_TRANSCRIPT_PATH when a session ends, extracting project-specific conventions and saving them as reusable skills to a configured directory.

Yes, you need jq installed as a dependency to parse the session transcript and extract learned skills from your Claude Code sessions, alongside setting up a Stop hook and configuring your learned-skills path.

To configure a Stop hook for saving error resolutions, point it to the evaluate-session.sh script, ensure CLAUDE_TRANSCRIPT_PATH is set, and edit config.json to specify the minimum session length and patterns to detect.

The pattern detection mechanism saves debugging techniques, error resolutions, user corrections, and project-specific conventions from session transcripts, storing them as learned skills in your configured ~/.claude/skills/learned/ directory.

Yes, you can set a minimum session length for transcript evaluation by editing the config.json file, ensuring that only sufficiently long Claude Code sessions are processed for pattern detection and skill extraction.