continuous-learning

You can extract reusable patterns from Claude Code session transcripts by configuring a Stop hook that evaluates session-end transcripts, filters for error resolutions and debugging techniques, and saves them as learned skills.

Session-end evaluation for continuous learning automatically inspects completed coding session transcripts to identify error resolutions, user corrections, and workarounds, then persists these patterns as reusable skills for future tasks.

To set up a Stop hook for saving debugging techniques, add the hook entry to your root SKILL.md with YAML frontmatter, configure the transcript path processing, set minimum session length thresholds, and specify a learned skills output directory.

Yes, you can filter extracted session patterns using configurable thresholds including minimum session length, extraction thresholds, and allow or ignore pattern lists to ensure only meaningful debugging techniques and project conventions are saved.

Continuous learning pattern extraction is designed specifically for repeated coding tasks across sessions, automatically capturing project-specific conventions and workarounds so you avoid rediscovering the same solutions in future Claude Code runs.

Automatic pattern extraction from session transcripts requires a root SKILL.md with YAML frontmatter and depends on session length meeting configured thresholds; sessions below the minimum length or lacking recognizable patterns will not produce learned skill outputs.