continuous-learning

You can extract reusable patterns from Claude Code sessions automatically by using a Stop hook that runs at session end, evaluating interactions and saving identified patterns as skills under ~/.claude/skills/learned/.

Continuous learning in Claude Code is the process of persisting session memory across interactions. It works by evaluating session length and extracting patterns to update learned skills, allowing past interactions to inform future tasks in different projects.

You can configure extraction thresholds for Claude Code session patterns by editing the config.json file. This configuration allows you to customize parameters like min_session_length and extraction_threshold to control when and how aggressively patterns are extracted.

Yes, the Claude Code pattern extraction Stop hook supports custom learned skills paths. You can specify the learned_skills_path parameter in your config.json file to define where the extracted skills are stored on your local machine.

Your Claude Code session patterns may not be saving because the session length did not meet the configured min_session_length threshold, or the extraction_threshold value is set too high for the current session's pattern density.