continuous-learning

You can automatically extract reusable patterns from Claude Code sessions by configuring a Stop hook. This hook evaluates the session upon ending, detects useful patterns, and saves them directly as learned skills to your local directory.

A Claude Code session needs a minimum of 10 messages by default to trigger pattern extraction. This threshold is configurable via a config.json file, allowing you to tailor when the session evaluation and skill harvesting process activates.

Yes, you can configure the output path for learned skills using a config.json file. By default, the extracted patterns are saved to the ~/.claude/skills/learned/ directory, but thresholds, paths, and patterns to detect are all adjustable.

Continuous learning in Claude Code workflows is achieved through a session hook mechanism. After a session ends, it checks for minimum messages, identifies extractable patterns, and writes them as learned skills for future use.

No external dependencies are required for the pattern extraction process. The skill operates independently within the Claude Code environment using a Stop hook and a configurable config.json file to manage thresholds and paths.

Session patterns may not be saved if the session lacks the minimum required messages, which defaults to 10. Check your config.json settings to ensure thresholds and the target output path are correctly configured for the Stop hook.