continuous-learning

To extract reusable patterns from conversation transcripts automatically, configure a Stop hook to evaluate session length and extraction thresholds, then save identified patterns like error resolutions and debugging techniques into skills/learned/.

Conversation patterns saved as learned skills include error_resolution, user_corrections, workarounds, debugging_techniques, and project_specific patterns, which are extracted based on configured thresholds and stored for future reuse.

Yes, you need jq installed to process the JSON data required for extracting learned skills from session transcripts, as it handles parsing the transcript available via the CLAUDE_TRANSCRIPT_PATH environment variable.

You configure minimum session length and extraction thresholds by defining min_session_length and extraction_threshold parameters in a config.json file, which guides the evaluation behavior when the Stop hook triggers.

You should use automated pattern extraction on conversation analysis when you want to capture debugging techniques and workarounds at the end of each session, provided the conversation meets the defined minimum session length criteria.

Extracted skills may not save to the learned skills path if the conversation length is below the min_session_length, the pattern score fails the extraction_threshold, or the CLAUDE_TRANSCRIPT_PATH is not properly configured.