continuous-learning

You can extract reusable patterns by configuring a Stop hook to evaluate completed Claude Code session transcripts, parse them for error resolutions and workarounds, and persist the detected patterns to a local learned skills directory.

Continuous learning in Claude Code debugging workflows is the automated process of identifying and saving hard-won fixes and project-specific conventions from session transcripts to reduce repeated effort in future development tasks.

No, pattern extraction does not add per-message latency because the session evaluator runs at the end of each Claude Code session through a Stop hook invocation, processing the transcript after your work is complete.

Yes, you can control what gets saved by configuring settings like minimum session length thresholds, specific pattern categories, and ignore lists to filter the extracted error resolutions and workarounds before persistence.

Yes, you need jq installed as a dependency to parse Claude Code session transcripts and extract reusable debugging patterns and project-specific conventions during the Stop hook evaluation process.

Stop-hook based pattern detection requires a minimum session length threshold to trigger, meaning short interactions are ignored, and extracted patterns are stored locally requiring manual review and curation for future activation.