continuous-learning

Automatically extracting debugging patterns from AI coding sessions requires evaluating transcripts via a stop hook to identify error resolutions, user corrections, and workarounds, then saving them as reusable skills.

Continuous learning for code sessions analyzes session transcripts to extract reusable patterns and project-specific conventions. It identifies error resolutions and debugging techniques, automatically converting them into new learned skills without manual intervention.

You can configure extraction sensitivity and ignored patterns for session learning by setting detection thresholds, defining session length minimums, and specifying pattern types to detect or ignore before the stop hook evaluates the transcript.

To save project-specific conventions and workarounds as reusable skills, configure the session evaluation stop hook to analyze your transcripts and automatically generate new skill files in a designated directory from the extracted solutions.

Continuous learning requires a minimum session length before extracting patterns, which you define by configuring session message thresholds. The quick start configuration demonstrates setting this threshold to sessions longer than 15 messages.

Limitations of automatic pattern extraction include reliance on a stop hook for session evaluation, meaning transcripts shorter than your configured message threshold are ignored, and unwanted patterns require manual configuration to filter out.