continuous-learning

You can extract reusable patterns from Gemini CLI session transcripts by automating continuous learning workflows, which detect error resolutions, workarounds, and debugging techniques to store as learned skills.

Continuous learning is the process of evaluating completed Gemini CLI session transcripts to identify repeatable patterns, such as user corrections or project-specific conventions, and persisting them for later skill reuse.

You automate continuous learning by adding a Stop hook in your .gemini/settings.json that runs an evaluation script, triggering pattern extraction when a Gemini CLI session ends.

Yes, jq is required as a dependency to parse and process session transcript data for configurable pattern detection and to store the extracted patterns in the learned-skills directory.

Yes, the evaluation script applies configurable pattern detection and allows you to specify an output directory to store the extracted reusable skills from your session transcripts.

The continuous learning script applies transcript message-count gating, requiring a minimum session length before evaluating the transcript, which prevents pattern extraction from overly short sessions.