continuous-learning

To extract reusable patterns from OpenCode session transcripts, run the stop-hook at session end to surface recurring actions and convert them into learned skills. The process identifies repeatable patterns and stores them as SKILL.md files for future reuse.

The best way to automatically save learned skills in OpenCode is using a session-stop hook that analyzes transcripts and writes extracted patterns directly to the ~/.config/opencode/skills directory, maintaining a centralized index for traceability.

Yes, you can configure extraction thresholds, keywords, and auto-approval settings to tune the pattern extraction flow. This allows you to manage output volume and ensure only significant recurring patterns are converted into learned skills.

Deduplication works by checking newly identified recurring patterns against existing entries in the centralized continuous-learning index. This prevents redundant learned skills from being written to the ~/.config/opencode/skills directory during pattern extraction.

Extracted OpenCode session patterns are stored as SKILL.md files under ~/.config/opencode/skills/<descriptive-slug>/, while metadata is tracked in a centralized learned/.continuous-learning-index.json file to enable knowledge reuse and traceability.