learn

Extract reusable patterns from the session, self-evaluate quality before saving, and determine the right save location (Global vs Project).

Apply the dog:learner skill to extract, evaluate, and save patterns from this session.

The skill will:

  1. Review the session for extractable patterns: error resolutions, debugging techniques, workarounds, project-specific conventions
  2. Self-evaluate quality of each candidate:
    • Is it specific enough to be actionable?
    • Is it general enough to apply in future sessions?
    • Does it already exist in skills/?
  3. Determine save location:
    • Global (~/.claude/skills/learned/): patterns usable across 2+ projects
    • Project (.claude/skills/learned/): project-specific knowledge
    • When in doubt: Global (easier to move inward than outward)
  4. Write skill file with: name, description, user-invocable: false, origin: auto-extracted, problem statement, solution with code examples, when to use
  5. Report: patterns extracted, save locations, files written

For complex sessions with 3+ agents: also run pattern-extractor agent to surface deeper architectural insights.

Additionally: after saving the skill file, append a JSONL entry to ~/.dog/projects/{slug}/learnings.jsonl using the learnings skill. This makes the learning searchable via /dog:learnings.

Entry schema:

{"ts": "<ISO8601>", "tags": ["<keyword>", ...], "title": "<first line>", "body": "<full description>", "files": ["<file>", ...], "commit_sha": "<HEAD>"}

Use /dog:learnings <query> to search across saved entries.