learn

name	learn
description	Manually trigger session evaluation and pattern extraction.

Run OmG learn.

Task: $ARGUMENTS

Protocol:

1. Load session history from the transcript.
2. Identify high-signal reusable patterns across the entire session:
   • Common error resolutions (why it failed, how it was fixed)
   • Reusable workarounds for environment/tool quirks
   • Project-specific conventions, styles, or patterns identified by the user
   • User-corrected agent behaviors (avoiding the same mistake)
3. DO NOT save simple one-time fixes or conversational fluff. Focus on patterns that repeat or are likely to recur.
4. Interactive Selection Phase:
   • List the identified patterns with IDs (e.g., 1, 2, 3...).
   • Ask the user: "Would you like to save ALL of these, or just specific ones (e.g., '1, 3')? (Default is ALL)"
   • If the user responds "all" (or just presses enter/affirms), proceed with all.
   • If the user provides specific IDs, only save those.
5. Extraction & Saving:
   • For each selected pattern, extract it into a markdown file with frontmatter:
     • name: unique identifier
     • description: what this pattern covers
     • globs: file patterns where this applies (optional)
   • Save to .omg/rules/learned/ (or configured learned_skills_path).
6. Update MEMORY.md if a pattern is project-critical.

Configuration:

• Respect rules from .omg/rules/learn.json if available.

Response:

• Confirm which patterns were saved.
• Provide file paths for the new rules.
• Suggest how these rules will be applied in future sessions.