Skip to content

Conversation

@adrien-hearthands
Copy link

Summary

  • Adds /scan-history skill that retrospectively analyzes Claude Code conversation history
  • Identifies learnings that weren't captured during original sessions
  • Presents candidates with confidence scores for user confirmation before creating skills
  • Supports --user flag to save skills at user-level instead of project-level

Closes #15

How it works

  1. Locates conversation history files (~/.claude/projects/<project>/*.jsonl)
  2. Parses JSONL and extracts user/assistant exchanges
  3. Identifies patterns indicating extractable knowledge (debugging sessions, error resolutions, workarounds)
  4. Scores and ranks candidates by reusability, non-triviality, specificity
  5. Presents summary table with candidates for user selection
  6. Creates skill files for confirmed selections

Test plan

  • Run /scan-history in a project with conversation history
  • Verify it finds and lists conversation files
  • Verify it presents candidate learnings with rationale
  • Confirm a candidate and verify skill file is created correctly
  • Test --user flag saves to ~/.claude/skills/

🤖 Generated with Claude Code

…history

Adds a new skill that retrospectively analyzes Claude Code conversation
history to identify and extract learnings that weren't captured during
original sessions. Presents candidates for user confirmation before
creating skill files.

Closes blader#15

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
@adrien-hearthands
Copy link
Author

'Needs refinement as tests shows the feature extraction is not great but it's a good base

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Extract learnings from history

1 participant