Project insights are lost when sessions end. This skill extracts patterns, quirks, and decisions from conversations and persists them to a knowledge base that survives context compaction and new sessions.
Download the skill ZIP
Upload in Claude
Go to Settings → Capabilities → Skills → Upload skill
Toggle on and start using
Test it
Using "learn". Use /learn to extract insights from our debugging session
Expected outcome:
- Knowledge Extraction Complete
- ───────────────────────────────────────
- Extracted:
- [Pattern] "API rate limiting with exponential backoff"
- Knowledge base now contains:
- - 5 patterns
- - 3 quirks
- - 7 decisions
Security Audit
SafePure prompt-based documentation skill with no code execution. Instructs AI to read/write markdown files in knowledge/learnings directory only. No network, scripts, or command execution. Behavior matches stated purpose.
Quality Score
What You Can Build
Build project memory
Capture team decisions and patterns so future sessions understand project conventions automatically
Train context-aware AI
Help AI assistants understand project-specific quirks and avoid repeated mistakes
Document institutional knowledge
Preserve architectural rationale and design choices for onboarding and long-term maintenance
Try These Prompts
Use /learn to extract the pattern from our debugging approach and save it to patterns.md
Use /learn to document the auth token quirk we discovered in quirks.md
Use /learn to record why we chose SQLite over PostgreSQL in decisions.md
Use /learn to extract all valuable insights from our conversation into the appropriate knowledge files
Best Practices
- Extract insights at natural stopping points during work sessions
- Use high confidence for verified insights, medium for inferred patterns
- Focus on project-specific knowledge, not generic programming concepts
Avoid
- Extracting obvious or trivial information that adds no value
- Recording insights without context or rationale
- Skipping extraction because insights seem "too small" to document