Learning System
Skills improve from your feedback.
How It Works
Every skill has a LEARN.md file. Before running, the skill reads its lessons. Your feedback shapes future behavior.
Run skill → Get output → Give feedback → Next run is better
LEARN.md Structure
Four categories of lessons:
| Category | Purpose | Example |
|---|---|---|
| Better Questions | Questions to ask | "Ask about i18n requirements" |
| Edge Cases | Special scenarios | "Handle multi-tenant features" |
| Workflow | Process improvements | "Generate JSON alongside PRD" |
| Common Mistakes | Things to avoid | "Don't assume single region" |
Adding Lessons
Use the learn-add skill:
/jaan-to:learn-add {target} "{lesson}"
Examples:
/jaan-to:learn-add pm-prd-write "Always ask about rollback strategy"
/jaan-to:learn-add pm-prd-write "Check for accessibility requirements"
Where Lessons Live
| Target | Location |
|---|---|
| Skills | jaan-to/learn/{skill}.learn.md |
| Templates | jaan-to/learn/{skill}.learn.md |
| Stacks | jaan-to/learn/{skill}.learn.md |
Three Learning Layers
1. Skill Learning
How the skill executes.
- Better questions to ask
- Edge cases to check
2. Template Learning
How outputs are formatted.
- Missing sections to add
- Phrasing improvements
3. Stack Learning
What context matters.
- Constraints that always apply
- Team norms to remember
Auto-Categorization
Lessons are categorized by keywords:
| Keywords | Category |
|---|---|
| ask, question, clarify | Better Questions |
| edge, special, handle | Edge Cases |
| workflow, process, step | Workflow |
| avoid, mistake, don't | Common Mistakes |
Viewing Lessons
Read the LEARN.md file directly:
jaan-to/learn/jaan-to:pm-prd-write.learn.md
When to Add Feedback
After any skill run where you notice:
- A question that should have been asked
- A section that was missing
- A pattern worth remembering
- A mistake to avoid
Good feedback makes the system smarter.