Imported from naj2r/claude-econ-paper-template (
.claude/skills/learn/SKILL.md). Install upstream withnpx skills add naj2r/claude-econ-paper-template --skill learn. Copyright stays with the author.
Learn
Extract reusable knowledge from the current session and persist it for future sessions.
Input: $ARGUMENTS — optional topic focus. If blank, review the full session.
What to Extract
Corrections ([LEARN] tags)
Look for any [LEARN:category] tags in the conversation where the user corrected Claude. Save as:
[LEARN:category] wrong assumption → correct behavior
Patterns
- Coding patterns that worked well or failed
- Data quirks that will recur (e.g., {{data source}} missing value format)
- Stata/R translation gotchas encountered
- File path conventions that are easy to get wrong
Preferences
- User workflow preferences discovered during the session
- Communication style preferences
- Tool usage patterns
Where to Save
| Type | Location |
|---|---|
| General learnings | MEMORY.md (update, don't duplicate) |
| Stata-specific | .claude/rules/stata-r-conventions.md |
| Data-specific | Relevant backmatter chapter (93 for problems) |
| New skill pattern | Create new skill in .claude/skills/ |
Process
- Review the session for corrections, discoveries, and patterns
- Check existing memory — don't duplicate what's already stored
- Categorize each learning (correction, pattern, preference, skill)
- Write to appropriate location
- Report what was saved and where
Quality Gates
- Don't save speculative conclusions from a single interaction
- Don't save session-specific context (current task, temp state)
- Do save stable patterns confirmed across multiple observations
- Do save explicit user requests ("always do X", "never do Y")