Skip to main content

​
/learn-eval Command

Extends /learn with a quality gate and save-location decision before writing any skill file.

​
Process

  1. Review session for extractable patterns
  2. Identify the most valuable/reusable insight
  3. Determine save location:
    • Global (~/.claude/skills/learned/): Generic patterns usable across 2+ projects
    • Project (.claude/skills/learned/): Project-specific knowledge
  4. Draft the skill file
  5. Self-evaluate using quality rubric
  6. Ask user to confirm with scores table
  7. Save to determined location

​
Quality Rubric

Dimension135
SpecificityAbstract principles onlyRepresentative example presentRich examples covering all patterns
ActionabilityUnclear what to doMain steps understandableImmediately actionable, edge cases covered
Scope FitToo broad or too narrowMostly appropriateName, trigger, and content perfectly aligned
Non-redundancyNearly identical to another skillSome overlap but uniqueCompletely unique value
CoverageCovers fraction of taskMain cases coveredMain cases, edge cases, and pitfalls covered

​
Command Syntax

/learn-eval
No arguments required - analyzes current session.

​
Evaluation Output

| Dimension | Score | Rationale |
|-----------|-------|-----------|  
| Specificity | 4/5 | Has code examples for main use cases |
| Actionability | 5/5 | Step-by-step instructions with commands |
| Scope Fit | 4/5 | Slightly broader than name suggests |
| Non-redundancy | 5/5 | Unique pattern not in existing skills |
| Coverage | 4/5 | Main cases covered, missing error handling |
| **Total** | **22/25** | |

**Save Location**: Global (~/.claude/skills/learned/)
**Filename**: async-error-handling.md

Proceed with save? (yes/no)

​
Requirements

  • All dimensions must score ≥ 3 before saving
  • If any dimension scores 1-2, improve draft and re-score
  • User must explicitly confirm before writing
  • Commands: /learn, /skill-create, /evolve