1. Identify a concrete, actionable pattern (not vague observation)
2. Verify the pattern is about skill structure/content (not application code)
3. Categorize correctly: Structural Pattern, Content Pattern, Anti-Pattern, or Skill-Specific
4. Get user confirmation before writing

Adding learnings without verification = cluttered, unhelpful knowledge base

This is not optional. Every learning requires disciplined validation. </EXTREMELY-IMPORTANT>

Update Skill Learnings

MANDATORY FIRST RESPONSE PROTOCOL

Before adding ANY skill learning, you MUST complete this checklist:

1. ☐ Identify the specific pattern or anti-pattern from the session
2. ☐ Verify this is about skill creation/improvement (not application code)
3. ☐ Categorize: Structural, Content, Anti-Pattern, or Skill-Specific
4. ☐ Formulate as actionable guidance (imperative mood)
5. ☐ Check for duplicates in existing learnings
6. ☐ Present learning to user for confirmation
7. ☐ Announce: "Adding skill learning: [category] - [brief description]"

Adding learnings WITHOUT completing this checklist = noise in the knowledge base.

Purpose

This skill captures learnings about creating and improving Claude Code skills.

What it captures:

• Structural patterns (required sections, quality signals)
• Content patterns (clarity, user interaction)
• Anti-patterns to avoid
• Skill-specific learnings

Output: Updates to .claude/learnings/skill-learnings.md → synced to appropriate skills

Does NOT:

• Capture learnings about application code (use update-agent-learnings instead)
• Modify existing skills without user confirmation

When to Use

• After creating a new skill
• After improving an existing skill
• After reviewing skill quality (like the ulpi-generate-hooks review)
• When discovering patterns that make skills more effective
• Invoke via /update-skill-learnings

Never add learnings proactively. Only when session revealed concrete patterns.

Step 1: Session Analysis

Gate: Identify at least one concrete skill pattern before proceeding to Step 2.

Review the current conversation/session for skill-related patterns:

1. Structural discoveries:

   • What sections were missing from a skill?
   • What structural patterns made skills more effective?
   • What formatting improved clarity?

2. Content discoveries:

   • What made instructions clearer?
   • What examples helped understanding?
   • What error handling was missing?

3. Anti-patterns discovered:

   • What caused confusion or errors?
   • What was missing that caused problems?
   • What patterns should be avoided?

Step 2: Learning Extraction

Gate: User confirms extracted learnings before proceeding to Step 3.

Present findings to user and ask for confirmation:

I identified the following skill learnings from this session:

**Structural Patterns:**
1. [Pattern description]

**Content Patterns:**
1. [Pattern description]

**Anti-Patterns:**
1. [Pattern description]

**Skill-Specific (if any):**
1. [Skill name]: [Learning description]

Should I add these to the skill learnings file?

Use AskUserQuestion to:

• Confirm the extracted learnings are accurate
• Allow user to refine or add additional learnings
• Get approval before making changes

Learning Classification

Each learning should specify:

Classification Guide

Step 3: Update Central Learnings File

Gate: Learnings file updated successfully before proceeding to Step 4.

1. Read current .claude/learnings/skill-learnings.md
2. Locate the appropriate section:
   • Structural patterns go under ## Structural Patterns → appropriate subsection
   • Content patterns go under ## Content Patterns → appropriate subsection
   • Anti-patterns go under ## Anti-Patterns → appropriate subsection
   • Skill-specific learnings go under ## Skill-Specific Learnings → appropriate skill
3. Add new learnings to the appropriate section
4. Update the "Last updated" timestamp
5. Write updated file

File Structure

The learnings file follows this structure:

# Skill Learnings

## Structural Patterns
### Required Sections
- [learning items...]

### Quality Signals
- [learning items...]

---

## Content Patterns
### Clarity
- [learning items...]

### User Interaction
- [learning items...]

---

## Anti-Patterns
### Structure
- [learning items...]

### Content
- [learning items...]

---

## Skill-Specific Learnings

### [skill-name]
- [learning specific to this skill...]

---

*Last updated: [date]*

Step 4: Verification

Gate: All checks pass before marking complete.

Check 1: Learning Added

• New learning appears in correct section
• No duplicate entries created
• Timestamp updated

Check 2: Learning Quality

• Learning is actionable (imperative mood)
• Learning is specific (not vague)
• Learning includes context if needed

Check 3: File Integrity

• File structure preserved
• No sections accidentally removed
• Markdown formatting valid

Pre-Update Checklist

Before updating, verify:

• Learning is concrete and actionable
• Learning is about skills (not application code)
• Category assignment is correct
• No duplicates exist
• User confirmed the learning

Error Handling

Safety Rules

Quality Checklist (Must Score 8/10)

Score yourself honestly before marking update complete:

Learning Identification (0-2 points)

• 0 points: Vague observation, not actionable
• 1 point: Somewhat specific but could be clearer
• 2 points: Concrete, actionable pattern

Classification (0-2 points)

• 0 points: Wrong category
• 1 point: Correct category, wrong subcategory
• 2 points: Correct category and subcategory

User Confirmation (0-2 points)

• 0 points: Added without showing to user
• 1 point: Showed but didn't wait for confirmation
• 2 points: Full confirmation before writing

File Integrity (0-2 points)

• 0 points: Broke file structure
• 1 point: Minor formatting issues
• 2 points: Clean update, structure preserved

Duplicate Check (0-2 points)

• 0 points: Added duplicate without checking
• 1 point: Checked but not thoroughly
• 2 points: Verified no duplicates exist

Minimum passing score: 8/10

Common Rationalizations (All Wrong)

These are excuses. Don't fall for them:

• "This pattern is obvious" → STILL document it explicitly
• "It's just a small insight" → Small insights compound into quality
• "The learnings file is already comprehensive" → STILL check for gaps
• "The user knows what they want" → STILL confirm before adding
• "This is similar to an existing learning" → Check if it's a duplicate or refinement

Failure Modes

Failure Mode 1: Vague Learnings

Symptom: Learnings like "skills should be clear" Fix: Be specific - "Use imperative mood, provide concrete examples"

Failure Mode 2: Wrong Category

Symptom: Structural pattern filed under Anti-Patterns Fix: Review category definitions before filing

Failure Mode 3: Duplicate Added

Symptom: Same learning exists in multiple places Fix: Search existing learnings before adding

Failure Mode 4: Application Code Learning

Symptom: Learning about TypeScript patterns added to skill learnings Fix: Redirect to update-agent-learnings skill

Quick Workflow Summary

STEP 1: SESSION ANALYSIS
├── Review session for skill patterns
├── Identify structural discoveries
├── Identify content discoveries
├── Identify anti-patterns
└── Gate: Concrete pattern identified

STEP 2: LEARNING EXTRACTION
├── Categorize the learning
├── Formulate as actionable guidance
├── Check for duplicates
├── Present to user
├── Get confirmation
└── Gate: User approved

STEP 3: UPDATE LEARNINGS FILE
├── Read current skill-learnings.md
├── Locate correct section
├── Add new learning
├── Update timestamp
├── Write file
└── Gate: File updated

STEP 4: VERIFICATION
├── Verify learning added correctly
├── Verify learning quality
├── Verify file integrity
└── Gate: All checks pass

Completion Announcement

When update is complete, announce:

Skill learning added.

**Quality Score: X/10**
- Learning Identification: X/2
- Classification: X/2
- User Confirmation: X/2
- File Integrity: X/2
- Duplicate Check: X/2

**Learning Added:**
- Category: [category]
- Subcategory: [subcategory]
- Learning: [brief description]

**File Updated:** .claude/learnings/skill-learnings.md

**Next steps:**
Review the skill-learnings.md file to see all captured patterns.

Integration with Other Skills

The update-skill-learnings skill integrates with:

• update-agent-learnings — Use for application code learnings, not skill learnings
• commit — Commit skill-learnings.md changes after update
• skill-creator (example-skills) — Reference skill-learnings.md when creating new skills

Workflow Chain:

Session reveals skill pattern
         │
         ▼
update-skill-learnings skill (this skill)
         │
         ▼
skill-learnings.md updated
         │
         ▼
Future skill creation benefits from learnings

Decision Guide:

Is the learning about...
         │
    ┌────┴────┐
    ▼         ▼
Skills?   Application Code?
    │         │
    ▼         ▼
This skill   update-agent-learnings

Examples

Example 1: Adding a Structural Pattern Learning

Session Issue: Created a skill without EXTREMELY-IMPORTANT block.

Learning Extraction:

• Category: Structural Patterns
• Subcategory: Required Sections
• Learning: "Always include <EXTREMELY-IMPORTANT> block after frontmatter with verification requirements"

Result:

1. Added to .claude/learnings/skill-learnings.md under Structural Patterns → Required Sections

Example 2: Adding an Anti-Pattern

Session Issue: Skill had bash code blocks that looked executable but were meant as instructions.

Learning Extraction:

• Category: Anti-Patterns
• Subcategory: Content
• Learning: "Avoid bash code blocks that look executable but are meant as agent instructions - rephrase as prose"

Result:

1. Added to .claude/learnings/skill-learnings.md under Anti-Patterns → Content

Example 3: Adding a Skill-Specific Learning

Session Issue: ulpi-generate-hooks was missing patterns that commit/create-pr/start have.

Learning Extraction:

• Category: Skill-Specific
• Skill: ulpi-generate-hooks
• Learning: "Apply same structural patterns as commit/create-pr/start for consistency"

Result:

1. Added to .claude/learnings/skill-learnings.md under Skill-Specific Learnings → ulpi-generate-hooks