majiayu000 Skill Generator (Meta-Skill)
This skill creates other skills.
When to Use
- User repeats similar instructions 2+ times in conversation
- User says "remember this", "save this workflow", "create a skill"
- You notice a pattern that should be automated
- User provides detailed methodology that should be reusable
Detection Patterns
Track these signals:
- Repetition: Same request type appears multiple times
- Detailed Instructions: User explains step-by-step how to do something
- Explicit Request: "Make this a skill", "Remember this process"
- API/Documentation Lookups: User asks for current docs repeatedly
Auto-Generation Process
Step 1: Detect Pattern
After conversation where user explains methodology:
# Record the pattern
./skill-learner record \
"User's request description" \
"User's detailed instructions on how they want it done"
Step 2: Acknowledge to User
🎓 I notice you've explained this workflow [N] times.
I'm tracking this pattern and will create a reusable skill after one more similar request.
This will let me automatically handle this in the future without needing detailed instructions.
Step 3: Generate Skill (after threshold)
System automatically creates .claude/skills/auto-[intent]-[id]/SKILL.md
Step 4: Confirm with User
✅ I've created a new skill for [intent]!
From now on, when you ask about [trigger phrases], I'll automatically:
1. [Step 1 from user's instructions]
2. [Step 2 from user's instructions]
3. [Step 3 from user's instructions]
The skill is saved in .claude/skills/auto-[name]/ and improves with each use.
You can edit it anytime to refine the approach.
Special Cases
API Documentation Skill
If user asks for API docs 2+ times:
---
name: auto-api-docs-lookup
description: Always search web for current API documentation. Use whenever API, SDK, or technical documentation is mentioned.
allowed-tools: [web_search, web_fetch, bash_tool, view]
---
# API Documentation Lookup
## When to Use
- User mentions API, SDK, library documentation
- Technical integration questions
- "Latest" or "current" version queries
## Process (Auto-search enabled)
1. **ALWAYS web_search first**: "[library name] official documentation [current year]"
2. Prioritize official sources (github.com/org/repo, docs.library.com)
3. Use web_fetch to read full current docs
4. Cross-reference version compatibility
5. Cite sources with dates
## User Instruction
[User's explained methodology goes here]
Analysis Skill
If user repeatedly explains analysis approach:
---
name: auto-analysis-methodology
description: Analysis workflow based on user's preferred methodology. Use when user asks to "analyze", "evaluate", or "assess".
---
# Analysis Methodology
## When to Use
- "Analyze this..."
- "Evaluate the..."
- "Break down..."
## User's Preferred Approach
[Captured from user's repeated instructions]:
1. [Step they always mention first]
2. [Their second step]
3. [How they want results formatted]
## Validation
[How user checks if analysis is complete]
Skill Improvement Loop
After each use of an auto-generated skill:
# In your response, include:
"""
📊 Skill Usage Feedback
This used the auto-generated '[skill-name]' skill.
Did this match your expectations? If not, I can refine the skill by:
- Adding more detail
- Adjusting the methodology
- Including additional validation steps
Just say "improve the [skill-name] skill" and explain what to change.
"""
Commands for Users
# Track a new pattern
./skill-learner track
# See what's being learned
./skill-learner list
# Manually create skill from current conversation
./skill-learner record "what user asked" "how they explained to do it"
Integration with Claude Code
When user interacts with you:
- After response: Check if this is a repetitive pattern
- Call:
./skill-learner record "user_request" "your_methodology" - System: Auto-generates skill at threshold
- Next time: Skill is automatically invoked
Example Workflow
First Request:
User: "Can you analyze this BRD? First read the executive summary, then assess market fit, then evaluate technical feasibility."
Claude: [Follows instructions, provides analysis]
Claude (internal): Records pattern via skill-learner
Second Similar Request:
User: "Analyze this other BRD following the same approach."
Claude: "🎓 I'm tracking this analysis pattern. One more similar request and I'll create a reusable skill."
Claude: [Executes analysis, records pattern]
Claude (internal): Skill auto-generated!
Third Request:
User: "Analyze the new warehouse BRD."
Claude: "✅ Using your custom analysis methodology skill..."
Claude: [Auto-invokes the learned skill, no detailed instructions needed]
Maintenance
Skills are stored in .claude/learning/patterns.json - this tracks:
- Pattern signatures
- Occurrence counts
- Example requests
- Generated skills
Clean up old patterns:
# Review learned patterns
./skill-learner list
# Edit/remove if needed
nano .claude/learning/patterns.json
Real-World Examples for Project Conductor
Example 1: BRD Analysis Pattern
# First time: You explain in detail how to analyze BRDs
# Second time: You explain again (slightly different BRD)
# System: Auto-creates "auto-analysis-methodology" skill
# Third time: Just say "analyze this BRD" - skill auto-invokes!
Example 2: API Documentation
# You ask: "What's the latest Express.js middleware documentation?"
# Claude: web_search → web_fetch → provides answer
# You ask again: "Check Socket.io docs for new authentication methods"
# System: Creates "auto-api-docs-lookup" skill with web-search enabled
# Future: Any API question automatically searches current docs first
Example 3: Deployment Checklist
# You ask: "Help me deploy to production - check env vars, run tests, build Docker"
# You ask again: "Deploy the staging environment following same steps"
# System: Creates "auto-deployment-checklist" skill
# Future: "Deploy to [env]" auto-invokes comprehensive checklist