Skill Manager

Purpose

Meta-skill for managing the complete lifecycle of all other Claude Skills. Ensures skills are version-controlled in Git and automatically synchronized across all Claude environments (Desktop, Code CLI, API).

Core Capabilities

• Create: Scaffold new skills from templates with proper structure
• Validate: Check YAML frontmatter, markdown syntax, security
• Deploy: Sync skills to Code CLI, API, or Desktop
• Version: Semantic versioning and Git integration
• Monitor: Watch for changes and auto-sync

Architecture Overview

Source of Truth

• Git Repository: C:\Users\Kurt Anderson\github projects\claude-skills\
• Branch Strategy:
  • main = Production skills (stable, tested)
  • dev = Experimental/WIP skills
  • Feature branches = New skill development

Deployment Targets

1. Claude Code CLI: ~/.claude/skills/ (symlinked to git repo)
2. Claude API: Via /v1/skills endpoint (programmatic)
3. Claude Desktop: Manual upload or API-based (if available)

Standard Skill Structure

skill-name/
├── SKILL.md              # Required: YAML frontmatter + instructions
├── README.md             # Human documentation
├── scripts/              # Optional: Executable code
│   ├── __init__.py
│   └── helper.py
├── resources/            # Optional: Templates, data files
│   └── template.json
├── tests/                # Validation tests
│   └── test_skill.py
└── .skillmeta            # Version, author, dependencies

Key Operations

1. Create New Skill

Scaffolds a new skill directory with proper structure and template content.

Usage Pattern: "Create a new skill called 'n8n-flow-builder' that helps design n8n workflows"

What Happens:

1. Runs scripts/create_skill.py with skill name and description
2. Creates directory structure with SKILL.md template
3. Opens SKILL.md for review and editing
4. User iterates with Claude on content
5. Validates with scripts/validate_skill.py

2. Validate Skill

Checks skill for correctness and security before deployment.

Validation Checklist:

• ✅ YAML frontmatter is well-formed
• ✅ Required fields present (name, description)
• ✅ Markdown syntax is valid
• ✅ Script files have proper permissions
• ✅ No suspicious network calls to untrusted domains
• ✅ File access restricted to safe directories
• ✅ External dependencies documented

Usage Pattern: "Validate the n8n-flow-builder skill before deploying"

3. Deploy Skill

Synchronizes skill to target environments.

Deployment Options:

• --env code: Deploy to Claude Code CLI (~/.claude/skills/)
• --env api: Deploy to Claude API via /v1/skills endpoint
• --env all: Deploy to all available environments

Usage Pattern: "Deploy the n8n-flow-builder skill to all environments"

What Happens:

1. Validates skill first (auto-validation)
2. For Code CLI: Creates symlink or copies to ~/.claude/skills/
3. For API: Zips skill, POSTs to /v1/skills endpoint
4. Logs deployment status
5. Commits to Git if requested

4. List and Search Skills

Find skills in the repository.

Usage Patterns:

• "What skills do I have?"
• "Show me all skills related to GitHub"
• "List skills that haven't been deployed yet"

Returns:

• Skill name and version
• Description
• Deployment status
• Last modified date

5. Update Existing Skill

Modify and redeploy skills with version bumping.

Usage Pattern: "Update the linear-orchestration skill to include new webhook format"

What Happens:

1. Reads current skill from git repo
2. Makes requested changes to SKILL.md or scripts
3. Shows diff for review
4. User approves changes
5. Validates updated skill
6. Deploys updated version
7. Git commit with semantic version bump (patch/minor/major)

Helper Scripts Reference

scripts/create_skill.py

Scaffolds new skills from template.

Arguments:

• --name: Skill name (kebab-case)
• --description: Brief description of what skill does
• --template: Optional template to use (default, api, workflow, etc.)

Example:

python scripts/create_skill.py --name "api-doc-generator" \
  --description "Generate OpenAPI specifications"

scripts/validate_skill.py

Validates skill structure and security.

Checks:

• YAML frontmatter syntax
• Required field presence
• Markdown lint
• Script security scan
• Dependency audit

Example:

python scripts/validate_skill.py skill-name/

scripts/deploy_skill.py

Deploys skills to target environments.

Arguments:

• skill_path: Path to skill directory
• --env: Target environment (code|api|all)
• --validate: Run validation first (default: true)
• --commit: Commit to git after deploy (default: false)

Example:

python scripts/deploy_skill.py n8n-flow-builder/ --env all --commit

Security Considerations

Validation Rules

Skills must pass all security checks before deployment:

1. Network Access: Scripts cannot call arbitrary URLs

   • Whitelist approved domains in skill config
   • Block suspicious patterns

2. File Access: Scripts restricted to safe directories

   • Skill directory itself
   • /mnt/user-data/ (Claude execution environment)
   • Explicitly approved paths only

3. Dependency Audit: External libraries documented

   • All imports declared in .skillmeta
   • Known vulnerabilities flagged
   • Version pinning required

4. Code Review: Human review for high-risk operations

   • Network calls
   • File system modifications
   • Environment variable access

Approval Workflow

• Auto-deploy: Skills that pass validation on main branch
• Manual review: New skills or those with network/file access beyond standard patterns

Integration with Your Workflow

Bootstrap Process

When starting a new session:

1. Claude Desktop loads skill-manager skill (this one)
2. Check git repo for latest skill updates
3. Auto-sync to Code CLI if changes detected
4. Report which skills are active and their versions

Continuous Sync Options

Option A: File Watcher (Recommended) Background daemon watches git repo for changes:

# scripts/skill_sync_daemon.py
# Watches for SKILL.md changes
# Auto-validates and deploys to Code CLI
# Logs all operations

Option B: n8n Automation (Optional) GitHub webhook triggers n8n workflow on push to main:

1. Validates all changed skills
2. Runs tests
3. Deploys to Code CLI (file sync)
4. Deploys to API (HTTP POST)
5. Posts deployment report to Slack
6. Creates Linear ticket if failures detected

Option C: Manual Sync (Simplest) Run deployment manually when skills change:

python scripts/deploy_skill.py --env all

Workflow Examples

Example 1: Create and Deploy New Skill

User: "Create a new skill for n8n workflow authoring"

Claude (using skill-manager):
1. ✅ Runs create_skill.py with name "n8n-flow-builder"
2. ✅ Generates SKILL.md with template structure
3. 📝 Opens for user review: "Here's the template. What should I include?"
4. 🔄 User iterates: "Add section on webhook patterns"
5. 🔍 Validates with validate_skill.py
6. ✅ User approves: "Looks good, deploy it"
7. 🚀 Runs deploy_skill.py --env all
8. 📦 Commits to git: "feat: add n8n-flow-builder skill v1.0.0"
9. 🔗 Pushes to GitHub
10. ✅ Reports: "Deployed to Code CLI and ready for API upload"

Example 2: Update Existing Skill

User: "Update linear-orchestration to include new status field format"

Claude (using skill-manager):
1. 📖 Reads linear-orchestration/SKILL.md from git
2. ✏️ Makes changes to include new status field
3. 👀 Shows diff: "Here's what I'm changing..."
4. ✅ User approves
5. 🔍 Validates changes
6. 🚀 Deploys updated version
7. 📦 Git commit: "feat(linear): add new status field format"
8. 🏷️ Bumps version: v1.2.0 → v1.3.0

Example 3: Search and Discover

User: "What skills do I have related to GitHub?"

Claude: Searches repository and returns:
- **github-coordinator** (v1.2.0) - Manage branches, PRs, CI/CD
  Last updated: 2025-10-15
  Status: Deployed to Code CLI ✅
  
- **github-actions-builder** (v0.5.0) - Create GH Actions workflows
  Last updated: 2025-10-10
  Status: Local only, not deployed ⚠️

Setup Instructions

Initial Setup

1. Ensure git repository exists at expected path

2. Initialize git if not already done:

   cd "C:\Users\Kurt Anderson\github projects\claude-skills"
   git init
   git add .
   git commit -m "Initial commit: skill-manager"
   

3. Set up Code CLI symlink (run once):

   ln -s "C:\Users\Kurt Anderson\github projects\claude-skills" ~/.claude/skills
   

4. Install Python dependencies (if using scripts):

   pip install pyyaml click watchdog
   

Warp Integration

Since Warp terminal owns system-level tasks:

• Run deployment scripts
• Manage git operations (commit, push, pull)
• Set up and maintain symlinks
• Execute file watchers as background services

Best Practices

Skill Naming Convention

• Use kebab-case: n8n-flow-builder, linear-orchestration
• Be descriptive but concise
• Prefix with domain if part of suite: github-coordinator, github-actions-builder

Version Numbering (Semantic Versioning)

• Major (X.0.0): Breaking changes to skill interface
• Minor (0.X.0): New features, backward compatible
• Patch (0.0.X): Bug fixes, documentation updates

Commit Message Format

type(scope): brief description

Examples:
feat(linear): add webhook callback format
fix(n8n): correct node connection validation
docs(skill-manager): update deployment instructions

Testing Before Deploy

1. Validate skill structure
2. Test in isolated Claude session
3. Verify all referenced files exist
4. Check script executables work
5. Review security implications

Troubleshooting

Common Issues

Issue: Skill not loading in Claude Code CLI Solution: Check symlink is correct, restart Claude Code

Issue: Validation fails on YAML frontmatter Solution: Check frontmatter has three dashes on separate lines, proper YAML syntax

Issue: Script can't be executed Solution: Verify script has execute permissions, check shebang line

Issue: Deployment to API fails Solution: Check API key is set, skill size under 8MB limit, max 8 skills per request

Next Steps

Once skill-manager is working:

1. Create your first domain skill (n8n-flow-builder recommended)
2. Set up automation (file watcher or n8n workflow)
3. Establish review process for new skills
4. Document your skill library in main README
5. Share useful skills with team (if applicable)

Notes

• This skill is self-referential (meta) - it manages itself too
• Start simple: manual deployment first, automation later
• Skills are portable: work with any LLM that can read files
• Git is your safety net: always commit before major changes
• Desktop Commander can be buggy: GitHub MCP is more reliable for repo operations

Resources

• Anthropic Skills Documentation
• Skills GitHub Repository
• Skills Cookbook

Auto-Trigger Pattern Recognition

When to Automatically Trigger skill-manager

Claude should automatically suggest skill creation when detecting these patterns:

Pattern 1: Repeated Workflow

User does same multi-step process twice
  → "This feels like a pattern we should capture in a skill"
  → Offer to create skill

Pattern 2: Complex Integration

User coordinates multiple MCPs successfully
  → "This cross-MCP workflow could be an integration-workflows entry"
  → Offer to add to integration-workflows skill

Pattern 3: MCP Enhancement

Discovering better way to use GitHub/Linear/n8n
  → "Should we add this to the relevant skill?"
  → Update existing skill + MCP repo (if forked)

Pattern 4: Efficiency Discovery

User: "Oh that's much better!"
Claude: "Want to capture this approach in a skill?"

Pattern 5: User Explicitly States

User: "This is how we always do it"
User: "Remember this pattern"
User: "We should capture this"
  → Immediately offer skill creation

Auto-Trigger Workflow

When pattern detected:

Step 1: Recognition

Claude: "💡 This feels like a repeatable pattern!
We've now [created X successfully / solved Y twice / found Z approach].
Should I create/update a skill to capture this?"

Step 2: If User Says Yes

Claude uses skill-manager to:
1. Determine if new skill or update existing
2. Draft skill content based on conversation
3. Show preview to user
4. Create/update skill files
5. Generate zip automatically
6. Prompt for upload (see below)

Step 3: Upload Prompt

Claude: "✅ Skill ready!

📦 Upload to Claude Desktop:
1. File location: C:\path\to\skill.zip
2. Go to: Settings > Capabilities > Upload skill
3. Drag and drop the zip file
4. Toggle skill ON

Should I wait while you do this, or continue?
(Say 'uploaded' when done)"

Step 4: Verification (Optional)

After user says "uploaded":
Claude: "Great! Let me verify it loaded correctly by checking if
I can see it in my skills list... [test query]"

Automatic Zip Generation

Every time a skill is created or updated:

Auto-Generate Process:

# Automatically runs after any skill modification
1. Navigate to skill directory
2. Create zip with SKILL.md at root (+ any other files)
3. Place in claude-skills/ parent directory
4. Report location to user

File Naming:

skill-name.zip          # For new/updated skills
skill-name-v1.2.0.zip   # For versioned releases (optional)

Zip Contents:

skill.zip/
├── SKILL.md         # At root (required by Claude)
├── README.md        # If exists
├── scripts/         # If exists
└── resources/       # If exists