• Creating new global skills or commands
• Setting up skills for a new project
• Synchronizing projects with updated global skills
• Organizing the centralized skill repository
• Troubleshooting skill discovery or activation issues
• Understanding the skill lifecycle

1. Create directory structure:
   bash

   mkdir -p $CLAUDE_METADATA/{skills,commands}
   cd $CLAUDE_METADATA

2. Set environment variable (add to

   ~/.zshrc

   or

   ~/.bashrc

   ):
   bash

   echo 'export CLAUDE_METADATA="$HOME/path/to/claude_data"  # Adjust to your actual path' >> ~/.zshrc
   source ~/.zshrc

3. Verify setup:
   bash

   echo $CLAUDE_METADATA
   # Should output your claude_data directory path

4. Create initial documentation:
   bash

   # Create README and QUICK_REFERENCE
   # (use templates from claude-skill-management skill)

5. Initialize git (recommended):
   bash

   cd $CLAUDE_METADATA
   git init
   git add .
   git commit -m "Initial centralized skill repository"

6. Create your first skill:
   bash

   mkdir -p $CLAUDE_METADATA/skills/my-first-skill
   # Create SKILL.md with frontmatter

7. Link to first project:
   bash

   cd ~/Workdir/my-project
   mkdir -p .claude/skills
   ln -s $CLAUDE_METADATA/skills/my-first-skill .claude/skills/

• Use

  $HOME

  not hardcoded paths for portability
• Source shell config after adding:

  source ~/.zshrc

• Verify in new terminals:

  echo $CLAUDE_METADATA

• Document for team members in README.md

mkdir -p $CLAUDE_METADATA/skills/your-skill-name

• Use

  kebab-case

  (lowercase with hyphens)
• Be descriptive but concise
• Examples:

  galaxy-tool-wrapping

  ,

  python-testing

  ,

  docker-workflows

• Specific use case 1
• Specific use case 2
• Specific use case 3

• Practice 1
• Practice 2

cat > $CLAUDE_METADATA/commands/your-category/command-name.md << 'EOF'
---
name: command-name
description: Brief description shown in /help
---

Your command prompt here. This will be expanded when the user types /command-name.

You can include:
- Multi-line instructions
- Variable references: {{variable_name}}
- Markdown formatting
- Code blocks

Example:
Check the status of all workflows for species {{species_name}}.
Show me which workflows are complete, running, or failed.
EOF

• Use

  kebab-case

• Start with verb:

  check-status

  ,

  debug-failed

  ,

  update-skills

• Be specific:

  deploy-production

  not just

  deploy

ln -s $CLAUDE_METADATA/skills/token-efficiency .claude/skills/token-efficiency

• Uses

  --quiet

  mode for commands automatically
• Reads log files efficiently (tail/grep instead of full read)
• Strategic file selection for learning mode
• Extends your Claude Pro usage 5-10x

ln -s $CLAUDE_METADATA/skills/claude-collaboration .claude/skills/claude-collaboration

• Explains when and how to update skills
• Documents skill lifecycle and version control
• Helps onboard team members
• Ensures consistency across projects

ln -s $CLAUDE_METADATA/skills/galaxy-automation .claude/skills/galaxy-automation

• Foundation for Galaxy workflow automation
• Required dependency for vgp-pipeline
• Useful for galaxy-tool-wrapping (Planemo testing)
• Reduces duplication across Galaxy-related skills

Set up Claude Code for this project. Show me available skills in $CLAUDE_METADATA and let me choose which ones to symlink.

1. Automatically symlink token-efficiency, claude-collaboration, python-environment-management (if not already present)
2. List all available skills and commands
3. Ask which additional ones you want
4. Create the symlinks
5. Verify everything works

• ✅ Repeated patterns or solutions
• ✅ New best practices
• ✅ Common errors and their fixes
• ✅ Token optimizations
• ✅ Workflow improvements

• ❌ One-time issues
• ❌ Experimental approaches (wait until proven)
• ❌ User-specific preferences
• ❌ Obvious information

Check what skills and commands are available in $CLAUDE_METADATA and compare with what's currently symlinked in this project. Show me what's new or missing, and let me choose which ones to add.

1. List current symlinks
2. List available skills in

   $CLAUDE_METADATA

3. Show what's new
4. Create symlinks for selected items

• Format:

  domain-subdomain

  or

  technology-purpose

• Examples:

  • galaxy-tool-wrapping

    (technology-purpose)

  • vgp-pipeline

    (project-type)

  • python-testing

    (language-purpose)

• Format:

  verb-noun

  or

  verb-target

• Examples:

  • check-status

    (verb-noun)

  • debug-failed

    (verb-state)

  • update-skills

    (verb-noun)

• ✅

  SKILL.md

  (required) - Main skill file
• ✅ Clear frontmatter with name and description
• ✅ "When to Use This Skill" section
• ⚠️

  reference.md

  (optional) - Detailed documentation
• ⚠️

  examples/

  (optional) - Example code/configs

• ✅ Frontmatter with name and description
• ✅ Clear prompt/instructions
• ✅ Examples if the command takes parameters

$CLAUDE_METADATA

1. Not using symlinks - Projects have copies instead
   bash

   # Check if it's a symlink
   ls -la .claude/skills/skill-name
   # Should show -> pointing to $CLAUDE_METADATA

2. Claude Code hasn't restarted - Skills loaded at session start
   • Fix: Restart Claude Code session
3. Editing wrong file - Multiple copies exist
   bash

   # Find all copies
   find ~/Workdir -name "SKILL.md" -path "*/skill-name/*"
   # Should only show one in $CLAUDE_METADATA

• galaxy-tool-wrapping/SKILL.md

  - Only Galaxy tools

• vgp-pipeline/SKILL.md

  - Only VGP workflows

• everything/SKILL.md

  - Galaxy + VGP + Docker + Python + ...

• Review session learnings
• Update skills with new patterns
• Commit changes with clear messages

• Audit all skills for conflicts
• Remove outdated information
• Reorganize if needed

final_script.sh

**Benefits:**
- Reusable across projects
- Single source of truth for logic
- Easy to maintain and update
- Parameter validation in one place
- Reduces duplication

**Example use case:**
Creating a global installer for backup systems that can be customized for any data file and directory structure. The template contains all the logic, and sed substitution customizes it for each project.

**Alternative: Template files:**
```bash

1. Create skill:
   bash

   mkdir -p $CLAUDE_METADATA/skills/docker-workflows
   # Create SKILL.md with frontmatter

2. Test in isolated project:
   bash

   mkdir -p /tmp/test/.claude/skills
   ln -s $CLAUDE_METADATA/skills/docker-workflows /tmp/test/.claude/skills/
   # Start Claude Code, test the skill

3. Link to real projects:
   bash

   cd ~/Workdir/real-project
   ln -s $CLAUDE_METADATA/skills/docker-workflows .claude/skills/

4. Share with team:
   bash

   cd $CLAUDE_METADATA
   git add skills/docker-workflows/
   git commit -m "Add docker-workflows skill"
   git push

1. Work with Claude, discover pattern
2. Tell Claude: "Add this pattern to the vgp-pipeline skill"
3. Claude updates:

   $CLAUDE_METADATA/skills/vgp-pipeline/SKILL.md

4. Review changes:

   git diff

5. Commit:

   git commit -m "Add WF8 troubleshooting pattern"

6. Push:

   git push

   (if using team repo)
7. All projects auto-updated via symlinks! 🎉

1. Create .claude directory:
   bash

   cd ~/Workdir/new-project
   mkdir -p .claude/skills .claude/commands

2. Symlink essential global skills:
   bash

   # Always include these
   ln -s $CLAUDE_METADATA/skills/token-efficiency .claude/skills/token-efficiency
   ln -s $CLAUDE_METADATA/skills/claude-collaboration .claude/skills/claude-collaboration
   ln -s $CLAUDE_METADATA/skills/python-environment-management .claude/skills/python-environment-management

3. Symlink ALL global commands:
   bash

   # Always include for all projects
   ln -s $CLAUDE_METADATA/commands/global/*.md .claude/commands/

4. Add project-specific skills (if needed):
   bash

   # For Galaxy projects:
   ln -s $CLAUDE_METADATA/skills/galaxy-automation .claude/skills/galaxy-automation

5. Tell Claude (or use

   /setup-project

   if already linked):

   I've set up the essential skills and commands. Show me other available skills in
   $CLAUDE_METADATA that might be relevant for [describe your project type].

   Or simply:

   /list-skills

6. Claude shows list, you choose project-specific skills
7. Claude creates additional symlinks
8. Commit symlinks to project:
   bash

   git add .claude/
   git commit -m "Add Claude Code configuration
   
   Essential global skills and commands:
   - token-efficiency (token optimization)
   - claude-collaboration (team best practices)
   - galaxy-automation (BioBlend & Planemo)
   - Global commands: /update-skills, /list-skills, /setup-project
   
   [Additional project-specific skills if added]"

9. Team members get symlinks via git pull
10. Team members point to their $CLAUDE_METADATA
    • They need to set

      $CLAUDE_METADATA

      in their shell config
    • Symlinks work automatically once environment variable is set

/update-skills

• Legacy files after reorganizations
• Duplicate documentation in different locations
• Outdated quick-start guides
• Superseded command files

1. Identify redundancies:
   bash

   # List all markdown files
   find $CLAUDE_METADATA -name "*.md" -type f | sort
   
   # Compare similar files
   diff file1.md file2.md
   
   # Search for overlapping content
   grep -r "specific topic" $CLAUDE_METADATA

2. Categorize files:
   • Skills (must be unique, in

     skills/*/SKILL.md

     )
   • Supporting docs (should be in skill subdirectories)
   • Commands (one version only, in

     commands/category/

     )
   • Root docs (only README.md, QUICK_REFERENCE.md)
3. Always backup before cleanup:
   bash

   cd $CLAUDE_METADATA
   mkdir -p .backup-$(date +%Y%m%d-%H%M%S)
   cp -r files-to-modify .backup-$(date +%Y%m%d-%H%M%S)/

4. Consolidation patterns:
   • Standalone docs → Move to

     skills/skill-name/reference.md

   • Legacy commands → Remove if superseded by new versions
   • Duplicate guides → Consolidate into single skill
   • Quick reference prompts → Replace with standardized QUICK_REFERENCE.md
5. Update skill to reference supporting docs:
   markdown

   ## Supporting Documentation
   
   This skill includes detailed reference documentation:
   - **reference.md** - Comprehensive guide
   - **troubleshooting.md** - Common issues and solutions

6. Verify structure:
   bash

   tree -L 3 $CLAUDE_METADATA
   # Should show clean, logical organization

• Reduces confusion about which file to use
• Improves discoverability via progressive disclosure
• Easier maintenance (single source of truth)
• Faster skill loading (no duplicate content)

1. Central repository - All skills in

   $CLAUDE_METADATA

2. Symlinks, not copies - Updates propagate automatically
3. Version control - Track changes with git
4. Essential global skills first - Always symlink token-efficiency, claude-collaboration, and python-environment-management
5. Selective activation - Link only relevant skills per project
6. Team collaboration - Share via git, everyone benefits

undefined