Skill Management

Skill Management

IMPORTANT: This skill should be activated BEFORE modifying any skill files!

You are an expert at creating and maintaining high-quality Claude Code skills. This skill helps you follow best practices and remember to commit changes to the skills repository.

When to Use This Skill (Read This First!)

✅ CORRECT Workflow

ALWAYS activate this skill FIRST when:

1. Creating a new skill in ~/.claude/skills/
2. Editing any existing SKILL.md file
3. Modifying skill-related files (EXAMPLES.md, REFERENCE.md, scripts, etc.)
4. User requests to create, improve, update, review, or refactor a skill
5. Discussing skill quality or effectiveness

The correct order is:

1. User asks to modify a skill (or you identify need to update one)
2. YOU ACTIVATE THIS SKILL IMMEDIATELY
3. You review best practices and quality checklist
4. You make changes following the guidelines
5. You commit changes to the skills git repository

❌ INCORRECT Workflow (Anti-pattern)

NEVER do this:

1. User asks to modify a skill
2. You directly edit the SKILL.md file
3. You commit the changes
4. Later realize you didn't follow best practices
5. You have to redo the changes

Examples of When to Activate

✅ "Can you update the literate-programming skill to be more emphatic?" → ACTIVATE THIS SKILL IMMEDIATELY, then plan changes

✅ "Create a new skill for handling API documentation" → ACTIVATE THIS SKILL IMMEDIATELY, then design skill

✅ "The code-review skill isn't triggering when it should" → ACTIVATE THIS SKILL IMMEDIATELY to review triggers

✅ Any task involving files in ~/.claude/skills/ → ACTIVATE THIS SKILL IMMEDIATELY

Remember

• Skills have specific quality requirements and best practices
• Following the checklist prevents having to redo work
• Git commits are REQUIRED after any skill modification
• Skill quality directly affects Claude Code effectiveness

Original "When to Use" Section

Invoke this skill proactively when:

1. Creating new skills - User requests a new skill or you identify a need for one
2. Modifying existing skills - Any edit to SKILL.md or related files in ~/.claude/skills/
3. Reviewing skills - User asks to review, improve, or refactor a skill
4. Skill quality questions - Discussing skill effectiveness, structure, or best practices
5. After skill changes - To verify git commit was performed

Core Principles (from Claude Code Documentation)

1. Conciseness

• Assume Claude is already intelligent
• Only include context Claude doesn't already possess
• Challenge each piece of information for necessity
• Keep SKILL.md under 500 lines
• Split additional content into separate files (REFERENCE.md, EXAMPLES.md, etc.)

2. Degrees of Freedom

Match instruction specificity to task fragility:

• High freedom (text instructions): Multiple valid approaches exist
• Medium freedom (pseudocode/patterns): Preferred patterns with acceptable variation
• Low freedom (specific scripts): Operations are fragile, exact sequences required

3. Progressive Disclosure

Use referenced files to load content on-demand:

• Keep direct references one level deep from SKILL.md
• Use separate reference files for different domains/features
• Structure long references with table of contents

Skill Structure Requirements

YAML Frontmatter (Required)

---
name: skill-name-here
description: What this skill does and when to use it. Max 1024 characters.
---

Name requirements:

• Maximum 64 characters
• Lowercase letters, numbers, and hyphens only
• No reserved words ("anthropic", "claude")

Description requirements:

• Maximum 1024 characters
• Non-empty, no XML tags
• Use third-person perspective
• State BOTH what the skill does AND when to use it
• Include specific trigger terms and contexts
• Be explicit about proactive invocation if applicable
• Avoid vague language ("helps with documents")

Effective Description Pattern

description: [What it does]. Use [proactively/when]: (1) [trigger condition],
(2) [keyword/phrase triggers], (3) [context triggers]. [Special instructions].

Example:

description: Write and analyze literate programs using noweb. Use proactively
when: (1) creating, editing, or reviewing .nw files, (2) user mentions
"literate quality" or "noweb", (3) requests to improve documentation.
This skill should be invoked BEFORE making changes to .nw files.

Three-Level Loading Architecture

Level 1 - Metadata (~100 tokens, always loaded):

• YAML frontmatter for discovery

Level 2 - Instructions (<5k tokens, loaded when triggered):

• Main SKILL.md body with procedures and best practices

Level 3 - Resources (unlimited, accessed as needed):

• Additional files: REFERENCE.md, EXAMPLES.md, FORMS.md
• Python scripts (executed via bash, output only enters context)
• Database schemas, templates, etc.

Content Guidelines

Organization Patterns

Templates: Provide strict format for critical outputs, flexible guidance for context-dependent work

Examples: Show input/output pairs demonstrating desired style and detail level

Workflows: Break complex operations into clear sequential steps with checklists

Feedback loops: Implement validate-fix-repeat cycles for quality-critical tasks

Writing Guidelines

• Use imperative/infinitive form - Write instructions using verb-first format (e.g., "To accomplish X, do Y" rather than "You should do X"). Maintain objective, instructional language for AI consumption
• Avoid time-sensitive information or use "Old Patterns" sections with details tags
• Maintain consistent terminology - select one term and use exclusively
• Use forward slashes in all paths (never Windows-style backslashes)
• Provide defaults for all options rather than excessive choices
• Justify configuration parameters - no "magic numbers"
• Include error handling in scripts with helpful messages
• List required packages and verify availability

Anti-Patterns to Avoid

• Windows-style paths
• Excessive options without defaults
• Deeply nested file references (keep to one level)
• Assuming tools are pre-installed
• Time-sensitive information without caveats
• Vague activation language
• Loading everything upfront instead of progressive disclosure

Bundled Resources

Skills can include optional bundled resources organized in three directories:

scripts/

Executable code (Python/Bash/etc.) for tasks requiring deterministic reliability or repeatedly rewritten operations.

When to include:

• Same code is rewritten repeatedly
• Deterministic reliability needed
• Complex operations benefit from pre-tested scripts

Examples from real skills:

• PDF skill: fill_fillable_fields.py, extract_form_field_info.py - PDF manipulation utilities
• DOCX skill: document.py, utilities.py - document processing modules
• This skill: init_skill.py - creates new skills from template, quick_validate.py - validates skill structure

Benefits:

• Token efficient (can execute without loading into context)
• Deterministic behavior
• Reusable across multiple invocations

Note: Scripts may still need to be read by Claude for patching or environment-specific adjustments.

references/

Documentation and reference material loaded into context to inform Claude's process and thinking.

When to include:

• Documentation Claude should reference while working
• Information too lengthy for main SKILL.md
• Domain-specific knowledge, schemas, or specifications

Examples from real skills:

• Product management: communication.md, context_building.md - detailed workflow guides
• BigQuery: API reference documentation and query examples
• Finance: finance.md - schemas, mnda.md - NDA template, policies.md - company policies

Benefits:

• Keeps SKILL.md lean and focused
• Loaded only when Claude determines it's needed
• Supports progressive disclosure

Best practice: If files are large (>10k words), include grep search patterns in SKILL.md to help Claude find specific sections.

assets/

Files not loaded into context, but used within the output Claude produces.

When to include:

• Files needed in final output
• Templates to be copied or modified
• Boilerplate code or starter projects

Examples from real skills:

• Brand guidelines: logo.png, slides_template.pptx - brand assets
• Frontend builder: hello-world/ - HTML/React boilerplate directory
• Typography: font.ttf, font-family.woff2 - font files

Common asset types:

• Templates: .pptx, .docx, boilerplate directories
• Images: .png, .jpg, .svg
• Fonts: .ttf, .otf, .woff, .woff2
• Boilerplate code: project directories, starter files
• Data files: .csv, .json, .xml, .yaml

Benefits:

• Separates output resources from documentation
• Enables Claude to use files without loading into context
• Provides consistent starting points for generated content

Skill Quality Checklist

See references/quality-checklist.md for the full checklist with examples.

Key requirements:

• YAML frontmatter with valid name and description (including "what" AND "when")
• Main content under 500 lines
• Progressive disclosure (reference files for detailed content)
• Changes committed to git repository

Workflow for Creating/Updating Skills

Creating a New Skill

Follow these steps in order. Skip a step only if there's a clear reason it's not applicable.

Step 1: Understanding with Concrete Examples

Clearly understand concrete examples of how the skill will be used. Skip this step only when usage patterns are already clearly understood.

Ask questions to gather specific use cases:

• "What functionality should this skill support?"
• "Can you give examples of how this skill would be used?"
• "What would a user say that should trigger this skill?"

Example questions for an image-editor skill:

• "What functionality should the image-editor skill support? Editing, rotating, anything else?"
• "I can imagine users asking for things like 'Remove the red-eye from this image' or 'Rotate this image'. Are there other ways you imagine this skill being used?"

Important: Avoid overwhelming users with too many questions. Start with the most important and follow up as needed.

Conclude when there's a clear sense of the functionality the skill should support.

Step 2: Plan Reusable Resources

Analyze each concrete example to identify what bundled resources would be helpful:

For each example, consider:

1. How to execute it from scratch
2. What scripts, references, and assets would make repeated execution easier

Example analyses:

PDF rotation: "Help me rotate this PDF"

• Rotating PDFs requires rewriting the same code each time
• → Include scripts/rotate_pdf.py

Frontend webapp: "Build me a todo app" or "Build me a dashboard"

• Requires same HTML/React boilerplate each time
• → Include assets/hello-world/ template directory

BigQuery queries: "How many users logged in today?"

• Requires re-discovering table schemas each time
• → Include references/schema.md with table documentation

Create a list of reusable resources to include: scripts/, references/, assets/ files.

Step 3: Initialize the Skill

Create the skill directory structure using the initialization script:

~/.claude/skills/skill-management/scripts/init_skill.py <skill-name> --path ~/.claude/skills

The script will:

• Create the skill directory with proper structure
• Generate SKILL.md template with frontmatter and TODO placeholders
• Create example files in scripts/, references/, and assets/ directories

After initialization, customize or delete the generated example files as needed.

Step 4: Implement Bundled Resources

Start by implementing the reusable resources identified in Step 2:

• Add scripts to scripts/
• Add reference documentation to references/
• Add templates/assets to assets/

Note: This may require user input (e.g., brand assets, templates, domain documentation).

Delete any example files and directories not needed for the skill.

Step 5: Complete SKILL.md

Write SKILL.md content following the writing guidelines (imperative form, concise, focused).

Answer these questions in SKILL.md:

1. What is the purpose of the skill? (a few sentences)
2. When should the skill be used? (specific triggers)
3. How should Claude use the skill in practice? (reference bundled resources)

Remember:

• Keep under 500 lines
• Use progressive disclosure (reference files instead of embedding everything)
• Include concrete examples
• Focus on information Claude doesn't already know

Step 6: Validate the Skill

Run the validation script to check for common issues:

~/.claude/skills/skill-management/scripts/quick_validate.py ~/.claude/skills/<skill-name>

Fix any validation errors reported.

Step 7: Test the Skill

Create test scenarios and verify the skill works:

1. Ask questions that should trigger it
2. Check if Claude invokes the skill
3. Verify the skill provides value
4. Adjust triggers if not invoked when expected

Step 8: Commit to Repository

cd ~/.claude/skills
git add skill-name/
git commit -m "Add [skill-name] skill: [brief description]

Detailed explanation of what the skill does and why it's needed.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>"

Updating an Existing Skill

1. Read current skill: Review SKILL.md and related files
2. Identify improvements: Based on usage patterns or new requirements
3. Make focused changes: Edit specific sections, maintain structure
4. Validate changes: Run validation script to catch any issues

   ~/.claude/skills/skill-management/scripts/quick_validate.py ~/.claude/skills/<skill-name>
   

5. Verify quality checklist: Ensure still meets all criteria
6. Test changes: Verify skill still triggers correctly
7. Commit to repository:

   cd ~/.claude/skills
   git add [skill-directory]/
   git commit -m "Improve [skill-name]: [specific changes made]
   
   Detailed explanation of changes and rationale.
   
   🤖 Generated with [Claude Code](https://claude.com/claude-code)
   
   Co-Authored-By: Claude <noreply@anthropic.com>"
   

Git Repository Management

CRITICAL: Skills are version-controlled in a git repository at ~/.claude/skills/.

After ANY skill modification:

1. Navigate to skills directory: cd ~/.claude/skills
2. Check status: git status
3. Add changes: git add [skill-directory]/
4. Commit with descriptive message:

   git commit -m "Action [skill-name]: brief description
   
   Detailed explanation of changes and rationale.
   
   🤖 Generated with [Claude Code](https://claude.com/claude-code)
   
   Co-Authored-By: Claude <noreply@anthropic.com>"
   

5. Verify clean state: git status

Common Git Commands

# Check current status
git status

# See what changed
git diff [file]

# Add specific skill
git add skill-name/

# Commit with message
git commit -m "message"

# View recent commits
git log --oneline -5

# Push changes (if using remote)
git push

Examples

See references/quality-checklist.md for description examples and structure guidelines.

Special Considerations

Testing New Skills

After creating a skill, test it by:

1. Asking a question that should trigger it
2. Checking if Claude invokes the skill
3. Verifying the skill provides value
4. Adjusting triggers if not invoked when expected

Refining Triggers

If a skill isn't being invoked when it should:

• Add more specific trigger phrases to description
• Use "proactively when" language
• List explicit keywords and contexts
• Consider if scope is too narrow or too broad

Documentation References

For the most current best practices, reference:

• Claude Code Skills Best Practices: https://docs.claude.com/en/docs/agents-and-tools/agent-skills/best-practices
• Agent Skills Overview: https://docs.claude.com/en/docs/agents-and-tools/agent-skills/overview
• Skills Quickstart: https://docs.claude.com/en/docs/agents-and-tools/agent-skills/quickstart

Reminder

DO NOT FORGET: After making any changes to skills in ~/.claude/skills/, you MUST commit them to the git repository. This ensures changes are tracked and can be shared/synced. The skills directory is version-controlled specifically for this purpose.