Skill creator

Skill Creator

Creates new Agent Skills through interactive wizard or template-based generation. A meta-skill for bootstrapping Claude Code capabilities.

When This Skill Activates

• User asks to "create a skill", "make a skill", "generate a skill", "scaffold a skill"
• User mentions "skill template", "new skill", "custom skill", "skill development"
• User wants to extend Claude with custom capabilities or workflows
• User asks about building, authoring, or writing skills

Workflow

Step 1: Choose Mode

Ask the user which mode they prefer:

1. Interactive Wizard: Guided Q&A to build a skill from scratch
2. Template-Based: Start from a pre-built template and customize
3. Hybrid: Template foundation + interactive customization

Step 2: Gather Requirements (Interactive Mode)

Ask these questions in order:

1. Skill Name: What should the skill be called? (kebab-case, max 64 chars)
2. Purpose: What does this skill do? (one sentence)
3. Trigger Phrases: When should Claude use this skill? (list 3-5 specific phrases users might say)
4. Tool Permissions: Which tools does the skill need? (Read, Write, Edit, Bash, Grep, Glob - request minimal)
5. Skill Type:
   • Simple: Single SKILL.md only
   • With Scripts: SKILL.md + Python/bash helper scripts
   • With Reference: SKILL.md + reference.md for technical docs
   • With Examples: SKILL.md + examples.md for use cases
   • Complex: All of the above
6. Target Location:
   • Personal: ~/.claude/skills/ (just for you)
   • Project: .claude/skills/ (shared with team)
   • Standalone: Portable plugin structure (for distribution)

Step 3: Select Template (Template Mode)

Available templates in templates/:

Template	Purpose	Best For
pr-reviewer	Review pull requests against standards	Teams with code review workflows
commit-helper	Generate commit messages from diffs	Projects with commit conventions
api-caller	Call external APIs with auth	Integrations with external services
code-analyzer	Analyze code quality and patterns	Codebases needing quality checks
data-processor	Process CSV/JSON data files	Data transformation workflows

Step 4: Generate Core Files

Always generate:

1. SKILL.md - Complete with:

   • YAML frontmatter (name, description, allowed-tools)
   • When This Skill Activates section
   • Main workflow/instructions
   • Examples section
   • Troubleshooting section

2. .claude-plugin/plugin.json - With:

   • Skill name and version
   • Description matching SKILL.md
   • Author info (prompt for name/email)
   • License (default: MIT)

3. README.md - With:

   • Skill overview
   • Installation instructions
   • Quick start example
   • Requirements

4. examples.md (if skill type includes it) - With:

   • 2-3 complete usage examples
   • Input/output samples
   • Common scenarios

Conditionally generate:

• scripts/ directory for automation skills
• reference.md for technical/API skills
• templates/ for generator skills

Step 5: Validate Before Writing

Before creating files, validate:

• Name is kebab-case (lowercase letters, numbers, hyphens only)
• Name is <= 64 characters
• Description is <= 1024 characters
• Description includes specific trigger phrases
• Tool permissions are minimal (don't request tools the skill won't use)
• YAML frontmatter is valid (proper --- delimiters)
• File paths use forward slashes

Step 6: Create Files

Use Write tool to create files in the selected location.

For target location = "standalone", create full plugin structure:

skill-name/
├── SKILL.md
├── .claude-plugin/
│   └── plugin.json
├── README.md
└── examples.md

Step 7: Verification

After creating files, tell the user:

1. Where files were created
2. How to verify the skill works:
   • What Skills are available? should list the new skill
   • Test with a query matching the description
3. Next steps (edit SKILL.md to customize, test with example queries)

Template Details

pr-reviewer.md

Reviews pull requests against team standards. Checks for common issues, provides structured feedback format, integrates with git workflow.

commit-helper.md

Generates conventional commit messages from git diffs. Supports custom commit conventions, formats summaries and body, handles commit types.

api-caller.md

Integrates with external APIs. Handles authentication (API keys, OAuth), rate limiting, error responses, request/response parsing.

code-analyzer.md

Analyzes code quality. Detects anti-patterns, suggests improvements, generates quality reports, supports multiple languages.

data-processor.md

Processes data files (CSV, JSON, YAML). Validates schemas, transforms data, handles large files, generates outputs.

Best Practices for Generated Skills

Description Writing

• Start with "Use when..." for clarity
• Include 3-5 specific trigger scenarios
• Mention what makes this skill unique
• Keep under 1024 characters

Tool Permissions

• Request only necessary tools
• Use specific constraints (e.g., Bash(python:*))
• Explain why each tool is needed

Progressive Disclosure

• Keep SKILL.md focused (under 500 lines)
• Reference supporting files for deep details
• Don't duplicate content across files

Common Patterns

Skill with Scripts

When a skill needs helper scripts:

## Utility Scripts

This skill includes helper scripts for complex operations:

```bash
# Validate input
python scripts/validate.py input.txt

# Transform data
python scripts/transform.py input.csv output.json

### Skill with Reference
For technical/API skills, reference detailed docs:

```markdown
## Technical Reference

For complete API details, see [reference.md](reference.md).

Skill with Environment Variables

If the skill needs environment setup:

## Requirements

Set these environment variables:
- `API_KEY`: Your service API key
- `ENDPOINT`: API endpoint URL

Install dependencies:
```bash
pip install requests pydantic

## Troubleshooting

### Skill Not Appearing
- Verify SKILL.md path is correct
- Check YAML frontmatter is valid
- Run `claude --debug` to see loading errors

### Description Not Triggering
- Add more specific trigger phrases
- Include keywords users would naturally say
- Test with different query phrasings

### Validation Errors
- Check name is kebab-case only
- Verify description length < 1024 chars
- Ensure YAML has proper `---` delimiters

## Examples

See [examples.md](examples.md) for complete walkthroughs of creating skills in each mode.