Skill Creator
Create effective Claude Code skills following Anthropic's best practices.
Quick Start
Create a minimal skill in 3 steps:
1. Create the directory
mkdir -p ~/.claude/skills/my-skill-name
# Or for project-specific: mkdir -p .claude/skills/my-skill-name
2. Create SKILL.md
---
name: my-skill-name
description: [Action verb] [what it does]. Use when [scenarios]. Triggers: "[keyword1]", "[keyword2]".
---
# My Skill Name
## Instructions
[Clear, step-by-step guidance for Claude]
## Examples
[Concrete usage examples]
3. Test the skill
Ask Claude: "What skills are available?" to verify it loads, then test with your trigger phrases.
Frontmatter Reference
The SKILL.md file must start with YAML frontmatter between --- markers:
| Field | Required | Constraints | Description |
|---|---|---|---|
name | Yes | lowercase, hyphens only, max 64 chars | Skill identifier (should match directory name) |
description | Yes | max 1024 chars | What it does + when to use (Claude uses this for matching) |
allowed-tools | No | comma-separated | Restrict tool access for security |
model | No | full model name | Override conversation model |
allowed-tools Examples
# Read-only skill
allowed-tools: Read, Grep, Glob
# File editing skill
allowed-tools: Read, Write, Edit, Grep, Glob
# Bash with specific commands
allowed-tools: Read, Bash(git:*, npm:*, docker:*)
# Full access (omit field entirely)
# (no allowed-tools field)
Writing Effective Descriptions
Claude matches user requests against skill descriptions using semantic similarity. A good description is critical for proper skill activation.
The Description Formula
[Action verb(s)] + [what it does] + [target domain].
Use when [specific trigger scenarios].
Triggers: "[keyword1]", "[keyword2]", "[keyword3]".
Good vs Bad Descriptions
Good (specific, action-oriented, includes triggers):
description: Create, modify, debug, and convert trading indicators for TradingView and NinjaTrader. Use when writing new indicators, adding features, or converting between platforms. Triggers: "create indicator", "Pine Script", "NinjaScript".
Bad (vague, no triggers):
description: Helps with trading stuff.
Description Checklist
- Starts with action verbs (create, configure, debug, analyze, generate)
- Explains what the skill does
- Specifies the domain/context
- Includes "Use when..." guidance
- Lists trigger keywords users would naturally say
- Under 1024 characters
File Organization
Single-File Skills
For simple skills under 500 lines:
my-skill/
└── SKILL.md
Multi-File Skills (Progressive Disclosure)
For complex skills, split content to keep SKILL.md under 500 lines:
my-skill/
├── SKILL.md # Essential info + links (under 500 lines)
├── reference.md # Detailed documentation
├── examples.md # Usage examples
└── scripts/
└── helper.py # Utility scripts (executed, not loaded)
Why progressive disclosure?
- Only SKILL.md loads initially
- Claude reads supporting files only when needed
- Keeps context focused and efficient
- Allows bundling comprehensive documentation
Linking Supporting Files
In SKILL.md, link to supporting files:
## Additional Resources
- For complete API details, see [reference.md](reference.md)
- For usage examples, see [examples.md](examples.md)
## Utility Scripts
To validate input, run:
\`\`\`bash
python scripts/helper.py input.txt
\`\`\`
Skill Locations
| Location | Path | Scope | Use Case |
|---|---|---|---|
| Personal | ~/.claude/skills/ | All your projects | General-purpose skills |
| Project | .claude/skills/ | This repository only | Project-specific skills |
| Plugin | skills/ in plugin dir | Anyone with plugin | Distributed skills |
Precedence: Enterprise > Personal > Project > Plugin
Common Mistakes
1. Vague Description
# Bad
description: Helps with documents
# Good
description: Extract text, fill forms, and merge PDF files. Use when working with PDFs or document extraction. Triggers: "PDF", "fill form", "merge documents".
2. SKILL.md Too Long
If over 500 lines, split into supporting files. Claude's context is shared with conversation history.
3. Missing Trigger Keywords
Include words users naturally say:
- "create", "make", "build", "generate"
- "fix", "debug", "troubleshoot"
- "convert", "transform", "migrate"
- Domain-specific terms
4. Overly Broad allowed-tools
# Bad - too permissive
allowed-tools: Read, Write, Edit, Bash, WebFetch, WebSearch
# Good - only what's needed
allowed-tools: Read, Grep, Glob
5. Hardcoded Absolute Paths
# Bad
See /Users/john/projects/docs/reference.md
# Good
See [reference.md](reference.md)
6. No "When to Use" Guidance
Always include a section explaining when to use (and not use) the skill.
When to Use This Skill
Use this skill when:
- Creating a new skill from scratch
- Updating an existing skill
- Learning skill best practices
- Debugging why a skill isn't triggering
Do NOT use for:
- General coding tasks
- Tasks unrelated to skill creation
Skill Templates
For copy-paste templates, see TEMPLATE.md.
Complete Examples
For full working examples, see EXAMPLES.md.
Pre-Flight Checklist
Before finalizing a skill, see CHECKLIST.md.
Workflow for Creating a New Skill
- Define the purpose - What problem does this skill solve?
- Choose location - Personal (
~/.claude/skills/) or project (.claude/skills/)? - Write description first - Get the semantic matching right
- Draft SKILL.md - Start with template, customize
- Add supporting files - If needed for progressive disclosure
- Test activation - Verify skill triggers on expected phrases
- Iterate - Refine based on real usage
Testing Skills
Verify Skill Loads
Ask Claude: "What skills are available?"
Test Trigger Phrases
Try phrases from your description's trigger keywords:
- "Help me [trigger keyword]"
- "I need to [action verb from description]"
- "[Domain term] assistance"
Debug Non-Triggering Skills
If skill doesn't activate:
- Check SKILL.md syntax (valid YAML, no tabs)
- Verify description includes semantic keywords
- Ensure no blank lines before opening
--- - Run
claude --debugfor error messages
