Skill Creator

Skill Creator

Generate well-structured SKILL.md files for Claude Code skills. Reads existing skills as reference, generates correct format, validates, and writes to .claude/skills/.

Prerequisites

No external CLI needed. Uses Read, Write, Edit, Grep, and Glob tools.

SKILL.md Format Reference

Every SKILL.md must follow this structure:

---
name: skill-name                    # kebab-case, matches directory name
description: >                      # Single paragraph, mentions trigger phrases
  What the skill does and when to use it.
allowed-tools:                      # Only tools actually needed
  - Bash
  - Read
  - Write
model: haiku|sonnet|opus            # haiku for simple, sonnet for complex
user-invocable: true                # Always true for user-facing skills
when_to_use: >                      # Explicit trigger phrases
  When to invoke. Triggers: "phrase 1", "phrase 2".
argument-hint: ""             # CLI-style hint shown in autocomplete
---

# Title

Brief one-line description.

## Prerequisites

Install command or "No install needed."

## Commands

Key CLI commands in a table or code blocks.

## Usage Examples

Realistic examples with bash code blocks.

## Rules

- Bullet list of behavioral rules
- Check if CLI installed first
- Handle missing env vars gracefully
- Keep responses concise

Workflow

Step 1: Gather Requirements

From user input, extract:

• Skill name: kebab-case identifier (e.g., my-skill)
• CLI tool(s): what commands power it
• Purpose: what task it accomplishes
• Trigger phrases: natural language patterns that should invoke it
• Model: haiku (simple/fast) or sonnet (complex/reasoning)

Step 2: Read Reference Skills

# Read 2-3 similar existing skills for pattern reference
ls /path/to/.claude/skills/

Use Read tool on relevant examples (e.g., apple-notes/SKILL.md, github-ops/SKILL.md).

Step 3: Check for Conflicts

# Check if skill already exists
ls .claude/skills/ | grep ""

Step 4: Generate SKILL.md

Generate the full SKILL.md content following the format reference above. Key rules:

• name field must match the directory name exactly
• description must include trigger phrases ("Use when user says...")
• when_to_use must list explicit trigger phrases after "Triggers:"
• allowed-tools must only include tools the skill actually uses
• Prerequisites section must include exact install commands
• Rules section must include: check CLI installed, handle missing env vars, keep output concise

Step 5: Create and Write

# Create the skill directory
mkdir -p .claude/skills/

Use Write tool to create .claude/skills//SKILL.md.

Step 6: Validate

After writing, re-read the file and verify:

• [ ] Frontmatter is valid YAML (no syntax errors)
• [ ] name matches directory name
• [ ] description is non-empty and mentions triggers
• [ ] when_to_use has explicit trigger phrases
• [ ] allowed-tools list is accurate
• [ ] model is one of: haiku, sonnet, opus
• [ ] argument-hint is present
• [ ] Has all required sections: Prerequisites, Commands, Usage Examples, Rules
• [ ] Rules include CLI-installed check and concise-output guidance

Usage Examples

Create a skill from description:

User: "create a skill for managing Docker containers — list, start, stop, exec"
→ Reads docker-related reference skills
→ Generates skill-name: docker-control
→ Creates .claude/skills/docker-control/SKILL.md
→ Validates format
→ Reports: "Created .claude/skills/docker-control/SKILL.md"

Create a skill with API key:

User: "create a skill for sending SMS via Twilio"
→ skill-name: twilio-sms
→ Includes TWILIO_ACCOUNT_SID + TWILIO_AUTH_TOKEN + TWILIO_FROM env vars
→ Creates .claude/skills/twilio-sms/SKILL.md

Rules

• Always read 2+ existing skills as format reference before generating
• Never create a skill that duplicates an existing one — check first with Glob
• Skill directory name must match the name frontmatter field exactly
• If user provides a vague description, ask 3 clarifying questions before generating:

1. What CLI tool(s) power this skill?
2. What are the 3 most common user requests?
3. Should it be haiku (fast/simple) or sonnet (reasoning required)?

• After writing, always re-read the file to confirm it was written correctly
• Report the created file path in the final response
• If the skill requires env vars, always include a "Set up" section in Prerequisites
• Keep the Rules section actionable — each rule should be a single clear directive

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: claude-world
• Source: claude-world/claude-agent
• License: MIT

Install and usage instructions live in the source repository linked above.