project-scaffolder: Command for Claude Code

Create a New Agent Skill

Create a new Agent Skill for specialized tasks.

Arguments

$ARGUMENTS should contain: <skill-name> [description]

What is a Skill?

A Skill is a folder containing:

• SKILL.md - Main instructions with YAML frontmatter
• Optional scripts and resources
• Optional reference documentation

Skills use progressive disclosure: only the name and description are loaded initially. The full skill content is loaded when relevant.

Skill Creation Process

Step 1: Gather Information

Ask the user:

1. Skill name: Short, descriptive (e.g., "pdf-forms", "api-testing")
2. Purpose: What task does this skill help with?
3. When to use: What triggers should activate this skill?
4. Required tools: Any external dependencies?

Step 2: Create Skill Directory

skills/<skill-name>/
├── SKILL.md           # Main skill file
├── scripts/           # Optional helper scripts
│   └── helper.py
└── resources/         # Optional templates/examples
    └── template.md

Step 3: Generate SKILL.md

Create with this structure:

---
name: <skill-name>
description: <brief one-line description for discovery>
---

# <Skill Name>

## When to Use This Skill

<Clear description of when Claude should activate this skill>

## Prerequisites

<List any required tools, dependencies, or setup>

## Instructions

### Step 1: <First Step>
<Detailed instructions>

### Step 2: <Second Step>
<Detailed instructions>

...

## Reference Files

<Point to any additional documentation or scripts>

## Examples

### Example 1: <Common Use Case>
<Input/output examples>

## Verification

<How to verify the skill worked correctly>

Step 4: Progressive Disclosure Tips

Structure the SKILL.md for progressive disclosure:

1. Top section: Most commonly needed info
2. Middle sections: Detailed procedures
3. Bottom sections: Edge cases and examples
4. Separate files: Rarely-used reference material

Step 5: Test the Skill

1. Simulate a task that should trigger the skill
2. Verify Claude loads the skill appropriately
3. Check that instructions are clear and complete
4. Iterate based on observed behavior