Agent Skill Builder

Creates new Claude skills from scratch or translates ChatGPT projects into the Claude skill format.

Trigger Conditions

Invoke this skill when the user:

• Wants to "create a skill" or "make a new skill"
• Wants to "translate a ChatGPT project" or "convert a ChatGPT Project to Claude"
• Mentions "migrate prompts" or "convert ChatGPT project"
• Asks about "slash commands" or "custom commands"
• Says "create a /command" or similar

Two Modes

Mode 1: Create New Skill (/skill-builder new [name])

Creates a fresh Claude skill with proper structure and best practices.

Mode 2: Translate ChatGPT Project (/skill-builder translate [name])

Converts an existing ChatGPT project into a Claude skill.

Core Principles (from Anthropic)

Concise is Key

The context window is a public good. Skills share it with system prompts, conversation history, other skills' metadata, and user requests.

Default assumption: Claude is already very smart. Only add context Claude doesn't already have. Challenge each piece of information: "Does Claude really need this explanation?" and "Does this paragraph justify its token cost?"

Prefer concise examples over verbose explanations.

Set Appropriate Degrees of Freedom

Match specificity to task fragility:

Think of Claude exploring a path: a narrow bridge with cliffs needs guardrails (low freedom), while an open field allows many routes (high freedom).

Progressive Disclosure

Skills use a three-level loading system:

1. Metadata (name + description) - Always in context (~100 words)
2. SKILL.md body - When skill triggers (<5k words ideal, <500 lines)
3. Bundled resources - As needed (unlimited, scripts execute without loading)

Key principle: Keep SKILL.md lean. Move variant-specific details to reference files.

Workflow: Create New Skill

Step 1: Understand with Concrete Examples

Ask the user:

1. Skill name: What should the /slash-command be called?
2. Purpose: What does this skill do?
3. Examples: "Can you give examples of how this skill would be used?"
4. Triggers: "What would a user say that should trigger this skill?"

Avoid overwhelming with questions. Start with the most important ones.

Step 2: Plan Reusable Contents

Analyze each example to identify:

• Scripts (scripts/): Code rewritten repeatedly or needing deterministic reliability
• References (references/): Documentation Claude should consult while working
• Assets (assets/): Files used in output (templates, images, fonts)

Step 3: Create Directory Structure

mkdir -p ~/.claude/skills/[skill-name]

Step 4: Write SKILL.md

Choose a structure pattern:

Workflow-Based (sequential processes)

## Overview → ## Workflow Decision Tree → ## Step 1 → ## Step 2...

Task-Based (tool collections)

## Overview → ## Quick Start → ## Task Category 1 → ## Task Category 2...

Reference/Guidelines (standards/specifications)

## Overview → ## Guidelines → ## Specifications → ## Usage...

Capabilities-Based (integrated systems)

## Overview → ## Core Capabilities → ### 1. Feature → ### 2. Feature...

Use this template:

---
name: [skill-name]
description: [What it does AND when to use it. Include specific triggers/contexts. Max 200 chars. This is the PRIMARY triggering mechanism.]
version: "1.0.0"
---

# [Skill Name]

[Brief description - 1-2 sentences]

## Instructions

[Step-by-step guidance for Claude]

## Output Format

[Expected output structure]

## Examples

[Concrete input/output pairs]

## Constraints

[Limitations and rules]

Step 5: Reference Files (if needed)

For skills with multiple domains or variants:

skill-name/
├── SKILL.md (overview + navigation)
└── references/
    ├── variant-a.md
    ├── variant-b.md
    └── variant-c.md

Reference from SKILL.md:

## Advanced Features

- **Form filling**: See [forms.md](references/forms.md)
- **API reference**: See [api.md](references/api.md)

Step 6: Test the Skill

1. Invoke directly: /skill-name
2. Let Claude auto-invoke by matching description keywords
3. Check /context to verify skill is loaded

Workflow: Translate ChatGPT Project to Claude Skill

Step 1: Extract ChatGPT Components

Request from user:

• ChatGPT system prompt/instructions
• Knowledge files (PDFs, docs, etc.)
• Conversation starters (optional)
• Actions/API configurations (if any)

Step 2: Map ChatGPT Project → Claude

Step 3: Convert Instructions

Step 4: Convert Knowledge Files

1. PDFs/Docs: Convert to Markdown
2. Large references: Put in references/ subfolder
3. Templates: Put in assets/ subfolder
4. Examples: Include in SKILL.md or references/examples.md

If files are large (>10k words), include grep search patterns in SKILL.md.

Step 5: Set Frontmatter

---
name: skill-name                    # lowercase, hyphens only
description: What AND when          # PRIMARY triggering mechanism
version: "1.0.0"                    # optional
disable-model-invocation: true      # if user-only
user-invocable: false               # if Claude-only
context: fork                       # if needs isolation
agent: Explore                      # if using subagent
allowed-tools: Read, Grep, Glob     # if restricting tools
---

Step 6: Update CLAUDE.md

Add entry to the skills table:

| `skill-name` | User wants to [action]; mentions "[keyword1]," "[keyword2]" |

ChatGPT-Specific Translation Notes

ChatGPT "Memory" → Claude

ChatGPT memory has no direct equivalent. Options:

• Store in ~/.claude/memory.json
• Use project-specific context files
• Include as "background" section in skill

ChatGPT "Actions" → MCP Servers

ChatGPT Actions (API calls) require MCP server setup in ~/.claude/mcp.json.

ChatGPT "Code Interpreter" → Claude Tools

Claude has native access to: Bash, Read, Write, Edit, Glob, Grep

ChatGPT Conversation Starters

Document in skill's "## Examples" section or CLAUDE.md usage hints.

What NOT to Include

Do NOT create extraneous documentation:

• README.md
• INSTALLATION_GUIDE.md
• QUICK_REFERENCE.md
• CHANGELOG.md

Skills should only contain information needed for an AI agent to do the job. No auxiliary context about creation process, setup procedures, or user-facing documentation.

Frontmatter Fields Reference

Writing Good Descriptions

The description is the PRIMARY triggering mechanism. Claude reads it to decide when to use the skill.

Bad:

description: Helps with code

Good:

description: Reviews code for security, performance, and best practices. Use when user asks to "review code," "check this code," "security audit," or wants feedback on code quality.

Include:

• What the skill does
• When to use it (keywords, triggers, scenarios)
• Max 200 characters

String Substitutions

Skill Locations

After Creation Checklist

• SKILL.md created with proper frontmatter
• Description includes what AND when (triggers)
• Knowledge files converted to Markdown in references/
• SKILL.md under 500 lines (split if larger)
• CLAUDE.md updated with skill entry
• Skill tested with /skill-name
• Auto-invocation tested with matching keywords
• No extraneous documentation files created