| name | skill-creator |
| description | Design Claude-style Agent Skills — multi-file bundles with SKILL.md, references/ guides, and scripts/ helpers. Use when the user asks to create, author, or package an Agent Skill that extends an AI agent's capabilities with specialized knowledge, workflows, or tool integrations. |
| license | MIT |
Skill Creator
When to Use This Skill
Apply when the user asks to create, design, or package an Agent Skill — a folder of files that extends an AI agent's capabilities. Skills are used by agents like Claude, Cursor, and other MCP-compatible clients.
A Skill is needed when:
- An agent needs specialized domain knowledge it doesn't have by default
- A workflow requires multi-step instructions with reference material
- Helper scripts would automate repetitive agent tasks
- You want to package expertise so any agent can use it
Related skills: For building live React/HTML artifacts use artifact-creator. For single-file system prompts/.cursorrules use prompt-creator.
Output Format
You MUST emit EACH file of the skill as its own block using this EXACT marker syntax, so the interface can reconstruct the folder tree:
=== FILE: SKILL.md ===
---
name: example-skill
description: One sentence describing exactly when this skill should be used.
license: MIT
---
# Example Skill
High-level instructions go here.
See references/patterns.md for advanced patterns.
=== FILE: references/patterns.md ===
# Advanced Patterns
Detailed reference material goes here.
=== FILE: scripts/run.py ===
print("hello")
Rules:
- ALWAYS start with
SKILL.md.
- Use forward slashes for folders.
- Do NOT wrap file bodies in extra markdown code fences — put the raw file content directly after each marker line.
- Briefly explain the skill's folder structure in the chat BEFORE emitting the file blocks.
SKILL.md Structure
---
name: skill-name
description: One sentence describing exactly when this skill should be used.
license: MIT
---
# Skill Title
## When to Use This Skill
[Trigger conditions — when should an agent invoke this?]
## Core Instructions
[The main guidance — concise, high-signal]
## When NOT to Use
[Boundaries — what's out of scope]
## Related Skills
[Links to companion skills, if any]
Folder Structure
my-skill/
├── SKILL.md # Required — entry point, concise instructions
├── references/ # Optional — in-depth docs the agent loads on demand
│ ├── api-details.md
│ ├── patterns.md
│ └── examples.md
├── scripts/ # Optional — runnable helper scripts
│ ├── setup.py
│ └── validate.sh
└── assets/ # Optional — templates or static files
└── template.tsx
Principles
See references/structure.md for the full anatomy of a well-designed skill, and references/examples.md for annotated example skills.
Keep SKILL.md concise
SKILL.md is the entry point — an agent reads it to decide if the skill is relevant and to get the core instructions. Keep it under ~300 lines. Push long, detailed material into references/ files and link to them:
For advanced easing curves, see **references/easing.md**.
Description is the trigger
The description in YAML frontmatter is what an agent uses to decide whether to invoke the skill. Make it specific:
- ✅
Use when the user asks to create GSAP scroll-triggered animations with pinning, scrub, or snap.
- ❌
A skill about animations.
Reference files are loaded on demand
Agents don't read every file upfront — they read SKILL.md, then load reference files when they need deeper detail. So reference files should be self-contained topics that an agent can pull when needed.
