Agent Skill
skill-creator
Published by Anthropic · Apache-2.0 (LICENSE.txt in the skill folder)
Creates, edits and benchmarks new agent skills, including description tuning for reliable triggering.
How it works
The skill-creator skill is for building other skills. It walks through creating a skill from scratch, editing an existing one, and improving the description so the agent selects it at the right moments rather than too often or never. It also describes evaluation: running a skill against test prompts and comparing results to see whether a change helped. That makes it a maintenance tool for teams standardizing repeated workflows. The output is a skill package you review and version yourself, and benchmark results are only as good as the test prompts you write.
Features and capabilities
- Scaffold a new skill
- Tune descriptions for reliable triggering
- Run evaluations against test prompts
- Improve existing skills iteratively
Primary use cases
- Creating a new custom skill
- Improving skill triggering
- Benchmarking a skill
- Refactoring an existing skill
Installation and setup
/plugin marketplace add anthropics/skills
/plugin install example-skills@anthropic-agent-skills
Other agents (open Agent Skills standard): the same folder loads by copying it into the host's skills directory — Codex reads .agents/skills in a repository or ~/.agents/skills; Cursor reads .agents/skills or .cursor/skills and also reads .claude/skills; the ChatGPT desktop app installs skills through its Skills panel, while ChatGPT on the web and mobile only loads skills packaged in a plugin. The commands above are Claude Code's marketplace; Anthropic documents that host only, so follow each host's own setup page.Setup instructions are reproduced from the publisher. Explore4.ai never installs or executes a third-party skill on your behalf.
Requirements
Claude Code, a paid Claude.ai plan, or the Claude API. Skills load as folders containing a SKILL.md file.
Licence, pricing and security
- License
- Apache-2.0 (LICENSE.txt in the skill folder)
- Pricing
- Free — The skill is free to use. Running it needs Claude Code, a paid Claude.ai plan, or Claude API access.
- Security review
- Not reviewedSource verification is recorded separately from any security assessment.
- Security notes
- No independent security review is published for this skill. Its instructions and scripts are public in Anthropic's repository; read them before use and grant only the access the task needs.
Compatible AI Agents
Only the agents and platforms the publisher documents as supported are listed here.
Related AI tools
Tools that commonly host or pair with this kind of capability.
