How to use the Skill Creator agent skill (skill-creator)

What the skill-creator skill makes the agent do

The skill tells the agent when a pattern is worth turning into a skill: when it is used repeatedly and the AI needs guidance, when project conventions differ from generic practice, when workflows need step-by-step instructions, or when decision trees help the AI choose an approach. It also says when not to bother, such as trivial or one-off patterns that normal documentation covers better.

When the agent writes a skill, it follows the structure in the source: a SKILL.md with frontmatter (name, description, license, metadata author and version), then sections in a fixed order. The description is kept to one quoted line with trigger words first, and the body is kept concise. Templates and schemas go in an assets/ folder, and links to local reference docs go in references/.

• Decides whether a skill is worth creating at all
• Structures SKILL.md in a fixed section order
• Keeps the frontmatter description short and valid
• Puts templates in assets/ and reference docs in references/

Install the skill-creator skill

Download the skill's ZIP from this page. Then choose one of two routes.

For a personal setup, unzip it into ~/.claude/skills/. For a single project, unzip it into that project's .claude/skills/ folder. Either way, the skill folder should contain SKILL.md and the references/skill-style-guide.md file.

In the Claude apps, you can instead upload the ZIP under Settings → Capabilities → Skills.

• Personal: ~/.claude/skills/
• Project: .claude/skills/ inside the repository

How to trigger it with natural requests

You do not need a special command. The skill's description lists trigger words such as new skills, agent instructions, and documenting AI usage patterns, so ask in those terms. The agent will load the skill when your request matches.

Be specific about the pattern you want captured. A request that names the repeated task and the conventions it should follow gives the agent more to work with than a bare request for a new skill.

• "Create a new skill for writing our database migration notes."
• "Write agent instructions for how we name API test files."
• "Document this recurring review workflow as a skill."

Examples of what to ask for

The skill is most useful when the thing you want to capture repeats. Good candidates are a team's commit or ticket format, a multi-step release checklist, or a set of project conventions that generic advice gets wrong.

Ask the agent to explain its decision first. Because the skill includes a rule against creating skills for one-off patterns, a good response should say whether the pattern qualifies and why.

What it needs and what it does not do

The source lists no requirements. It does not run scripts and does not use the network. The skill only changes how the agent writes instruction files; it does not run code or touch your data by itself.

The skill also depends on a particular format. It assumes SKILL.md-style instructions and refers to a style guide. If your project does not use that format, the rules will not fit well. If the repo has docs/skill-style-guide.md, the agent is told to follow that first, and otherwise to use the bundled references/skill-style-guide.md copy.

• Requires: nothing listed
• Runs scripts: no
• Uses network: no

Safety notes and review

Generated skills become instructions that an agent follows later. A bad or overbroad skill can steer future behaviour, and the instructions are unverified text. Read every new SKILL.md before you commit it or rely on it, and check that its trigger words and scope are what you intended.

Keep new skills small. The source sets a target of 180 to 450 tokens for the skill body, with a hard maximum of 1000, so a sprawling generated file is a warning sign.

• Review the description, scope and every instruction before use
• Check that referenced files exist locally
• Remove any skill that duplicates an existing one