Writing Skills Skill

Writing Skills is the meta skill of Superpowers (skill source). It is used when creating a new skill, editing an existing one, or checking that a skill works before deployment, and its central idea is that writing a skill is test-driven development applied to process documentation. The Superpowers contribution guide requires it for any skill change.

Its iron law mirrors TDD: "No skill without a failing test first", and that applies to edits too.

Key Features

• TDD mapping: the test case is a pressure scenario run with a subagent; RED is the agent breaking the rule without the skill; GREEN is compliance with the skill; REFACTOR is closing loopholes.
• When to write one: techniques that were not obvious, reusable across projects and broadly useful. Not for one-offs, project conventions, or rules a regex could enforce.
• Skill types: technique, pattern and reference, each with its own testing approach.
• Descriptions are triggers: the description should say when to use the skill ("Use when..."), not summarize its workflow. The skill explains that a workflow summary led an agent to do one review where the skill required two.
• Bulletproofing: close each loophole explicitly, build a rationalization table from real test runs, and add a red-flags list.
• Token efficiency: keep skills lean and move heavy reference or reusable tools into separate files.
• Supporting files: Anthropic's skill best practices, persuasion principles, graphviz conventions and a guide to testing skills with subagents.

Use Cases

• Writing your own team skills for Claude Code, Codex or Gemini CLI.
• Fixing a skill that agents keep ignoring or misapplying.
• Contributing to Superpowers itself.

Pricing

Free and open source under the MIT license. Prime Radiant, the company behind Superpowers, sells commercial support to enterprises. The practical cost is tokens: process skills add questions, reviews and subagent runs.

Getting Started

Superpowers installs as one plugin, so this skill arrives with the rest of the library (15 skills in v6.4.1). In Claude Code run /plugin install superpowers@claude-plugins-official. Cursor uses /add-plugin superpowers, Gemini CLI uses gemini extensions install https://github.com/obra/superpowers, and the README lists commands for Codex, GitHub Copilot CLI, OpenCode, Pi and other harnesses. Skills trigger on their own; you can also ask for one by name. Then ask the agent to "write a skill for X"; expect it to run a baseline scenario before drafting.

Limitation: pressure testing with subagents takes real time and tokens, and results vary by model. It is heavier than simply writing a markdown file.

FAQ

Where do personal skills live?

In the runtime's skills directory, ~/.claude/skills/ on Claude Code; several runtimes also read ~/.agents/skills/.

Is this the same as Anthropic's skill creator?

No. Skill Creator focuses on templates; this skill focuses on testing that agents actually comply.

Alternatives

• Skill Creator: Anthropic's official meta skill.
• Claude Skills: how Claude Code loads skills.
• Skills Explorer: browse existing skills before writing a new one.

Conclusion

The most rigorous guide to writing agent skills we have listed. Worth reading even if you never install Superpowers. More in the skills hub.