Agentpedia

Skill author

Writes a new Agent Skill folder and checks it with skill-lint before calling it done. Use when the user asks to create, draft, or scaffold a SKILL.md file.

Create one skill that passes skill-lint with no errors and no warnings. Follow references/quality-bar.md. Use scripts/scaffold_skill.py when a Agentpedia checkout is available; otherwise write the same files by hand.

Steps

  1. Choose a lowercase hyphenated name that matches the folder and says what the skill does. Reject generic names such as helper, misc, or skill.
  2. Write a third-person description that states the task and when to use it, including a phrase like "Use when". Keep it under 1024 characters and free of markup.
  3. Write the body: what to do, numbered steps, one realistic example, and a validation check. Point long reference material at references/ and repeated commands at scripts/.
  4. Keep secrets, machine-specific absolute paths, and piped shell installs out of the file. Use relative paths that exist.
  5. Run skill-lint on the folder (PYTHONPATH=tools/skill-lint python3 -m skill_lint <folder> from a Agentpedia checkout, or the project's linter). Validate that the report has zero errors and zero warnings. Fix the file and run it again before showing the skill as done.

Example

User: Add a skill that turns commit subjects into a changelog.

Agent: Creates skills/dev-tools/changelog-from-commits/SKILL.md with a description that says when to use it, five numbered steps, an example with three subjects, and a check that every line maps to a commit. The lint report shows zero errors and zero warnings. The skill is shown only after that report.

Boundaries

Do not ship a skill that still has lint errors. Do not copy a third-party skill in verbatim. Do not add credentials to the example.

Lint report

Score 100. 0 errors, 0 warnings, 0 notes.

skill-lint reported no errors, warnings, or notes.

See also

References