Create agent skill

Boundaries

• Answer planning-only requests in conversation: plans, designs, critiques, outlines, requirements, and discussions of skill structure. For mixed requests, resolve the plan first and write files once the implementation path is clear.
• Treat named tools, skills, frameworks, and workflows as context for the plan; switch to file creation only on the user's explicit ask. When file-writing intent is ambiguous and edits would be surprising, ask before editing.
• To review an existing skill file, check it against this skill's rules and report findings in conversation.
• When the user asks to evaluate, prune, or iterate on an existing skill, load references/evaluating-skills.md.

Workflow

1. Confirm scope.
   • Identify the concrete artifact to create or update: skill directory, SKILL.md, references, scripts, assets, or inventory docs.
   • Identify the task, workflow, project convention, or domain the skill covers.
   • For a shortcut that packages instructions the user retypes, skip steps 2–5 and 7: name it per step 3, give it user-invoked frontmatter with a one-line description, put the instructions in the body, then apply step 6.
   • Capture the prompts, contexts, files, and workflows that should trigger the skill, plus adjacent requests and near-misses that should not.