Ask the user these 3 questions in a single message:

Name — lowercase hyphenated slug (e.g. my-skill)
Trigger — when should Claude invoke this? The description is the only thing Claude sees when deciding which skill to load — be specific: first sentence says what it does, second says "Use when [triggers/keywords]". Max 1024 chars.
Behavior — describe what the skill does, step by step. Include: any early-exit conditions and what to output, the expected output format (if any), and any exact strings the skill should produce (use quotes).

After receiving answers:

Create skills/<name>/SKILL.md with frontmatter (name + description) and a compact body derived from the behavior description. Use backticks for exact output strings. End with No explanation. only if the skill produces a specific output.
Add "./skills/<name>" to the skills array in .claude-plugin/plugin.json
Verify against checklist (fix silently if needed):
- Description includes "Use when [triggers]"
- No time-sensitive information
- SKILL.md under 100 lines
- Exact output strings in backticks
Output exactly Created skills/<name>/SKILL.md — run /reload-plugins to activate. — then, on a new line, optionally suggest a better name or description if the given ones are suboptimal (e.g. `better-name` might be more descriptive than `given-name`). Skip if both are already good.

create-skill