| name | skill-authoring |
| description | Use this skill when creating, reviewing, or improving SKILL.md files, agent skills, workflow instructions, trigger descriptions, progressive disclosure structure, bundled skill examples, or skill-selection behavior. Also use it when the user asks for best practices for skills. |
Skill Authoring
Principles
- Put the trigger contract in frontmatter:
name is short and stable; description says when to use the skill and includes realistic user-intent keywords.
- Keep the body procedural, not encyclopedic. Include what the model is likely to miss: local conventions, edge cases, validation steps, and preferred defaults.
- Use progressive disclosure. Keep
SKILL.md small; reference extra files only when the runtime can load them on demand.
- Prefer concrete checklists, gotchas, and output templates over generic advice.
- Avoid instructions that try to override user intent, system safety rules, or unrelated tasks.
Description Checklist
- Starts with "Use this skill when..."
- Names the task class and adjacent phrases users are likely to type.
- Includes near-boundaries where it should trigger.
- Avoids claiming every generic task in the domain.
- Stays under 1024 characters.
Body Checklist
- One clear workflow.
- A concise output contract.
- Non-obvious gotchas.
- Validation or self-check steps when the task can be verified.
- No broad background material the model already knows.
RustyLLM Constraint
RustyLLM injects the selected SKILL.md text into the system prompt. It does not execute skill scripts or lazily read references/ files, so example skills for RustyLLM should be self-contained unless the user explicitly provides external content.
