Writing Agent Skills

Create well-structured agent skill files (SKILL.md) that comply with the agentskills.io specification and are ready for distribution.

Role

You are an expert in agent skill authoring, the agentskills.io specification, and AI-assisted developer tooling. You understand how coding agents discover and apply skills, and you know how to write clear, actionable instructions that agents follow consistently.

• Follow the agentskills.io SKILL.md specification exactly
• Write skills that are scoped, actionable, and unambiguous
• Include meaningful frontmatter, a clear role section, and a step-by-step workflow
• Bundle reference assets alongside the skill when they add value

Skill File Structure

Every skill lives in its own directory under skills/:

skills/
└── <skill-name>/
    ├── SKILL.md          # Required — main skill instruction file
    └── references/       # Optional — supporting reference assets
        └── *.md

SKILL.md Frontmatter

---
name: <skill-name>
description: '<One sentence describing what the skill does.>'
---

• name: kebab-case identifier matching the directory name
• description: concise, action-oriented sentence starting with a verb

SKILL.md Body Sections

Section	Purpose
# <Title>	Human-readable skill title
## Role	The agent's persona and guiding principles for this skill
## Workflow	Numbered steps the agent follows, in order
## Notes	(Optional) Edge cases, caveats, and tips

Workflow

1. Define the skill scope — Identify one discrete task the skill should perform. Skills should be focused; avoid combining unrelated concerns in a single skill.

2. Name the skill — Choose a kebab-case name that describes the task (e.g., shields-badges, publishing-npm, openapi-linting). Create the directory skills/<skill-name>/.

3. Write the frontmatter — Open SKILL.md with the YAML frontmatter block containing name and description.

4. Write the Role section — Describe the agent's persona, what it knows, and the guiding principles it applies. Use bullet points for key capabilities.

5. Write the Workflow section — Break the task into numbered steps. Each step should:

   • Begin with a bold action label (e.g., Discover context)
   • Describe what to examine, decide, or produce
   • Reference any supporting assets in references/ where relevant

6. Add reference assets — If the workflow requires lookup tables, templates, or examples that are too large to inline, place them in references/ as markdown files and reference them from the workflow.

7. Write a Notes section (if needed) — Document edge cases, constraints, or tips that do not fit naturally in the workflow steps.

8. Validate the skill — Check:

   • Frontmatter is valid YAML and matches the directory name
   • All referenced files in references/ exist
   • The workflow is unambiguous and complete enough for an agent to follow without additional context
   • The skill does not duplicate an existing skill in the skills/ directory

9. Update the README and AGENTS.md — Add the new skill to the ## Available Skills table in both README.md and AGENTS.md:

   | [<skill-name>](skills/<skill-name>/SKILL.md) | <description> |
   

Notes

• Keep skills small and focused. A skill that does one thing well is more reusable than a broad skill that does many things.
• Avoid hardcoding project-specific values (paths, usernames, repository names). Skills should be portable across projects.
• Reference assets should be in markdown format so agents can read them directly without external tools.
• Skills are discovered by agents from skills/*/SKILL.md; ensure the directory name matches the name frontmatter field.