Skip to content

Writing Your First Agent Skill

A 6-chapter course for developers on writing a skill from scratch. SKILL.md anatomy, the references/ vs scripts/ split, and the 10 most common authoring mistakes.

By skills-il·Version v1.0.0·Last updated: 23-05-2026·GitHub
6 chapters~24 minDevelopersBeginner
0Write a Review
10

What you'll learn

  • Understand what a skill really is to the LLM at routing time, and when to pick a skill vs MCP / system prompt / CLI
  • Write a SKILL.md with a description the LLM routes correctly (Use when / Do NOT use for patterns)
  • Distinguish metadata.json from YAML frontmatter and understand why skills-il uses a separate file
  • Decide between references/ and scripts/ for any supporting content
  • Ship a skill end-to-end through the pipeline: scaffold, push, sync, trust score, publication
  • Run the 10-minute pre-publish checklist and pass the 10 most common rejection reasons

Who this is for

Developers who know git, Markdown, and JSON, and want to write a Claude agent skill for the skills-il catalog. No prior Claude Skills experience required.

Course curriculum

Click "Start course" to unlock chapters
  1. 1.
    What is a skill, really?
    A skill is the simplest way to teach an AI assistant to handle a specific task: a single Markdown file with YAML frontmatter at the top, optionally accompanied by reference files and executable scripts. The agent reads the file on demand when the description matches the user's request. That is the entire model. Everything else (catalog, sync pipeline, trust score, install counts) is infrastructure around that one file.
    ~4 min
  2. 2.
    Anatomy of a great SKILL.md
    A great SKILL.md does three things at once: it tells the LLM when to load itself (the description), it teaches the LLM how to perform the task (the body), and it gives the LLM enough examples to handle edge cases without escalating to the user. If any of these three are missing, the skill either does not get loaded (description failure), does the wrong thing (body failure), or asks the user too many clarifying questions (examples failure).
    ~4 min
  3. 3.
    Extending the spec (metadata.json + bilingual content)
    The skill format (the spec developed by Anthropic, accepted by Claude Code, Cursor, Windsurf, and Claude Desktop) is intentionally minimal: `SKILL.md` with three frontmatter fields, plus the optional `references/` and `scripts/` subfolders. That is enough for a personal skill or a skill you share by zipping a folder and sending it to a colleague. But if you want to put your skill in a catalog, support multiple languages, or expose extra metadata to consumers, you need to extend the spec.
    ~4 min
  4. 4.
    references/ vs scripts/ (when to use which)
    A skill folder can include two optional subfolders: `references/` and `scripts/`. They serve different purposes and choosing the wrong one is a frequent rejection reason. The rule is simple: `references/` for static content the LLM reads, `scripts/` for executable code the LLM runs. The hard part is recognizing which side of the line your content falls on.
    ~4 min
  5. 5.
    Sharing your skill
    You have written SKILL.md, optionally added references/ and scripts/, and the skill works. Now the practical question: how do you actually use it, share it with others, or publish it to a catalog? This chapter covers four scenarios in order of effort.
    ~3 min
  6. 6.
    The 10-minute pre-share checklist
    This chapter is short on purpose. It is the checklist you run BEFORE sharing your skill with anyone (a colleague, a public catalog, or your future self after a long break). Ten minutes spent here saves multiple back-and-forth rounds with reviewers and prevents the most common authoring mistakes.
    ~5 min

Skills to install to act on what you've learned

What's new

  • 1.0.0
    Initial release. Six chapters covering how to write a first agent skill: the mental model (skill vs MCP / system prompt / CLI), SKILL.md anatomy, the skills-il extension (metadata.json + SKILL_HE.md), references/ vs scripts/, the publication flow, and a 10-minute pre-publish checklist. Catalog exemplars referenced: israeli-id-validator, israeli-phone-formatter, hebrew-seo-geo-toolkit.(24-05-2026)

Reviews (...)

0Write a Review

Use at your own risk. Terms of Use · Security