Software Engineering Glossary

Agent Skills

Also known as: Skills SKILL.md Claude Code skills Cursor Skills

Definition

Agent Skills are portable folders that teach an AI coding agent how to do one job. Each skill is a directory with a SKILL.md file: YAML frontmatter for the name and description, plus Markdown instructions the agent follows when the task matches. Claude Code, Cursor, and other tools that implement the Agent Skills standard can load the same files. The agent keeps only the name and description in context until it actually needs the body, which is progressive disclosure applied to prompts.

Popular reads

View All
Payment System Design: Ledger, Idempotency, and Settlement
Payment System Design: Ledger, Idempotency, and Settlement Jul 18, 2026
X Algorithm Explained: How the Open Source Recommendation System Works
X Algorithm Explained: How the Open Source Recommendation System Works Jan 22, 2026
Cursor Skills: How to Create and Use Agent Skills
Cursor Skills: How to Create and Use Agent Skills Jun 23, 2026
The Complete HTMX Guide: From Zero to Production
The Complete HTMX Guide: From Zero to Production Dec 22, 2025
Complete Guide to Graph Data Structure: BFS, DFS, Adjacency List vs Matrix
Complete Guide to Graph Data Structure: BFS, DFS, Adjacency List vs Matrix Jan 20, 2026
Transactional Outbox Pattern: Never Lose an Event Again
Transactional Outbox Pattern: Never Lose an Event Again Apr 07, 2026

Key Takeaways

  • A skill is a folder with a SKILL.md, not a hidden product setting. Commit it and the whole team gets the same playbook.
  • The description is the routing key. A vague description means the skill never fires.
  • Put standing rules in CLAUDE.md (or Cursor Rules). Put multi-step workflows in a skill so they load only when needed.
  • Claude Code adds extra frontmatter on top of the open spec: slash commands, allowed-tools, context: fork, and live shell injection before the model sees the skill.

How It Works

  1. At session start the agent reads each skill’s name and description, about a hundred tokens each.
  2. When your request matches a description, or you type /skill-name, it loads the full SKILL.md body.
  3. Linked files under references/ and executables under scripts/ load only if the instructions point at them.
  4. In Claude Code, a !command`` line runs before the model sees the skill, so the prompt can include a live git diff or PR body.

Where It Is Used

  • Teams commit .claude/skills/ so Claude Code reviews pull requests, writes commits, and deploys the same way for everyone.
  • The same SKILL.md format works in Cursor under .cursor/skills/, which is why one repo can serve both agents.
  • Anthropic ships bundled Claude Code skills such as /code-review and /debug, and lets you override them with a project skill of the same name.

Related glossary terms