AI agent skills

An agent skill is a reusable set of instructions you hand to an AI assistant so it does a job the same way every time — a code review checklist, a writing style, a research routine. Instead of retyping the same prompt, you save it once and the assistant loads it when the work calls for it.

How each platform packages a skill

The idea is the same everywhere; the file format and the place you put it are not.

Claude

A skill folder holding a SKILL.md file with instructions, plus any scripts or reference files it needs.

ChatGPT

Custom instructions and uploaded reference files, saved on a custom GPT or a project.

Cursor

Rule files kept in the repository so every developer on the project gets the same guidance.

GitHub Copilot

Repository instruction files that Copilot reads before it suggests code.

Grok

System-level instructions attached to a conversation or workspace.

MCP servers

A server that exposes tools an assistant can call, rather than text instructions.

Perplexity

Space-level instructions and sources that shape every answer inside that space.

What makes a skill worth saving

  • It describes a repeatable job, not a one-off question.
  • It states the steps and the output shape, so results stay consistent.
  • It names the edge cases you keep having to correct by hand.
  • It stays short enough to read; long skills get ignored by people and models alike.

Browse ready-made skills

I built Skill Finder Plus for exactly this: a browsable library of agent skills across the platforms above, so you can start from something that already works.

  • TypeScript
  • React
  • AI SDK
  • Supabase
  • hCaptcha