Agent skills
An agent skill is a folder with a SKILL.md file that teaches an AI coding agent how to do one job. The agent reads each skill's name and description at startup and loads the full instructions only when a task needs them, so a skill it is not using costs little more than its name and description.
The SKILL.md format
A skill folder holds SKILL.md and, optionally, scripts/, references/ and assets/. SKILL.md starts with YAML frontmatter, then Markdown instructions. The Agent Skills specification requires two fields: name (up to 64 lowercase letters, numbers and hyphens, matching the folder name) and description (up to 1,024 characters, saying what the skill does and when to use it). Optional fields are license, compatibility, metadata and the experimental allowed-tools. In Claude Code every field is optional and only description is recommended. Both sources checked 2026-10-28.
--- name: pdf-processing description: Extract PDF text, fill forms, merge files. Use when handling PDFs. --- # PDF processing Step-by-step instructions for the agent...
Where each agent loads skills
Each path below comes from the agent's own documentation, linked in the last column and checked 2026-10-28. Project folders sit in your repository; personal folders sit in your home directory.
| Agent | Project | Personal | Source |
|---|---|---|---|
| Claude Code | .claude/skills/ | ~/.claude/skills/ | Docs |
| Codex Scans .agents/skills in every folder from the working directory up to the repository root. | .agents/skills/ | ~/.agents/skills/ | Docs |
| Cursor Also reads .claude/skills/, .codex/skills/ and the matching home folders for compatibility. | .agents/skills/.cursor/skills/ | ~/.agents/skills/~/.cursor/skills/ | Docs |
| GitHub Copilot Supported locations vary by Copilot surface. | .github/skills/.claude/skills/.agents/skills/ | ~/.copilot/skills/~/.agents/skills/ | Docs |
| Gemini CLI | .gemini/skills/.agents/skills/ | ~/.gemini/skills/~/.agents/skills/ | Docs |
| OpenCode | .opencode/skills/.claude/skills/.agents/skills/ | ~/.config/opencode/skills/~/.claude/skills/~/.agents/skills/ | Docs |
Install with vskill
The open-source vskill CLI detects which agents you have and puts the skill where each one looks for it. It supports 56 install targets (vskill 1.2.16, counted 28 Oct 2026). The command is below.
Check a skill before you install it
A skill can tell your agent to run commands and can ship its own scripts. Every listing on Verified Skill shows its security scan, trust tier and source. See the trust center and how we verify for what is checked and what a scan does not guarantee.
Browse skills
Start with the Claude Code skills hub, pick a category below, or search the full registry. To decide what belongs in a skill and what belongs in AGENTS.md or CLAUDE.md, read AGENTS.md vs CLAUDE.md vs skills and the SpecWeave skills guide.
npx vskill@latest install <owner>/<repo>/<skill>
Top 20 agent skills by trust score
anthropics/anthropic-cookbook
DevOps · 41,482 starsanthropics/anthropic-cookbook
DevOps · 33,610 starsanthropics/anthropic-cookbook
Security · 34,691 starsanthropics/anthropic-sdk-csharp
AI/ML · 328 starsanthropics/anthropic-sdk-go
AI/ML · 1,179 starsanthropics/anthropic-sdk-java
AI/ML · 342 starsanthropics/anthropic-sdk-python
AI/ML · 3,937 starsanthropics/anthropic-sdk-ruby
AI/ML · 371 starsanthropics/anthropic-sdk-typescript
AI/ML · 2,130 starsanthropics/claude-agent-sdk-demos
AI/ML · 2,585 starsanthropics/claude-agent-sdk-demos
AI/ML · 1,630 starsanthropics/claude-agent-sdk-demos
AI/ML · 1,636 starsanthropics/claude-agent-sdk-demos
AI/ML · 2,764 starsanthropics/claude-agent-sdk-demos
AI/ML · 2,326 starsanthropics/claude-agent-sdk-demos
AI/ML · 1,660 starsanthropics/claude-agent-sdk-demos
AI/ML · 1,636 starsanthropics/claude-code-action
AI/ML · 5,919 starsanthropics/claude-code
AI/ML · 91,563 stars- anthropics/claude-code/claude-codeCertified
anthropics/claude-code
Docs · 72,305 stars anthropics/claude-code
AI/ML · 145,657 stars
Common questions
Are agent skills and Claude Code skills the same thing?
Yes, the format is the same. Claude Code skills follow the Agent Skills open standard, and Claude Code adds features of its own such as invocation control and running a skill in a subagent.
Do I need both name and description in the frontmatter?
The Agent Skills specification requires both. Claude Code only recommends description and falls back to the folder name, so write both if you want the skill to work in every agent.
Where do I put a skill so several agents find it?
Per the docs linked in the table above, Codex, Cursor, GitHub Copilot, Gemini CLI and OpenCode all read .agents/skills/. Claude Code reads .claude/skills/, which Cursor, Copilot and OpenCode also read. vskill writes to the right folder for each agent it finds.
Can a skill run code on my machine?
It can. The specification lets a skill ship a scripts/ folder of executable code that the agent may run. Check the security scan and source on the skill's page before you install it, and read how we verify for what the scan covers.