A small, verified, security-aware directory of Agent Skills (SKILL.md format).
概览
A list of Agent Skills (the SKILL.md format used by Claude Code, Cursor, Codex, and more), curated for trust, not a 1,400-skill dump. A skill is a folder with a SKILL.md that teaches an agent a workflow. The ecosystem exploded, and so did the risk: a skill is code the agent trusts, and Snyk found prompt injection in while Trail of Bits showed the public scanners are bypassable in under an hour. Most lists optimize for count. This one optimizes for signal: the official sources, a tight set of genuinely useful skills, and a first-class security section the volume lists skip. - until you have read its SKILL.md and any scripts. Treat installing one like adding a dependency, or running a stranger's shell script. - The model loads a skill based on its description; a poisoned description is how injection gets in. - Use a skill scanner (below), and prefer skills from teams you can name. - A good skill keeps SKILL.md short and links out to details the agent pulls only when needed.
README
以下为原文 README 节选，完整内容请查看原文。
Awesome Agent Skills 
A small, verified, security-aware list of Agent Skills (the SKILL.md format used by Claude Code, Cursor, Codex, and more), curated for trust, not a 1,400-skill dump.
A skill is a folder with a SKILL.md that teaches an agent a workflow. The ecosystem exploded, and so did the risk: a skill is code the agent trusts, and Snyk found prompt injection in 36% of tested skills while Trail of Bits showed the public scanners are bypassable in under an hour. Most lists optimize for count. This one optimizes for signal: the official sources, a tight set of genuinely useful skills, and a first-class security section the volume lists skip.
Browse and filter: agent-skills.agentpostmortem.com
Start here: how to think about skills
- A skill is untrusted code until you have read its SKILL.md and any scripts. Treat installing one like adding a dependency, or running a stranger’s shell script.
- The trigger description is the security boundary. The model loads a skill based on its description; a poisoned description is how injection gets in.
- Scan before you install. Use a skill scanner (below), and prefer skills from teams you can name.
- Author for progressive disclosure. A good skill keeps SKILL.md short and links out to details the agent pulls only when needed.
The one rule: volume is not trust. A verified skill from a known team beats a hundred you have not read.
46 entries, auto-refreshed weekly. Star counts updated 2026-09-28. Browse the filterable version at agent-skills.agentpostmortem.com.
Official and spec
- anthropics/skills
* 178.8k: Anthropic’s official public repo: production document skills (docx, pdf, pptx, xlsx), a skill template, skill-creator, and the spec. - agentskills/agentskills
* 25.8k: Specification and documentation repository for the open Agent Skills standard. - Agent Skills Specification: The open, vendor-neutral standard for the SKILL.md format, adopted beyond Claude.
- Agent Skills overview (Claude docs): Official docs on what skills are, how discovery and loading works, and how to build them.
- Introducing Agent Skills (Anthropic): The launch announcement of Agent Skills as an open standard, plus the design deep-dive.
- Claude Skills gallery: Anthropic’s official gallery of prebuilt skills (PowerPoint, Excel, Word, PDF, and more).
Skill collections
推荐工具
换一个关键词，或者移除筛选条件。
安装
npx skillfish add royalpinto007/awesome-agent-skills