# Agent Skills

- **Website:** <https://agentskills.io>
- **Primary alias:** `agentskills.io`

## Description

Agent Skills is an open standard and documentation project that helps people give AI agents specialized capabilities and expertise. Developed originally by Anthropic and released as an open format, it provides a consistent way to package knowledge, instructions, and resources for use across compatible AI tools and agentic clients.

A skill is a folder centered on a required SKILL.md file, which contains at least a name, description, and task instructions. Creators can also include scripts, reference materials, templates, and other assets. Skills are designed to capture domain expertise and repeatable workflows, including company-, team-, or user-specific processes, in portable, version-controlled packages.

Agents use skills through progressive disclosure: they first identify relevant skills from their names and descriptions, then load detailed instructions when needed, and finally follow those instructions, using bundled resources as appropriate. This approach helps agents handle many skills without loading every instruction at once. The site offers a format specification, quickstart, creation and optimization guidance, implementation advice, and a showcase of compatible clients.

## Industries

- 🖥 Computers Electronics and Technology
- 🖥 Artificial Intelligence and Machine Learning _(under Computers Electronics and Technology)_

## Links

- [github](https://github.com/agentskills)

## Logos & icons

- logo _(primary)_ — PNG — [download](https://cdn.brandfetch.io/idSIp6pjfg/w/180/h/180/theme/dark/logo.png?c=1bxid64Mup7aczewSAYMX&t=1793162042081)

## Colors

| Name | Hex | Theme |
| --- | --- | --- |
| Tundora | #404040 | dark |
| White | #ffffff | light |

## Fonts

- var(--font-inter) — asset _(custom)_
- var(--c15t-font-family) — asset _(custom)_

---

_Source: <https://brandfetch.com/agentskills.io>_
_See [/llms.txt](/llms.txt) for a full list of agent-readable pages, and [/auth.md](/auth.md) for how agents authenticate._