What is an Agent Skill?

An Agent Skill is a folder that contains a SKILL.md file. That file tells your coding agent how to carry out one specific task, and the agent reads it only when the task calls for it.

What a skill contains

The core of a skill is the SKILL.md file. It starts with a YAML header (the frontmatter) followed by instructions in Markdown:

---
name: accessibility-audit
description: Checks a web page for accessibility problems and suggests fixes. Use when asked for an accessibility, WCAG or a11y audit.
---

# Accessibility audit

1. Open the page and list the interactive components.
2. Check contrast, labels and keyboard navigation.
3. Return a table of issues, sorted by severity.

According to the specification, two fields are required:

• name: at most 64 characters, lowercase letters, digits and hyphens, matching the folder name;
• description: at most 1,024 characters. It says what the skill does and when to use it. The agent reads it to decide whether to load the skill.

Other fields are optional, such as license, compatibility or metadata. The body has no mandated format: steps, examples, edge cases.

The folder can also hold supporting files. The specification suggests three subfolders by convention:

Why the agent loads it on demand

A skill is not pasted into every conversation. Agents load it in three steps, which the specification calls progressive disclosure:

1. Discovery: at startup, the agent reads only the name and description of each installed skill.
2. Activation: if your request matches a description, it reads the full SKILL.md.
3. Execution: it follows the instructions and opens supporting files only when a step needs them.

So you can install many skills without weighing down each exchange. The trade-off is that everything rests on the description: if it is vague, the agent will not recognise when to use the skill. The specification also recommends keeping SKILL.md under 500 lines and moving detail into separate files.

Where the format comes from

Anthropic published the format as an open standard in December 2025, with a specification at agentskills.io. The company later announced it was donating the standard to the Agentic AI Foundation to keep it open and vendor-neutral, and said more than 70 agent platforms had integrated it.

In practice, a skill written for one agent can serve another. Claude Code follows the standard and adds its own options, such as invocation control. A skill that sticks to the standard fields stays portable.

Which agents read it

These three agents document the format and their folders:

Find these agents on their pages: Claude Code, Codex, Cursor. Many other agents read the format too: the catalog lists about thirty, each with its install folder.

What PalateKit adds

Skills do not cover everything a coding agent uses. PalateKit gathers in one catalog, in English and French:

• Agent Skills, in the format described above;
• DESIGN.md files, which describe a project's visual system so the agent produces a consistent interface;
• motion design templates;
• prompts for text, image and video;
• rules, agents and hooks.

These other resource types are not skills in the sense of the specification: they are files the agent reads or that you paste, with their own locations. Each page says where to put them.

Installing a resource means copying files. Nothing runs during installation, and the catalog does not offer resources that ship scripts. The three methods (install prompt, command, ZIP) are described in Install your first resource, and the difference from a plain prompt in Resource or plain prompt?.