Agent Skills (agentskills.io)

Agent Skills started at Anthropic and were released as an open standard. agentskills.io is where that standard lives: the specification, a quickstart, and a showcase of clients that implement it.

The mechanism it documents best: progressive disclosure

Skills load in three stages, and this is the detail that explains why you can have many of them without paying for all of them:

1. Discovery — at startup, the agent loads only each skill’s name and description. Enough to know when it might apply.
2. Activation — when a task matches a description, the full SKILL.md is read into context.
3. Execution — the agent follows the instructions, optionally running bundled scripts or loading referenced files.

Full instructions load only when the task calls for them. That is the whole reason skills scale where a longer system prompt does not.

The format

my-skill/
├── SKILL.md          # Required: metadata + instructions
├── scripts/          # Optional: executable code
├── references/       # Optional: documentation
├── assets/           # Optional: templates, resources
└── ...

SKILL.md requires name and description at minimum.

Adoption beyond Anthropic

The client showcase lists a substantial number of agents implementing the format — among them OpenCode, Goose, OpenHands, Cursor, Roo Code, Gemini CLI, GitHub Copilot and VS Code, Amp, Letta, Kiro, Factory and Codex.

That matters for a practical reason rather than a political one: a skill you write is not locked to Claude. Of everything in the agent stack, skills are the most portable artefact.