Agent skill

A skill is a folder an agent harness reads: a markdown file describing when and how to perform a task, often with helper scripts, and sometimes with tool definitions or credentials it expects. Registries distribute them with one-line installers, and a harness loads the ones relevant to a prompt. In crypto, skills wrap wallets, trading venues, chain queries and payment rails.

Because a skill runs with the agent's permissions, what it contacts and asks for matters more than what it claims. Sato Hub publishes a static disclosure per skill — hosts contacted, whether it generates keys, requests credentials, pipes remote content to a shell, or grants tools — read from the skill's files, plus the registry's own scan status attributed to the registry.

A disclosure describes; it never says safe. A skill with a clean disclosure can still be wrong, and a skill that requests a private key may be doing exactly what a trading skill needs. The disclosure exists so the reader can decide.