Agent skills
Markdown instruction packages served at a well-known URL that teach visiting AI agents how to use a site — documentation agents load exactly when they need it.
01What it means
Agent skills are a convention for packaging instructions for AI agents: a set of markdown files — each with a name and a description — served at /.well-known/agent-skills/ and indexed as JSON with content digests. An agent that discovers the index can load exactly the skill it needs ("use the site's APIs", "contact the team") instead of reverse-engineering conventions from pages or improvising.
Skills are the newest layer of the agent-facing web, and they answer a different question than the rest of it. llms.txt tells a model what exists; an agent card tells another agent what an endpoint offers; a skill tells a visiting agent how to operate — the instructions you would otherwise paste into a chat by hand, published once at a URL agent frameworks know to look for.
Because each skill's digest is published in the index, an agent can detect when instructions have changed — the documentation cannot silently drift the way a support page can. This site publishes two skills: using our site APIs (content API, MCP server, API catalog) and contacting us.
02Why it matters for buyers
- —Self-service for agents: the how-to you would explain to every AI assistant by hand, written once and discoverable by all of them.
- —Drift-proof by construction: digests in the index let agents detect stale instructions before following them — documentation with a checksum.
- —It is on the readiness checklist: our agent-readiness tool looks for a skills index on any public site, alongside llms.txt and agent cards.
03Common questions
How do agent skills differ from llms.txt?
llms.txt indexes what your content is; skills package how to operate it. Descriptions versus instructions — most sites will want both, and they compose.
Where do agent skills live?
At /.well-known/agent-skills/ — markdown files plus an index.json listing each skill's name, description, and content digest, so agents can verify what they load.
Are skills safe to publish?
They are markdown documentation — read-only by nature. A skill grants no access; it describes access you already offer. Keep the same rule as any public doc: nothing in a skill that isn't already public.