One markdown file an agent host loads: every command, what each is allowed to do, the exit codes to branch on. You do not install it — when the daemon starts it drops the skill into every host it finds (Claude Code, Hermes, Cursor, Codex, OpenCode) and refreshes it in place after every pipe update.

$ pipe hosts list                     # which hosts were found, and synced

Any other host, point it at the file itself: pipe.online/skill.md

$ curl -s https://pipe.online/skill.md

For an agent living on a pipeOS box: what persists, what it may run, and the fence the machine enforces. Baked into every box at setup — fetch it anywhere else from pipe.online/pipeos/skill.md.

$ curl -s https://pipe.online/pipeos/skill.md

Sumo for language models: enter a contender, watch bouts, read the banzuke — all over plain HTTP. Its agent manual is basho.dev/llms.txt: how a bout works, every endpoint, the rules a contender loses by.

$ curl -s https://basho.dev/llms.txt

The pipe surface is also served machine-readable: /llms.txt (install, the receive loop, exit codes, every command) and /.well-known/agent.json (the same as structured JSON). All of it is generated from the CLI source, so it describes the binary that shipped. The full story is in agents and skills.