Skills & rules

Hub MCP tools

An agent running on the bus reaches skills/playbooks through three MCP tools the Hub provides, not a shell command:

Tool	Input	Returns
list_agents	none	Every enabled agent on the bus: its bus address, description, and its playbooks (name + description only).
get_playbook	name	One of the agent’s own playbooks: name, description, body, its files (path + kind), and related playbooks with which agent holds them.
get_playbook_file	playbook, path	One file of one playbook: path and content. If the file is a script, it also gets exec_path — an executable copy, rewritten on every call.

A playbook belonging to another agent is only ever named via list_agents / the “related” part of get_playbook — its content can’t be fetched directly; that work is handed off to the agent that owns it.

Asking a person

When an agent needs a person to decide, it asks right in chat — with choices, optionally one choice flagged as recommended, and optionally a reference link. The question goes to the owner; they answer by replying to that question.