gateway-skill

---

name: gateway_skill description: Complete toolset for interacting with the monoU Gateway. Discover topology (online agents, L3 nodes and their capabilities), query per-agent skills and cron jobs, delegate tasks to other agents, invoke L3 node capabilities (browser-node / sandbox-node via node.invoke, analogous to MCP servers), manage sessions, and push messages to connectors (e.g. Feishu). Use when the user asks about: who is online, what agents can do, scheduled tasks, delegating work to another agent, using browser or sandbox capabilities, sending messages to Feishu/channels, or viewing/sending to sessions. Requires Gateway connection.

Gateway Skill

Interact with the monoU Gateway. All tools in this skill call through the Gateway WebSocket.

Architecture

L1 Connectors (Feishu, Control UI)
        ↓
L2 Gateway  ← all tools here route through Gateway
        ↓              ↓
L3a Agent          L3b Node
(Runner)      (browser-node / sandbox-node)
               declares capabilities on connect;
               invoked via node.invoke (like MCP servers)

L3 Nodes connect with role=node and declare capabilities (e.g. ["browser"], ["sandbox"]). Use gateway_node_invoke to call their commands — this is the MCP-equivalent mechanism in monoU, not skill files.

Tools

Topology

Tool	Use
gateway_agents_list	List all connected agents (agentId, deviceId, online, lastHeartbeatAt).
gateway_nodes_list	List L3 nodes (nodeId, capabilities) and L1 connectors (e.g. Feishu).

Skills & Cron

Tool	Use
gateway_skills_status	Get an agent's skill list (what it can do). Optional agentId, defaults to current agent.
gateway_cron_list	List an agent's scheduled jobs. Optional agentId, defaults to current agent.

Agent Delegation

Tool	Use
gateway_agent_send_to_session	Send a message to another agent's session; that agent executes and replies in that session. Required: targetAgentId, message. Optional: sessionKey (defaults to agent:<targetAgentId>:main).

Node Invocation (MCP-style)

Tool	Use
gateway_node_invoke	Invoke a command on an L3 node. Use gateway_nodes_list first to find nodeId and capabilities. browser-node: browser_fetch, browser_click, browser_fill, browser_links, browser_screenshot, browser_pages, browser_switch, browser_new_tab. sandbox-node: system.run, system.which.

Sessions

Tool	Use
sessions_list	List all sessions (sessionKey, sessionId, updatedAt, displayName).
sessions_preview	Same but fewer fields — quick overview.
sessions_send	Send a message to a session by sessionKey; triggers agent reply in that session.

Message Push

Tool	Use
send_message	Push a message to a connector channel (e.g. Feishu group/DM). Get connectorId from gateway_nodes_list connectors. Required: connectorId, chatId, text.

Common Patterns

• "Who is online?" → gateway_agents_list
• "Is there a browser node?" → gateway_nodes_list (check capabilities field)
• "What can agent X do?" → gateway_skills_status(agentId)
• "Ask agent X to do Y" → gateway_agents_list to confirm online → gateway_agent_send_to_session
• "Open a URL with the browser" → gateway_nodes_list to find node with capabilities: ["browser"] → gateway_node_invoke(nodeId, "browser_fetch", { url })
• "Run a command in sandbox" → gateway_nodes_list for capabilities: ["sandbox"] → gateway_node_invoke(nodeId, "system.run", { command })
• "Send a Feishu message" → gateway_nodes_list for connectorId → send_message(connectorId, chatId, text)
• "Why isn't Feishu push working?" → gateway_nodes_list to check connector status
• "Did agent X's cron run?" → gateway_cron_list(agentId)

Guidelines

• All tools require Gateway. If not connected (no GATEWAY_URL set when agent started), tools return an error.
• For browser interactions beyond gateway_node_invoke (auto node selection, screenshot parsing), use browser_skill — it wraps node.invoke internally.
• Always run gateway_agents_list before delegating to confirm the target agent is online.

More Scenarios

See references/scenarios.md for discovery, capability queries, cron troubleshooting, orchestration, and combined usage examples.