ratking

name	ratking
description	Use this skill when you need to communicate with a peer Pi coding agent — to delegate a task, ask a question, sync data between machines, or coordinate work across agents. Activate when the user asks you to "talk to", "send to", "ask", "ping", or "coordinate with" another agent, or when you need information that a peer agent has access to.

ratking — Peer-to-Peer Agent Communication

You have tools that let you send messages to other Pi coding agents and receive their replies. Agents are equals — there is no orchestrator. Any agent can initiate communication.

Your Tools

Tool	What It Does
list_agents	See who is in the pool (get their ids and names)
send_to_agent	Send a prompt to a peer; returns a messageId immediately
await_response	Block until the peer replies (default 60s timeout)
poll_response	Non-blocking check — returns response or null

Standard Workflow

1. list_agents        → confirm peer is online, get their id
2. send_to_agent      → send your prompt, save the messageId
3. await_response     → wait for peer's reply
4. (repeat as needed)

Example: Ask a peer for data

1. Call list_agents — find "prod" agent with id "abc-123..."
2. Call send_to_agent { toAgentId: "abc-123...", prompt: "Please give me the schema of the users table with all PII fields redacted" }
   → returns messageId "msg-xyz..."
3. Call await_response { messageId: "msg-xyz..." }
   → peer returns: "users(id, email_hash, created_at, plan_tier)"
4. Use that information locally

Rules

• Always call list_agents first — never guess an agentId.
• Never send raw PII in your prompt. If you need sensitive data from a peer, instruct it to redact/anonymise before returning.
• Include enough context in your prompt for the peer to act autonomously. The peer's LLM does not see your conversation history.
• Define a clear end state — tell the peer what "done" looks like to prevent loops.
• Use poll_response when you want to continue doing local work while waiting; use await_response when you must have the answer before proceeding.
• Increase timeoutMs for long-running peer tasks (e.g., 120000 for 2 minutes).

Connection Setup (human)

The user must connect the agent to the pool before these tools work:

/jcoms              — connect + list peers (auto-names this agent)
/jcoms name prod    — set this agent's display name to "prod"
/jcoms disconnect   — leave the pool
/jcoms ping <id>    — test round-trip to a specific agent

When Things Go Wrong

• "not connected to pool" — ask the user to run /jcoms first
• await_response timeout — the peer may be busy or offline; try poll_response later or ask the user to check the peer's terminal
• "agent not found" — call list_agents again; the peer may have reconnected with a new id