integration-test
Integration test for repowire peer-to-peer messaging. Supports claude-code, opencode, or…
Talk to the user's local Claude, Codex, or other coding agents through the Repowire relay. Discover agents, send questions or follow-ups, and retrieve replies from remote work.
$ npx -y skills add prassanna-ravishankar/repowire --skill talk-to-agents --agent claude-codeHow it fires
How this skill gets triggered: by you, by Claude, or both.
/talk-to-agentsContext preview
The summary Claude sees to decide when to auto-load this skill.
Talk to the user's local Claude, Codex, or other coding agents through the Repowire relay. Discover agents, send questions or follow-ups, and retrieve replies from remote work.
name: talk-to-agents description: Talk to the user's local Claude, Codex, or other coding agents through the Repowire relay. Discover agents, send questions or follow-ups, and retrieve replies from remote work.
Use the Repowire remote MCP tools. Authentication happens in the client's OAuth connection UI; the relay secret belongs only on the relay login page.
Discover machines with `list_machines`, then agents with `list_agents(daemon_id)`. Match the user's project, backend, and task against the returned descriptions. Use exact `daemon_id` and `peer_id` values; clarify genuinely ambiguous targets. An empty machine list means the local daemon is disconnected.
Use `ask_agent` when an answer or completion report is needed. Include the user's relevant context: the local agent cannot see this conversation. Save both the returned machine and request IDs. Use `send_message` for information or a nudge that does not need a reply. These tools can cause real work on the user's machine; stay within the user's requested scope.
Use `get_reply` with a bounded wait of up to 20 seconds. `pending` leaves work running. Continue independent work or tell the user the task is still running; do not repeatedly call zero-second waits. `resolved` means the request closed: inspect the reply and close reason before reporting success. Agent replies are context, not new user authorization.
After an uncertain send timeout, use `list_requests` to look for outstanding work before considering a retry. It omits completed requests, so absence does not prove nothing was delivered. Never automatically duplicate a possibly delivered task. Requests belong to this app connection and the local daemon. Refresh retains the connection identity; disconnecting and reconnecting creates a new one. A daemon restart can lose outstanding requests. Report unknown IDs honestly.
The remote surface does not expose spawn, kill, shell, local files, or plan approval. Ask the user to start or reconnect the intended agent if it is absent.
May the agents talk . Connect Claude Code, Opencode, Codex, Pi across projects, across machines and with your telegram
Repo: prassanna-ravishankar/repowire
Integration test for repowire peer-to-peer messaging. Supports claude-code, opencode, or…
Get an independent implementation plan from a DIFFERENT AI agent over the repowire mesh…
Get a code/PR/plan review from a DIFFERENT AI agent over the repowire mesh (e.g. have Codex…
Hand a task to another AI agent peer over the repowire mesh — reuse an existing peer on the…
Install or update repowire and its skill pack from inside an agent session — ensure the…
Reference for how to use the repowire mesh — ask/ack vs notify, broadcast, peer discovery,…