cross-agent-review

Cross-agent review

Ask a peer running a different agent backend to review your work, so the review is genuinely independent (not the same model checking itself).

Resolve the reviewer backend

Pick the reviewer in this order:

1. Explicit argument — if the user named a backend (e.g. "review with codex"), use it.
2. Configured default — else read it:

   repowire config get skills.default_reviewer_backend
   

   (Prints the value, or nothing if unset. --json for the raw value.)
3. Safe fallback — if still unset, pick any online peer whose backend differs from yours, or ask the user which backend to use. Never default to your own backend — that defeats the purpose. Never hardcode a specific backend.

You can see who is available and on which backend:

• MCP: list_peers()
• CLI: repowire peer list

Run the review

1. Find (or spawn) a peer on the chosen backend.
2. Send it a review request with the concrete diff/PR/plan to review. Be specific about what to check.
   • MCP: ask(peer_name, "Review this diff for correctness + bugs: <context>")
   • The tracked ask/ack lifecycle is MCP-only — there's no CLI equivalent for a non-blocking ask (repowire peer ask is a synchronous test utility, not this).
3. ask is non-blocking and returns a correlation_id. The reviewer closes the thread with ack(corr_id, <their review>) — that reply comes back to you.
4. Apply the review; iterate with ask(reply_to=corr_id, ...) for follow-ups.

If no different-backend peer is available and the user wants one, spawn it (see the delegate skill) — but only with the user's go-ahead.