inter-agent-comms-reviewer

Agent messaging security requires end-to-end message signing, nonce or timestamp replay bounds, and per-edge authentication to prevent spoofing. Reviewing live communication paths validates that senders are properly authenticated and side-effecting commands enforce idempotency.

Topology enforcement in multi-agent systems confirms that only approved communication paths exist between agents. Reviewing the orchestrator mesh validates per-edge access control to ensure unauthorized agents cannot send instructions or data to receivers outside their allowed edges.

Validating MCP server identity requires pinning server certificates and enforcing receiver-side handling that treats authenticated messages as untrusted input. Reviewing agent-to-MCP-server connections confirms pinned identity and verifies message integrity against tampering.

Planner-worker queues require end-to-end signing to ensure message integrity and prevent tampering. Reviewing the queue validates that side-effecting commands include expiry controls and idempotency checks to defend against replay attacks on the shared bus.

Authenticated agent messages must be treated as untrusted data because sender authentication only validates identity, not content safety. Reviewing receiver-side handling ensures the system does not blindly execute message content as trusted instructions, preventing injection attacks.

Shared bus authentication alone does not guarantee message integrity or replay protection across the full communication path. Reviewing the bus topology reveals gaps in end-to-end signing, expiry enforcement, and per-edge access control that leave agents vulnerable to tampering.