Why infrastructure agents should only connect outbound

A management agent that listens on a public port becomes a target. An agent that only makes outbound HTTPS requests needs no open port, works behind NAT, and can be revoked centrally.

The trade-off is latency: commands are fetched by polling or over an outbound WebSocket instead of being pushed. For deployment workflows with explicit approval, that trade is acceptable.

Signed requests, short expiries and replay protection make the outbound channel safe to rely on.