houmao-agent-loop-pairwise-v3

To initialize a workspace-aware multi-agent tree loop, you must author a deterministic plan artifact with a canonical output directory, verify mailbox gateway posture, and write run-owned memo blocks before issuing a controlled start command.

You can control AI agent orchestration loops using manual lifecycle actions such as peek, ping, pause, resume, recover_and_continue, stop, and hard-kill, all governed by explicit guardrails and fail-closed behavior checks.

No, you cannot use this workflow without email or mailbox capabilities because the system enforces fail-closed checks for email mailbox capability and required participants before initializing the routing packet structure.

Routing packet validation for multi-agent messaging checks packet coverage manually or automatically, routing messaging and gateway responsibilities through owning skills while keeping runtime recovery state outside the authored workspace contract.

No, this skill does not mutate previous version behaviors because it extends v2 behaviors safely without mutating the stable v2 skill or other related stable skills in your environment.