Share multi-agent best practices

[bloodcarter]

Hello, would you be open to share your multi-agent setup and best practices?

I am heavily experimenting with hcom but bumping into issues related to the fact that agents talking to each other instead of being a blind judges (for example). One way could be cresting custom hcom run commands, but maybe there is a better way to avoid biased implementation and review?

Thanks.

[aannoo]

I dunno I guess it's a general AI problem where they converge and get biased. You could try one agent as orchestrator who routes tasks and only gives specific details to implementer/tester/reviewer. Or what you said a script that does it. You can also set it up so it's spawning a new agent and killing more, instead of letting an agent run long and it getting more biased.

[hopscotchor]

Hi, I’m new to hcom and found the multi-agent workflow very interesting.

Would you be open to a small docs PR for this issue? I’m thinking of adding a minimal “multi-agent chat” example using --tag, --thread, hcom send, and hcom events, aimed at new users who want the simplest safe starting point before trying more structured workflows like debate.

The PR would be docs-only and small.