Run as squad (CLI) or initialize agent from copilot?

[tihomir-kit]

Hi,

Could you please shed some light on the intended use of squad in this sense? I started using it <v0.5.4 and then it was ran from the copilot cli.

What are the differences in your opinion, are there any obvious pros / cons running squad vs copilot --agent squad, does one work better than the other in your view, does it even matter?

Thanks

[tihomir-kit]

I'm asking because I've switched to using squad CLI and I think now more often the model that is supposed to be used is incorrect, and I'm not really sure that the background agents are being spawned. I have a feeling that sometimes the coordinator just kind of "impersonates" the desired squad agent, but does it on the "main thread" if that makes sense, and also using haiku.. Which feels like a bit of a handicap, and I'm not sure how to get it under control.

[bradygaster]

Both work, but here's the difference:

• squad CLI — runs Squad as a standalone interactive shell (the REPL). You get the full Squad experience: agent coordination, slash commands, dispatch syntax (@AgentName do something), autonomous mode, etc.
• copilot --agent squad — runs Squad as an agent within the Copilot CLI. This means Squad operates as a sub-agent under Copilot's coordination.

Recommendation: Use copilot --agent squad directly for the richest experience. The standalone CLI has features (REPL commands, shell history, agent monitoring) that don't fully translate when running as a nested agent. The REPL isn't quite done yet.

As Squad matures, the standalone CLI is where we're investing the most. The copilot --agent path is useful if you want to mix Squad with other Copilot agents, but for pure Squad workflows, go direct.