maxConcurrent is ignored for foreground subagents

[WinPooh32]

What I can see

All subagents run in parallel ignoring maxConcurrent setting.

Expected behaviour

maxConcurrent == 1 : only one subagent executed at the same time. Agents execution is queued.

Why does it matter?

Running parallel agents with local models are spending much more time on solving tasks due to cache misses on switching chat sessions. It hurts very much. Queued agents solve this problem.
Asking local model to not spawn many subagents is not reliable.

[tintinweb]

By design, fg agents are exempt from macConcurrent. I dont think it makes much sense to combine the limits because under normal circumstances you want your main agent to be available for work while bg work is done. The difference between normal main agent foreground work and a blocking foreground subagents is marginal and we dont want to starve the main session just because the bg pool is full (it would also not starve if main agent would just do the work directly)

[WinPooh32]

Main problem is not maxConcurrent setting itself.
It would be good to limit foreground subagents parallel execution.

[tintinweb]

Would this be addressed with a maxConcurrentForeground?

[WinPooh32]

Yes, that would be nice

[WinPooh32]

maxConcurrentForeground=1 works well as expected. Thank you.