Parallel subagents' responses are getting combined & truncated

[utkesmer]

The model spawned a few subagent for a deep research session, parallel tasks.
When all agents are done, the subagents' responses are all combined and only then given to the model.
Since the combined response as a tool result was too long, most of the results got truncated.

Time and work is gone. Had to re-do it all with a new "write reports as md files" instruction.

I think this should be handled better.