[Bug] Agent loop stops when reasoning token budget exhausted

[CodeReclaimers]

Affected area

Agent core

What happened?

When using models that tend to reason for a long time (like deepseek-v4-flash), it looks like they hit the max token budget and just stop, so that I have to prompt it to continue whatever it was thinking about. I can bump maxTokens to work around this, but it seems like just stopping when hitting the reasoning token budget is something that ought to be fixed.

Steps to reproduce

1. Start prime-agent
2. Switch to a reasoning model (I've seen this with deepseek-v4-flash and openrouter auto beta)
3. Give it a prompt that causes long reasoning sessions.

Expected behavior

I expected the reasoning to be cut off and the model forced to respond.

Prime Agent version

0.7.2

Environment

Linux

Additional context

No response

[phrit]

same thing happens to me