[Bug]: Agent Loop can exceed more than 800 steps

[danishcontractor]

Bug Description

I have trajectories in my environment that never exit (balanced mode, GPTOSS-120B) and exceed 800 steps. In agent_loop.py

 return self.graph.astream(
            state if state else Command(resume=resume.model_dump()) if not both_none else None,
            config={
                "recursion_limit": 135,
                "callbacks": callbacks,
                "thread_id": self.thread_id,
            },
            stream_mode="updates",
        )

the value is set to a 135 and its not clear why this happens.

Steps to Reproduce

airline Test Suite_0.json

Sample trajectory

Expected Behavior

Agent should exit without running for so many steps.

Actual Behavior

Agent continues to run -- see trajectory sample.

Environment

Ubuntu, GPT-OSS-120B, Balanced Mode

Error Logs / Screenshots

Attached previously

Configuration

No response

Additional Context

No response

Checklist

• I have searched existing issues to ensure this is not a duplicate
• I have provided all the required information above
• I am using the latest version of CUGA