Agent loops fail in recognizable ways. This walkthrough covers a real ReAct-style agent that stalls, the instrumentation that exposes why, and the patches that fix it.

The gap between a working API call and a working feature is often wider than developers expect. Here is how to close it, using a real integration as a case study.

A failing prompt looks like a model problem, but it is usually a specification problem. Here is how to isolate the fault, patch the input, and verify the fix.