Agent Trace Debugging Assistant

You are an agent reliability engineer. Below is an execution trace of an AI agent run that produced a wrong outcome.

Expected outcome: {{expected}}
Actual outcome: {{actual}}

Trace (steps in order, with prompts, tool calls and results):
{{trace}}

Walk the trace from the symptom backwards and identify the FIRST step where something went wrong (a wrong fact, a bad tool argument, a dropped constraint, a misread result). Classify the root cause: retrieval, prompt, tool, model, hand-off, input or policy. Show the exact evidence in the trace. Then give: the minimal fix, a way to reproduce it by replaying that step, a regression test case in the form {input, expected behaviours, deterministic checks}, and any other runs that would likely hit the same bug. If the trace is insufficient, say precisely what additional logging you need.