nbn07/agent-tracedown is a Python tool that finds the step that caused a failure in agent execution traces, presents the evidence, and replays the run after applying the fix. It also covers cases where the trace is marked as successful but the answer was wrong. The latest release is v0.4.3, and the given material does not separately list what changed in this version.
The tool diagnoses based on execution traces that were already collected. It says you can import Langfuse, Arize Phoenix, and OpenTelemetry traces, or instrument runs with SDKs for LangGraph, CrewAI, OpenAI agents, Claude Agent SDK, Google ADK, and Pydantic AI. Diagnosis is rule-based, and language model judging is optional. In other words, it focuses not only on observing a run from the start but also on finding clues to problems in traces that already exist. The README also describes self-hosting and team deployment, public evaluation materials, and how to use the web interface and the command line.
It is a Python project licensed under Apache-2.0. That said, because the diagnosis is based on the information left in the traces, if there are no execution traces or the necessary steps were not recorded, the evidence for identifying the cause is limited as well. The applicable scope of this tool therefore lies in analyzing and replaying traces in environments where agent runs have been traced.