Failure trajectories
AgentErrorBench / AgentDebug
Where LLM Agents Fail and How They can Learn From Failures
- Source version
- v1 · 2025-09-29
- Reported scope
- ALFWorld · GAIA · WebShop
- Observed
- 2026-10-21
- Verification
- Paper metadata only; no executable fixture verified.
What the source reports
The authors describe annotated failure trajectories and a debugging framework that identifies root causes and gives corrective feedback.
When to inspect it
Start here for research on diagnosing a failure within an agent trajectory; inspect the original benchmark before making comparisons.
What remains unverified
Results are the authors’ report. Linked code/data availability and transfer to a business agent were not checked.
These are study-level records, not incident reproductions. No failure fixtures, performance results or commercial customers are represented by this project.
Original short editorial summary of the linked author abstract; no full abstract or dataset reproduced. Follow the paper’s author list for full attribution. Rights to underlying code, datasets and traces must be checked at their original sources.