CUA-Recovery: Benchmarking Error Recovery in Long-Horizon Computer-Use Agents
As computer-use agents (CUAs) move to long-horizon workflows spanning multiple applications, undetected errors can silently corrupt persistent state and cause cascading failures in subsequent actions.
ProofPaper ↗
Key points
- Reliable error detection and recovery have become critical bottlenecks for building robust CUAs.
- However, existing benchmarks focus primarily on end-to-end task performance, providing limited insight into whether agents can detect and recover from errors during execution.
- We introduce CUA-Recovery, a benchmark comprising test and training suites across multiple desktop applications.
- We construct it using ReRail, a framework for synthesizing long-horizon tasks and trajectories in a persistent desktop environment.
Sources (1)
- [1]CUA-Recovery: Benchmarking Error Recovery in Long-Horizon Computer-Use AgentsarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 20, 05:06 AM
As computer-use agents (CUAs) move to long-horizon workflows spanning multiple applications, undetected errors can silently corrupt persistent state and cause cascading failures in subsequent actions.
Reliable error detection and recovery have become critical bottlenecks for building robust CUAs.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 20, 2026Do Diffusion LLMs Scale? Testing Mercury 3 + Decide
- Oct 20, 2026Evolving Agent Interfaces
- Oct 20, 2026Cache Engineering for Agents
- Oct 19, 2026Game of Phones: Harnessing Game Theory and LLMs in 'Bot Wars' to Counteract Phone Scams
- Oct 19, 2026Agent World Model in the Wild: Computer-Use Agent Environment Synthesis
- Oct 19, 2026TorchNeuron: running HF models on AWS Trainium is easier than ever