CUA-SWE: When Computer-Use Agents Meet Visual Software Engineering
CUA-SWE tests whether agents can repair software by using the running app’s visual interface, not just the codebase.
The paper introduces a benchmark, environment, and evaluation pipeline for tasks that mix code edits, command execution, GUI interaction, screenshots, and verification. Its tasks span four software engineering domains and include cases where required information is only visible inside the app. Each task has deterministic tests to check the final software behavior. HF Daily Papers' note
The paper introduces a benchmark, environment, and evaluation pipeline for tasks that mix code edits, command execution, GUI interaction, screenshots, and verification. Its tasks span four software engineering domains and include cases where required information is only visible inside the app. Each task has deterministic tests to check the final software behavior. HF Daily Papers' note
score 5