Skip to content

Latest commit

 

History

History
21 lines (11 loc) · 813 Bytes

File metadata and controls

21 lines (11 loc) · 813 Bytes

An Empirical Study of Proactive Coding Assistants in Real-World Software Development

Why It Matters

ProCodeBench shows that proactive coding assistants trained or evaluated on synthetic traces can misread real development behavior.

Proactivity Signal

The agent must infer coding intent and assistance opportunities from real IDE trajectories rather than explicit user prompts.

Evaluation Setup

The paper uses real VS Code traces and compares them with simulated development trajectories to expose sim-to-real differences.

Key Limitations

The benchmark is specific to software development and may not generalize to other computer-use domains without new trace data.

Use For

Use this for real-world proactive coding evaluation, sim-to-real analysis, and IDE trace-based benchmark design.