Adversarial Agent Collaboration for Correctness Improvements of C to Safe Rust Translation

In three linesACToR, an adversarial multi-agent loop, improves C-to-Rust translation using a translator agent and discriminator agent that iteratively compete. On 63 real-world C utilities (avg 473 LOC), the system achieves 90% test pass rate with zero human intervention, improving correctness by 36.7% over non-adversarial baselines.

Summary generated by Claude — human-verified