execution-feedback

CP-Agent presents a calibrated risk-controlled approach for feedback-driven competitive programming using large language models, achieving significant improvements on benchmarks without parameter updates.