FlowEvo: Self-Evolving Agents through the Co-Evolution of Workflows and Executable Skills
FlowEvo turns successful agent workflows into reusable executable skills during inference.
The paper presents a training-free framework that keeps a persistent skill bank instead of discarding procedures after a task run. Retrieved skills can be executed directly or used as context for building new workflows, while skills linked to negative transfer are suppressed. With GPT-4o-mini, FlowEvo reports the best accuracy among eight baselines across ALFWorld, HumanEval, MBPP, GSM8K, and MATH-500. On ALFWorld, it reaches 85.6% accuracy while using roughly one third as many tokens as the strongest baseline.
HF Daily Papers' note
The paper presents a training-free framework that keeps a persistent skill bank instead of discarding procedures after a task run. Retrieved skills can be executed directly or used as context for building new workflows, while skills linked to negative transfer are suppressed. With GPT-4o-mini, FlowEvo reports the best accuracy among eight baselines across ALFWorld, HumanEval, MBPP, GSM8K, and MATH-500. On ALFWorld, it reaches 85.6% accuracy while using roughly one third as many tokens as the strongest baseline.
HF Daily Papers' note
score 5