Retrieval-Augmented Skill Optimization via Cross-Harness Adaptation
RASO uses existing public agent skills as prior knowledge instead of starting optimization from rollouts alone.
The paper proposes Retrieval-Augmented Skill Optimization, which pulls relevant skills from an external corpus and adapts them to a target task and harness. Its two stages create an initial skill without agent rollouts, then update that skill using retrieved knowledge guided by execution feedback. The authors report consistent gains across four agent benchmarks and two models against baselines that do not use retrieval-augmented initialization and updating. ArXiv · AI/CL/LG's note
The paper proposes Retrieval-Augmented Skill Optimization, which pulls relevant skills from an external corpus and adapts them to a target task and harness. Its two stages create an initial skill without agent rollouts, then update that skill using retrieved knowledge guided by execution feedback. The authors report consistent gains across four agent benchmarks and two models against baselines that do not use retrieval-augmented initialization and updating. ArXiv · AI/CL/LG's note
score 4