An Agent-Oriented Pluggable Experience-RAG Skill for Experience-Driven Retrieval Strategy Orchestration

Researchers present Experience-RAG Skill, an agent-oriented system that dynamically selects optimal retrieval strategies based on task context, rather than using a single fixed pipeline. The system achieves competitive performance across diverse question-answering tasks by leveraging experience memory to orchestrate retrieval, demonstrating that strategy selection can be implemented as a reusable agent component.

The paper addresses a fundamental limitation in retrieval-augmented generation systems: the assumption that one retrieval pipeline works equally well across different task types. Different information retrieval challenges—factoid questions, multi-hop reasoning, and scientific verification—benefit from distinct retrieval strategies, yet most systems apply static approaches. Experience-RAG Skill encapsulates dynamic strategy selection as a pluggable agent layer, creating a more flexible architecture.

This work emerges from ongoing efforts to make RAG systems more adaptive and context-aware. The broader trend reflects the AI community's recognition that one-size-fits-all solutions underperform in production environments handling heterogeneous workloads. By treating retrieval orchestration as a modular agent skill rather than hard-coding it into workflows, the system gains portability and reusability across different applications.

For developers and AI teams, this approach offers practical benefits: improved performance on diverse tasks while maintaining architectural simplicity compared to complex routing mechanisms. The competitive nDCG@10 score of 0.8924 across multiple benchmark datasets suggests the method generalizes effectively. This architecture pattern—encapsulating decision logic as reusable agent components—has implications for building more maintainable and scalable AI systems.

The significance lies not in incremental performance gains but in the architectural insight that strategy selection belongs in the agent layer rather than upper-level workflows. This paradigm could influence how future RAG systems are designed, particularly in enterprise settings where multiple retrieval strategies must coexist. The technique opens questions about what other workflow components could benefit from similar modularization.

• →Experience-RAG Skill dynamically selects retrieval strategies based on task context, outperforming fixed single-retriever approaches.
• →The system achieves 0.8924 nDCG@10 across diverse benchmarks without changing the underlying retriever pool.
• →Retrieval orchestration encapsulated as a reusable agent skill offers better portability than hard-coded workflow logic.
• →Different task types (factoid QA, multi-hop reasoning, scientific verification) benefit from distinct retrieval preferences.
• →The approach suggests retrieval strategy selection is a solvable agent-level problem suitable for production RAG systems.