From Relevance to Execution Utility: Reward-Aware Dynamic Execution Gating for Skill-Based LLM Agents

2026-08-10

AI Summary

This work addresses a critical limitation in existing skill retrieval methods, which, despite generating plausible skill compositions, lack mechanisms to assess execution utility—leading to substantial computational waste on ineffective reasoning steps. To remedy this, the authors propose RADEG, a lightweight decision layer that introduces an execution utility prediction mechanism between retrieval and execution. RADEG generates contrastive data for the same query through local perturbations (deletion, addition, or substitution) and employs a warm-started logistic regression head coupled with reward-aware dynamic gating to enable online adaptation without retraining. Experiments on 288 held-out queries demonstrate that RADEG significantly reduces unnecessary executions while preserving most downstream task rewards, consistently outperforming relevance-based and random baseline strategies across varying execution budgets.

Abstract

Agent skills are increasingly used to equip large language model (LLM) agents with reusable procedural knowledge. Although recent work has substantially improved skill retrieval due to the increasing skill libraries, retrieving a plausible skill bundle does not guarantee that executing it is worthwhile. Since every skill-conditioned rollout is computationally expensive, deciding whether a retrieved bundle should be executed has become an increasingly important challenge. To this end, we introduce the Reward-Aware Dynamic Execution Gate (RADEG), a lightweight, retriever-agnostic decision layer between skill retrieval and agent execution. RADEG learns a low-cost surrogate model that predicts the execution utility of a query--bundle pair before the expensive rollout is launched. To obtain informative supervision while controlling for task difficulty, we locally perturb each retrieved bundle by deleting, adding, or replacing one skill, producing matched same-query rollouts that isolate the effect of bundle composition on verifier reward. During deployment, RADEG updates only a warm-started logistic head as new verifier feedback becomes available, enabling inexpensive adaptation of the execute/skip boundary without retraining either the retriever or the agent. Under a query-level held-out evaluation on 288 collected rollouts, RADEG substantially reduces unnecessary agent executions while preserving a large fraction of the downstream verifier reward. It consistently outperforms relevance-based and random gating across different execution budgets, demonstrating that execution-aware surrogate modeling provides a practical and cost-effective complement to skill retrieval.

Problem

• skill-based LLM agents
• execution utility
• reward-aware gating
• skill retrieval
• computational cost

Innovation

• execution utility
• reward-aware gating
• skill-based LLM agents
• surrogate modeling
• dynamic execution gating