Task Decomposition-Guided Reranking for Adaptive Agent Skill Retrieval

2026-07-07

AI Summary

Existing skill selection methods struggle to achieve precise task-skill matching due to their neglect of semantic ambiguity between tasks and skills, as well as the dynamic coupling between task difficulty and skill applicability. To address this, this work proposes SkillReranker, a framework that, during inference, constructs a directed acyclic execution graph through semantic decomposition to identify subtask intervals and incorporates a task-decomposition-guided reranking mechanism. This mechanism dynamically adjusts the ranking of candidate skills by integrating cross-encoders with large language models. Evaluated on ALFWorld and ScienceWorld, the proposed approach significantly improves task completion rates while reducing both environmental interaction steps and token consumption, outperforming current state-of-the-art baselines.

Abstract

Skill usage can significantly enhance the ability of modern agent systems to complete complex tasks. However, the growing scale of skill libraries makes accurate skill selection increasingly challenging. In real-world scenarios, ambiguous semantic matching often arises between a specific task requirement and multiple generic yet semantically similar candidate skills. Moreover, existing methods tend to overlook the dynamic influence of task difficulty and skill applicability when selecting the optimal target skill set. To address these issues, we propose SkillReranker, an inference-time reranking framework for adaptive skill selection. Specifically, we first perform semantic decomposition on both the task and skill sides, yielding informative subtask and execution-state descriptions as well as transition-state descriptions that characterize each skill's functionality. These descriptions are then used to construct a directed acyclic execution graph, where intermediate task states are modeled as nodes and candidate skills as edges, thereby establishing a structured task-skill correspondence. On this basis, SkillReranker determines whether each state node satisfies the split condition to identify subtask intervals. For each task interval, we employ a cross-encoder to perform comprehensive scoring over candidate skills and select the most suitable ones to form the final target skill set. Experiments on ALFWorld and ScienceWorld with three backbone LLMs show that SkillReranker effectively improves task performance, reduces environment interaction steps, and lowers token consumption compared with existing skill selection baselines.

Problem

• skill retrieval
• task decomposition
• semantic ambiguity
• adaptive selection
• agent systems

Innovation

• task decomposition
• skill retrieval
• reranking
• execution graph
• adaptive selection