ASA: Training-Free Representation Engineering for Tool-Calling Agents

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Wang, Youjin, Zhou, Run, Fu, Rong, Cao, Shuaishuai, Zeng, Hongwei, Lu, Jiaxuan, Fan, Sicheng, Zhao, Jiaqiao, Pan, Liangming
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917256555921408
author Wang, Youjin
Zhou, Run
Fu, Rong
Cao, Shuaishuai
Zeng, Hongwei
Lu, Jiaxuan
Fan, Sicheng
Zhao, Jiaqiao
Pan, Liangming
author_facet Wang, Youjin
Zhou, Run
Fu, Rong
Cao, Shuaishuai
Zeng, Hongwei
Lu, Jiaxuan
Fan, Sicheng
Zhao, Jiaqiao
Pan, Liangming
contents Adapting LLM agents to domain-specific tool calling remains notably brittle under evolving interfaces. Prompt and schema engineering is easy to deploy but often fragile under distribution shift and strict parsers, while continual parameter-efficient fine-tuning improves reliability at the cost of training, maintenance, and potential forgetting. We identify a critical Lazy Agent failure mode where tool necessity is nearly perfectly decodable from mid-layer activations, yet the model remains conservative in entering tool mode, revealing a representation-behavior gap. We propose Activation Steering Adapter (ASA), a training-free, inference-time controller that performs a single-shot mid-layer intervention and targets tool domains via a router-conditioned mixture of steering vectors with a probe-guided signed gate to amplify true intent while suppressing spurious triggers. On MTU-Bench with Qwen2.5-1.5B, ASA improves strict tool-use F1 from 0.18 to 0.50 while reducing the false positive rate from 0.15 to 0.05, using only about 20KB of portable assets and no weight updates.
format Preprint
id arxiv_https___arxiv_org_abs_2602_04935
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle ASA: Training-Free Representation Engineering for Tool-Calling Agents
Wang, Youjin
Zhou, Run
Fu, Rong
Cao, Shuaishuai
Zeng, Hongwei
Lu, Jiaxuan
Fan, Sicheng
Zhao, Jiaqiao
Pan, Liangming
Software Engineering
Artificial Intelligence
Adapting LLM agents to domain-specific tool calling remains notably brittle under evolving interfaces. Prompt and schema engineering is easy to deploy but often fragile under distribution shift and strict parsers, while continual parameter-efficient fine-tuning improves reliability at the cost of training, maintenance, and potential forgetting. We identify a critical Lazy Agent failure mode where tool necessity is nearly perfectly decodable from mid-layer activations, yet the model remains conservative in entering tool mode, revealing a representation-behavior gap. We propose Activation Steering Adapter (ASA), a training-free, inference-time controller that performs a single-shot mid-layer intervention and targets tool domains via a router-conditioned mixture of steering vectors with a probe-guided signed gate to amplify true intent while suppressing spurious triggers. On MTU-Bench with Qwen2.5-1.5B, ASA improves strict tool-use F1 from 0.18 to 0.50 while reducing the false positive rate from 0.15 to 0.05, using only about 20KB of portable assets and no weight updates.
title ASA: Training-Free Representation Engineering for Tool-Calling Agents
topic Software Engineering
Artificial Intelligence
url https://arxiv.org/abs/2602.04935