Newest papers

Robust Soft Skill Adaptation via Behavioral Localization in Tool-Using Agents

Zijian Jiang, Xilu Wang · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.30041 · cited by 0 works (OpenAlex) · open-access PDF

French

Les compétences textuelles peuvent étendre les capacités des agents utilisant des outils, mais les descriptions de compétences longues augmentent les coûts de calcul. Les prompts doux fournissent une alternative compacte, mais l'apprentissage à partir de trajectoires réussies entières peut également introduire des changements comportementaux inutiles. Nous constatons que l'effet utile d'une compétence textuelle est très concentré dans un petit sous-ensemble de positions de trajectoire. Sur la base de cette observation, nous proposons la localisation comportementale, qui compare l'agent gelé avec et sans la compétence textuelle pour identifier les positions où la compétence augmente à la fois la probabilité de tokens réussies et modifie significativement la prochaine distribution de

Machine translation (Meta M2M100); the English original follows.

Abstract

Text skills can extend the capabilities of tool-using agents, but long skill descriptions increase computational costs. Soft prompts provide a compact alternative, yet learning from entire successful trajectories may also introduce unnecessary behavioral changes. We find that the useful effect of a text skill is highly concentrated in a small subset of trajectory positions. Based on this observation, we propose Behavioral Localization, which compares the frozen agent with and without the text skill to identify positions where the skill both increases the likelihood of successful tokens and meaningfully changes the next token distribution. We use these localized positions to supervise a short soft prompt. Experiments on SpreadsheetBench with a frozen Qwen3.6-35B-A3B agent show that the top 10\% of trajectory positions capture 96.12\% of the total positive skill gain. In the shared prompt setting, selective supervision improves success from 30.36\% with full trajectory to 36.79\%. Under task-specific conditions, using the top 10\% localized positions achieves 57.38\%, outperforming both the full text skill and the on-policy OPCD baseline. These results show that a small set of behaviorally relevant trajectory positions can provide effective supervision of soft skills.

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Trajectory, Set (abstract data type), Computer science, Adaptation (eye), Artificial intelligence, Security token

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange