SKILL0: In-Context Agentic Reinforcement Learning for Skill Internalization

arXiv cs.LG / 4/3/2026

💬 OpinionSignals & Early TrendsIdeas & Deep AnalysisModels & Research

共有:

Key Points

The paper introduces SKILL0, an in-context reinforcement learning approach that aims to internalize agent skills into model parameters rather than relying on inference-time skill retrieval.
SKILL0 uses a training curriculum that gradually withdraws full skill context, while grouping skills offline by category and rendering them with interaction history as compact visual prompts for learning tool invocation and multi-turn task completion.
A Dynamic Curriculum evaluates each skill file’s on-policy helpfulness and retains only skills that continue to improve performance within a decaying token/interaction budget, eventually enabling fully zero-shot behavior without runtime retrieval.
Experiments show SKILL0 improves over a standard RL baseline, reporting +9.7% on ALFWorld and +6.6% on Search-QA, while keeping per-step context under 0.5k tokens.
The authors release code at a public GitHub repository, supporting reproducibility and further exploration of skill internalization.

Abstract

Agent skills, structured packages of procedural knowledge and executable resources that agents dynamically load at inference time, have become a reliable mechanism for augmenting LLM agents. Yet inference-time skill augmentation is fundamentally limited: retrieval noise introduces irrelevant guidance, injected skill content imposes substantial token overhead, and the model never truly acquires the knowledge it merely follows. We ask whether skills can instead be internalized into model parameters, enabling zero-shot autonomous behavior without any runtime skill retrieval. We introduce SKILL0, an in-context reinforcement learning framework designed for skill internalization. SKILL0 introduces a training-time curriculum that begins with full skill context and progressively withdraws it. Skills are grouped offline by category and rendered with interaction history into a compact visual context, teaching he model tool invocation and multi-turn task completion. A Dynamic Curriculum then evaluates each skill file's on-policy helpfulness, retaining only those from which the current policy still benefits within a linearly decaying budget, until the agent operates in a fully zero-shot setting. Extensive agentic experiments demonstrate that SKILL0 achieves substantial improvements over the standard RL baseline (+9.7\% for ALFWorld and +6.6\% for Search-QA), while maintaining a highly efficient context of fewer than 0.5k tokens per step. Our code is available at https://github.com/ZJU-REAL/SkillZero.