Paper 2604.02268
SKILL0: In-Context Agentic Reinforcement Learning for Skill Internalization
- Published
- Apr 2026
- Research lab
- Independent
- Citations
- 47
- GitHub
- 360 stars
01 In brief
Summary
SKILL0 is a reinforcement learning framework that internalizes agent skills into model parameters, eliminating the need for inference-time skill retrieval.
It uses in-context RL during training, where skills are provided as visual context and progressively withdrawn via a Dynamic Curriculum that evaluates each skill's on-policy helpfulness.
Experiments on ALFWorld, Search-QA, and WebShop show SKILL0 outperforms standard RL baselines (e.g., +9.7% on ALFWorld, +6.6% on Search-QA, +10.1% on WebShop accuracy) and achieves competitive results against skill-augmented methods like SkillRL, while maintaining low token costs (under 0.5k per step).
The framework includes relevance-driven skill grouping and a helpfulness-based filter, rank, and select mechanism.
Ablations confirm the importance of the dynamic curriculum and helpfulness filtering.
SKILL0 demonstrates that skills can be internalized, enabling zero-shot autonomous behavior without runtime skill retrieval.
02 From the paper
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, +6.6\% for Search-QA, and+10.1\% for WebShop), 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.