ZJU-REAL/SkillZero

[EMNLP 2026] Official code for "SKILL0: In-Context Agentic Reinforcement Learning for Skill Internalization"

What it solves

SKILL0 is designed to address the challenge of skill internalization in AI agents. It aims to improve how agents learn and internalize skills from their environment and interactions, allowing them to perform better on complex tasks like those found in ALFWorld and Search-QA.

How it works

SKILL0 is an in-context reinforcement learning (RL) framework. It uses agentic RL to enable the model to internalize skills, moving from relying on external guidance or prompts to internalizing those capabilities within the policy itself.

Who it’s for

Researchers and developers working on reinforcement learning, agentic AI, and the ability of LLM-based agents to acquire and internalize complex skills through experience.

Highlights

  • In-Context RL: Leverages in-context learning to facilitate the internalization of skills.
  • Performance Gains: Demonstrates substantial improvements over standard RL baselines on the ALFWorld and Search-QA benchmarks.
  • Environment Support: Includes built-in support for complex environments like ALFWorld and Search-QA.

Related

  • Project
    rllm-org/rllm

    An open-source framework for training language agents with reinforcement learning, allowing the same agent code to be used across different backends, sandboxes, and benchmarks.

  • Dispatch
    Stochastic Neural Networks for Hierarchical Reinforcement Learning

    OpenAI researchers propose a framework using Stochastic Neural Networks and information-theoretic regularizers to pre-train interpretable skills that accelerate learning in downstream tasks with sparse rewards.

  • Project
    TsinghuaC3I/MARTI

    An open-source framework for training LLM-based multi-agent systems using reinforcement learning, featuring tree search augmentation for complex reasoning and code generation.

  • Dispatch
    OpenAI RL²: Fast Reinforcement Learning via Slow Reinforcement Learning

    OpenAI introduces RL², a method that uses a recurrent neural network (RNN) to learn a reinforcement learning algorithm, allowing agents to adapt to new tasks in a few trials by encoding the learning process itself into the RNN weights.