Toward Skill-Native LLMs: Skill Entropy for Benchmarking and Training Long-Horizon Reasoning

• URL: http://arxiv.org/abs/2608.05139v1
• Date: Wed, 05 Aug 2026 17:57:16 GMT
• Title: Toward Skill-Native LLMs: Skill Entropy for Benchmarking and Training Long-Horizon Reasoning
• Abstract summary: Long-horizon reasoning demands that the model switch between distinct skills inside a reasoning chain.<n>We introduce Skill Entropy, a measure of the difficulty of switching from one skill to another.<n>We propose Skill-Entropy RL, an RL framework where the model predicts not only the answer at each step but also the skill used to produce it.
• Score: 72.09826781730662
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Long-horizon reasoning in recent LLMs demands that the model switch between distinct skills inside a reasoning chain, such as first doing a math derivation, then using the result to plan a schedule. We call such problems cross-skill long-horizon tasks: multi-step tasks whose steps require different reasoning skills and depend on earlier outputs. Existing benchmarks often evaluate individual skills, lacking a principled way to measure how well a model switches between skills. We address this gap from both the evaluation and training sides. We introduce Skill Entropy, a measure of the difficulty of switching from one skill to another. We then propose Skill^2-Bench, a benchmark of cross-skill long-horizon tasks built over 558 skills across 9 verifiable and open-ended domains. Each task is assigned a task-level skill-entropy score and grouped into three difficulty levels. Evaluating 8 frontier and 4 open-source models on Skill^2-Bench reveals a skill-switching gap: accuracy decreases on higher-entropy tasks. We then turn skill entropy from a benchmark scale into a training signal. We propose Skill-Entropy RL, an RL framework where the model predicts not only the answer at each step but also the skill used to produce it. The reward combines step-level correctness with a skill-entropy reward that measures the alignment between the model-predicted skill sequence and the gold skill sequence. On Qwen3-4B-Instruct and Qwen3-1.7B, Skill-Entropy RL improves the Skill^2-Bench score from 34.4% to 68.4% and from 14.6% to 40.1%, respectively, outperforming competitive baselines. The same pipeline can be applied to off-the-shelf training data such as OpenR1-Math, indicating that skill entropy is a reusable training signal. Code available at: https://github.com/Gen-Verse/Skill-Entropy-RL

Related papers

• Not All Skills Help: Measuring and Repairing Agent Knowledge [47.498075849595374]
  ASSAY is a framework that separates generation from curation.<n>It computes a per-skill causal attribution on a small development set.<n>It restructures the library offline, and suppresses skills with negative predicted effect for each test task.
  arXiv  Detail & Related papers  (2026-06-13T16:37:38Z)
• SkillsInjector: Dynamic Skill Context Construction for LLM Agents [16.631471381875816]
  Existing methods treat skill injection as a static step, selecting skills with fixed criteria, fixing the budget in advance, and leaving descriptions unchanged.<n>We propose SkillsInjector, a two-stage adaptive method that jointly addresses these decisions.<n>Across tau2-bench, SkillsBench, and ALFWorld, SkillsInjector achieves the highest score, improving over the strongest baseline by 3.9, 6.1, and 7.3 percentage points, respectively.
  arXiv  Detail & Related papers  (2026-05-28T11:44:32Z)
• SkillGraph: Skill-Augmented Reinforcement Learning for Agents via Evolving Skill Graphs [70.1970574147839]
  We propose SKILLGRAPH, a framework that represents reusable skills as nodes in a directed graph.<n>SKILLGRAPH retrieves not just individual skills, but an ordered skill subgraph that can guide multi-step decision making.<n> Experiments show that SKILLGRAPH achieves state-of-the-art performance against memory-augmented RL methods.
  arXiv  Detail & Related papers  (2026-05-12T12:21:49Z)
• Skill-R1: Agent Skill Evolution via Reinforcement Learning [84.35984979949502]
  Skill-R1 is a reinforcement learning framework for instance-level recurrent skill optimization from verifiable rewards.<n>It preserves black-box compatibility with both open- and closed-source models while making adaptation substantially cheaper than model-level updates.
  arXiv  Detail & Related papers  (2026-05-10T06:19:15Z)
• SkillGen: Verified Inference-Time Agent Skill Synthesis [60.927977774369516]
  SkillGen is a framework that synthesizes a single auditable skill from trajectories generated by a base agent.<n>It identifies reusable success patterns, recurring failure modes, and behaviors that appear in nearby successes but are missing from failures.<n>A key novelty in SkillGen is that we model agent skills as interventions to empirically verify the net effect of skills on the overall performance.
  arXiv  Detail & Related papers  (2026-05-09T19:24:11Z)
• SkillsBench: Benchmarking How Well Agent Skills Work Across Diverse Tasks [61.89812116484928]
  Agent Skills are structured packages of procedural knowledge that augment LLM agents at inference time.<n>We present SkillsBench, a benchmark of 86 tasks across 11 domains paired with curated Skills and deterministic verifiers.<n>We show curated Skills raise average pass rate by 16.2 percentage points(pp), but effects vary widely by domain.<n>Self-generated Skills provide no benefit on average, showing that models cannot reliably author the procedural knowledge they benefit from consuming.
  arXiv  Detail & Related papers  (2026-02-13T07:06:06Z)
• Pretrained Bayesian Non-parametric Knowledge Prior in Robotic Long-Horizon Reinforcement Learning [10.598207472087578]
  Reinforcement learning (RL) methods typically learn new tasks from scratch, often disregarding prior knowledge that could accelerate the learning process.<n>This work introduces a method that models potential primitive skill motions as having non-parametric properties with an unknown number of underlying features.<n>We utilize a non-parametric model, specifically Dirichlet Process Mixtures, enhanced with birth and merge, to pre-train a skill prior that effectively captures the diverse nature of skills.
  arXiv  Detail & Related papers  (2025-03-27T20:43:36Z)
• Skill Reinforcement Learning and Planning for Open-World Long-Horizon Tasks [31.084848672383185]
  We study building multi-task agents in open-world environments. We convert the multi-task learning problem into learning basic skills and planning over the skills. Our method accomplishes 40 diverse Minecraft tasks, where many tasks require sequentially executing for more than 10 skills.
  arXiv  Detail & Related papers  (2023-03-29T09:45:50Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.