Hugging Face Daily Papers · May 29, 2026 · 6 min read

Skill0.5: Joint Skill Internalization and Utilization for Out-of-Distribution Generalization in Agentic Reinforcement Learning

Mirrored from Hugging Face Daily Papers for archival readability. Support the source by reading on the original site.

Like Read original ↗

Existing skill-based RL methods force a rigid choice between full externalization (prohibitive context overhead) and full internalization (overfitting and knowledge conflicts). Skill0.5 resolves this by jointly combining general skill internalization with task-specific skill utilization, treating the two inherently different skill types with distinct optimization strategies.\nA dynamic difficulty-aware router streams tasks into mastery tiers:\n<ul>\n<li>Hard tasks: internalize general skills via privileged distillation to build a cognitive foundation;</li>\n<li>Medium tasks: standard RL to maximize task success;</li>\n<li>Easy tasks: diagnostic probing that penalizes shortcuts and enforces faithful task-specific skill utilization.</li>\n</ul>\nExperiments on ALFWorld and WebShop show that Skill0.5 outperforms both memory-based and skill-based RL baselines across in-distribution and out-of-distribution scenarios.\n","updatedAt":"2026-05-29T05:27:38.492Z","author":{"_id":"642b9861bb77f8456634b048","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/642b9861bb77f8456634b048/VrNmmcdgX7FufQmdP5YaG.jpeg","fullname":"Zichen Ding","name":"heroding77","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":12,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.8510344624519348},"editors":["heroding77"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/642b9861bb77f8456634b048/VrNmmcdgX7FufQmdP5YaG.jpeg"],"reactions":[],"isReport":false}},{"id":"6a1a4055ff93f365649edae0","author":{"_id":"63d3e0e8ff1384ce6c5dd17d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1674830754237-63d3e0e8ff1384ce6c5dd17d.jpeg","fullname":"Librarian Bot (Bot)","name":"librarian-bot","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":359,"isUserFollowing":false},"createdAt":"2026-05-30T01:41:41.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"This is an automated message from the [Librarian Bot](https://huggingface.co/librarian-bots). I found the following papers similar to this paper. \n\nThe following papers were recommended by the Semantic Scholar API \n\n* [SKILLC: Learning Autonomous Skill Internalization in LLM Agents via Contrastive Credit Assignment](https://huggingface.co/papers/2605.27899) (2026)\n* [Predictive Representations for Skill Transfer in Reinforcement Learning](https://huggingface.co/papers/2604.07016) (2026)\n* [SKILL0: In-Context Agentic Reinforcement Learning for Skill Internalization](https://huggingface.co/papers/2604.02268) (2026)\n* [SkillMaster: Toward Autonomous Skill Mastery in LLM Agents](https://huggingface.co/papers/2605.08693) (2026)\n* [CoRe-Code: Collaborative Reinforcement Learning for Code Generation](https://huggingface.co/papers/2605.24812) (2026)\n* [SkillOS: Learning Skill Curation for Self-Evolving Agents](https://huggingface.co/papers/2605.06614) (2026)\n* [Skill-R1: Agent Skill Evolution via Reinforcement Learning](https://huggingface.co/papers/2605.09359) (2026)\n\n\n Please give a thumbs up to this comment if you found it helpful!\n\n If you want recommendations for any Paper on Hugging Face checkout [this](https://huggingface.co/spaces/librarian-bots/recommend_similar_papers) Space\n\n You can directly ask Librarian Bot for paper recommendations by tagging it in a comment: `@librarian-bot recommend`","html":"This is an automated message from the <a href=\"https://huggingface.co/librarian-bots\">Librarian Bot</a>. I found the following papers similar to this paper. \nThe following papers were recommended by the Semantic Scholar API \n<ul>\n<li><a href=\"https://huggingface.co/papers/2605.27899\">SKILLC: Learning Autonomous Skill Internalization in LLM Agents via Contrastive Credit Assignment</a> (2026)</li>\n<li><a href=\"https://huggingface.co/papers/2604.07016\">Predictive Representations for Skill Transfer in Reinforcement Learning</a> (2026)</li>\n<li><a href=\"https://huggingface.co/papers/2604.02268\">SKILL0: In-Context Agentic Reinforcement Learning for Skill Internalization</a> (2026)</li>\n<li><a href=\"https://huggingface.co/papers/2605.08693\">SkillMaster: Toward Autonomous Skill Mastery in LLM Agents</a> (2026)</li>\n<li><a href=\"https://huggingface.co/papers/2605.24812\">CoRe-Code: Collaborative Reinforcement Learning for Code Generation</a> (2026)</li>\n<li><a href=\"https://huggingface.co/papers/2605.06614\">SkillOS: Learning Skill Curation for Self-Evolving Agents</a> (2026)</li>\n<li><a href=\"https://huggingface.co/papers/2605.09359\">Skill-R1: Agent Skill Evolution via Reinforcement Learning</a> (2026)</li>\n</ul>\n Please give a thumbs up to this comment if you found it helpful!\n If you want recommendations for any Paper on Hugging Face checkout <a href=\"https://huggingface.co/spaces/librarian-bots/recommend_similar_papers\">this</a> Space\n You can directly ask Librarian Bot for paper recommendations by tagging it in a comment: <code><a href=\"/librarian-bot\">@librarian-bot</a> recommend</code>\n","updatedAt":"2026-05-30T01:41:41.723Z","author":{"_id":"63d3e0e8ff1384ce6c5dd17d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1674830754237-63d3e0e8ff1384ce6c5dd17d.jpeg","fullname":"Librarian Bot (Bot)","name":"librarian-bot","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":359,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.7468864321708679},"editors":["librarian-bot"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/1674830754237-63d3e0e8ff1384ce6c5dd17d.jpeg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2605.28424","authors":[{"_id":"6a186ace9245bafffe463cf4","user":{"_id":"66f547c5405af2cd3256ba27","avatarUrl":"/avatars/fc52bac1785191af0eb9a1d108c4397e.svg","isPro":false,"fullname":"JP Zhu","user":"JPZhu","type":"user","name":"JPZhu"},"name":"Jiapeng Zhu","status":"claimed_verified","statusLastChangedAt":"2026-05-29T09:34:58.510Z","hidden":false},{"_id":"6a186ace9245bafffe463cf5","name":"Jianxiang Yu","hidden":false},{"_id":"6a186ace9245bafffe463cf6","name":"Yibo Zhao","hidden":false},{"_id":"6a186ace9245bafffe463cf7","name":"Chengcheng Han","hidden":false},{"_id":"6a186ace9245bafffe463cf8","name":"Qi Gu","hidden":false},{"_id":"6a186ace9245bafffe463cf9","name":"Xunliang Cai","hidden":false},{"_id":"6a186ace9245bafffe463cfa","name":"Xiang Li","hidden":false},{"_id":"6a186ace9245bafffe463cfb","name":"Weining Qian","hidden":false}],"publishedAt":"2026-05-27T00:00:00.000Z","submittedOnDailyAt":"2026-05-29T00:00:00.000Z","title":"Skill0.5: Joint Skill Internalization and Utilization for Out-of-Distribution Generalization in Agentic Reinforcement Learning","submittedOnDailyBy":{"_id":"642b9861bb77f8456634b048","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/642b9861bb77f8456634b048/VrNmmcdgX7FufQmdP5YaG.jpeg","isPro":false,"fullname":"Zichen Ding","user":"heroding77","type":"user","name":"heroding77"},"summary":"Equipping large language models with explicit skills has emerged as a promising paradigm for enabling autonomous agents to solve complex tasks. Agent skills can be inherently divided into general skills for broad cognitive transfer and task-specific skills for dynamic execution. However, existing skill-based reinforcement learning (RL) methods typically force a rigid choice between full externalization, which incurs prohibitive context overhead, and full internalization, which risks overfitting and knowledge conflicts. To address this dilemma, we propose Skill0.5, a novel agentic RL framework that explicitly differentiates skill treatments by combining general skill internalization with task-specific skill utilization. Driven by a dynamic, difficulty-aware router, Skill0.5 streams tasks into distinct mastery tiers to apply tailored optimization strategies: it internalizes general skills via privileged distillation to build a cognitive foundation for hard tasks, while using diagnostic probing on easy tasks to penalize shortcuts and enforce specific skill utilization. Experiments on ALFWorld and WebShop demonstrate that Skill0.5 outperforms both memory-based and skill-based RL baselines, yielding performance improvements across both in-distribution and out-of-distribution scenarios.","upvotes":17,"discussionId":"6a186ace9245bafffe463cfc","githubRepo":"https://github.com/JasonZhujp/Skill0_5","githubRepoAddedBy":"user","ai_summary":"Skill0.5 is a novel agentic reinforcement learning framework that combines general skill internalization with task-specific skill utilization through a dynamic, difficulty-aware router to improve performance in complex task environments.","ai_keywords":["skill-based reinforcement learning","agentic RL","privileged distillation","diagnostic probing","dynamic router","mastery tiers","cognitive foundation","out-of-distribution scenarios"],"githubStars":6,"organization":{"_id":"62c1482aac1b639c2d873235","name":"ECNU","fullname":"East China Normal University","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/1656834075081-62c14609ac1b639c2d87192c.png"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"66f547c5405af2cd3256ba27","avatarUrl":"/avatars/fc52bac1785191af0eb9a1d108c4397e.svg","isPro":false,"fullname":"JP Zhu","user":"JPZhu","type":"user"},{"_id":"642b9861bb77f8456634b048","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/642b9861bb77f8456634b048/VrNmmcdgX7FufQmdP5YaG.jpeg","isPro":false,"fullname":"Zichen Ding","user":"heroding77","type":"user"},{"_id":"64eb4f5507987950ae5e2b0f","avatarUrl":"/avatars/77dc21b195bc94490e45bfe208abfbc4.svg","isPro":false,"fullname":"Jiapeng Zhu","user":"JasonZhujp","type":"user"},{"_id":"64d09c16c0c627dfa7f22599","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/64d09c16c0c627dfa7f22599/TCV-PmAmPcbRpd2Nc11CL.jpeg","isPro":false,"fullname":"jianxiangyu","user":"ffjasonyu","type":"user"},{"_id":"65840dd94b2f0e9f340acdd1","avatarUrl":"/avatars/aa4f8a23e63af370a25fba5f7d2cefbe.svg","isPro":false,"fullname":"Aquarius","user":"AquariusAQ","type":"user"},{"_id":"6800ee4cb4d9835c1ddc5b76","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/X3HvO7o_owLGR-WiUmsy6.png","isPro":false,"fullname":"yujing","user":"allacnobug","type":"user"},{"_id":"690d8ff1238a29cc19018445","avatarUrl":"/avatars/1b4cab1c4b04cd72ee63a3dd040ae5f4.svg","isPro":false,"fullname":"qinnan cai","user":"qinnancai0115","type":"user"},{"_id":"6475a1094766357252e5cb17","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6475a1094766357252e5cb17/QcvbjQLbuyDSU-XH_lW75.jpeg","isPro":false,"fullname":"zhaoyibo","user":"ybyby624","type":"user"},{"_id":"69707f67d62a921e227e94e2","avatarUrl":"/avatars/e438539bcdef63a590a6015e9574819f.svg","isPro":false,"fullname":"Zixuan Yang","user":"Luli3220","type":"user"},{"_id":"6708edcae69f6e30a816af9f","avatarUrl":"/avatars/c4daa9b0cb2f4bb2a7db0e78b22034cb.svg","isPro":false,"fullname":"Yao","user":"distant-yuan","type":"user"},{"_id":"6a192694dc238681b74ba29b","avatarUrl":"/avatars/29cde8ed0077dc7692dcf28f184beeb5.svg","isPro":false,"fullname":"uncertainty-hash","user":"Uncertainty-Hash","type":"user"},{"_id":"660d0633f04ebfca62704513","avatarUrl":"/avatars/7e7416e5d0dc4ec4380aa5b9c4ea03bb.svg","isPro":false,"fullname":"Jiaming Zhang","user":"DonFinliani","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"organization":{"_id":"62c1482aac1b639c2d873235","name":"ECNU","fullname":"East China Normal University","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/1656834075081-62c14609ac1b639c2d87192c.png"},"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2605/2605.28424.md"}">

Papers

arxiv:2605.28424

Skill0.5: Joint Skill Internalization and Utilization for Out-of-Distribution Generalization in Agentic Reinforcement Learning

Published on May 27

· Submitted by

Zichen Ding on May 29

East China Normal University

Upvote

Authors:

Jiapeng Zhu ,

Abstract

Skill0.5 is a novel agentic reinforcement learning framework that combines general skill internalization with task-specific skill utilization through a dynamic, difficulty-aware router to improve performance in complex task environments.

AI-generated summary

Equipping large language models with explicit skills has emerged as a promising paradigm for enabling autonomous agents to solve complex tasks. Agent skills can be inherently divided into general skills for broad cognitive transfer and task-specific skills for dynamic execution. However, existing skill-based reinforcement learning (RL) methods typically force a rigid choice between full externalization, which incurs prohibitive context overhead, and full internalization, which risks overfitting and knowledge conflicts. To address this dilemma, we propose Skill0.5, a novel agentic RL framework that explicitly differentiates skill treatments by combining general skill internalization with task-specific skill utilization. Driven by a dynamic, difficulty-aware router, Skill0.5 streams tasks into distinct mastery tiers to apply tailored optimization strategies: it internalizes general skills via privileged distillation to build a cognitive foundation for hard tasks, while using diagnostic probing on easy tasks to penalize shortcuts and enforce specific skill utilization. Experiments on ALFWorld and WebShop demonstrate that Skill0.5 outperforms both memory-based and skill-based RL baselines, yielding performance improvements across both in-distribution and out-of-distribution scenarios.

View arXiv page View PDF GitHub 6 Add to collection

Community

heroding77

Paper submitter 1 day ago

Existing skill-based RL methods force a rigid choice between full externalization (prohibitive context overhead) and full internalization (overfitting and knowledge conflicts). Skill0.5 resolves this by jointly combining general skill internalization with task-specific skill utilization, treating the two inherently different skill types with distinct optimization strategies.

A dynamic difficulty-aware router streams tasks into mastery tiers:

Hard tasks: internalize general skills via privileged distillation to build a cognitive foundation;
Medium tasks: standard RL to maximize task success;
Easy tasks: diagnostic probing that penalizes shortcuts and enforces faithful task-specific skill utilization.

Experiments on ALFWorld and WebShop show that Skill0.5 outperforms both memory-based and skill-based RL baselines across in-distribution and out-of-distribution scenarios.