{"id":52101816,"url":"https://github.com/amazon-science/reskill","last_synced_at":"2026-08-04T16:30:53.166Z","repository":{"id":373453674,"uuid":"1258876319","full_name":"amazon-science/reskill","owner":"amazon-science","description":"An easy-to-configure and extensible veRL extension for agent RL training with skill co-evolution.","archived":false,"fork":false,"pushed_at":"2026-07-27T21:14:15.000Z","size":3344,"stargazers_count":22,"open_issues_count":1,"forks_count":2,"subscribers_count":1,"default_branch":"main","last_synced_at":"2026-07-27T23:08:02.483Z","etag":null,"topics":[],"latest_commit_sha":null,"homepage":"https://amazon-science.github.io/reskill/","language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"apache-2.0","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/amazon-science.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":"CONTRIBUTING.md","funding":null,"license":"LICENSE","code_of_conduct":"CODE_OF_CONDUCT.md","threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null,"notice":"NOTICE","maintainers":null,"copyright":null,"agents":null,"dco":null,"cla":null}},"created_at":"2026-06-04T02:13:35.000Z","updated_at":"2026-07-27T21:13:00.000Z","dependencies_parsed_at":null,"dependency_job_id":null,"html_url":"https://github.com/amazon-science/reskill","commit_stats":null,"previous_names":["amazon-science/reskill"],"tags_count":null,"template":false,"template_full_name":"amazon-archives/__template_Apache-2.0","purl":"pkg:github/amazon-science/reskill","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/amazon-science%2Freskill","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/amazon-science%2Freskill/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/amazon-science%2Freskill/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/amazon-science%2Freskill/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/amazon-science","download_url":"https://codeload.github.com/amazon-science/reskill/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/amazon-science%2Freskill/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":36283782,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-07-20T02:08:10.276Z","status":"online","status_checked_at":"2026-08-04T02:00:06.901Z","response_time":57,"last_error":null,"robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":true,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":[],"created_at":"2026-08-04T16:30:51.069Z","updated_at":"2026-08-04T16:30:53.161Z","avatar_url":"https://github.com/amazon-science.png","language":"Python","funding_links":[],"categories":[],"sub_categories":[],"readme":"# ReSkill\n\n*An easy-to-configure, extensible veRL extension that brings the Anthropic\nSkill Creator into agentic RL training. Full control over skill versioning,\nsampling, bundle testing, and skill-policy co-evolution.*\n\nOfficial code for the paper:\n**ReSkill: Reconciling Skill Creation with Policy Optimization in Agentic RL**.\n\n[![Paper](https://img.shields.io/badge/Paper-arXiv%202606.01619-b31b1b)](https://arxiv.org/abs/2606.01619) [![Project Page](https://img.shields.io/badge/Project-Page-232F3E)](https://amazon-science.github.io/reskill/) [![veRL](https://img.shields.io/badge/built%20on-veRL%20v0.6.1-blue)](https://github.com/verl-project/verl/tree/d62da4950573d7a4b7ef2362337952e7ab59e78d) [![License](https://img.shields.io/badge/License-Apache%202.0-green.svg)](LICENSE)\n\n---\n\n## 🔥 News\n\n- **[2026-06]** 🎉 Paper and codebase are now public. More are on the way... stay tracked!\n\n---\n\n## 🧩 System Overview\n\n\u003cp align=\"center\"\u003e\n  \u003cimg src=\"assets/reskill_system_overview.png\" alt=\"ReSkill overview: RL-in-the-loop skill creation and reconciled skill-policy updates\" width=\"920\"\u003e\n\u003c/p\u003e\n\n\u003cp align=\"center\"\u003e\n  \u003cem\u003e(a) Inspired by Anthropic's human-in-the-loop Skill Creator, ReSkill recasts skill creation as an RL-in-the-loop process. (b) Compared with decoupled skill-update methods, ReSkill exposes a highly configurable loop for jointly evolving skills and policies.\u003c/em\u003e\n\u003c/p\u003e\n\nReSkill combines three pieces:\n\n- **RL training with per-turn skill customization**: veRL handles distributed RL, while\n  ReSkill follows the [verl-agent](https://github.com/langfengq/verl-agent)\n  design of decomposing multi-turn agent rollouts and adds skill loading into\n  each turn.\n- **RL-in-the-loop skill creation**: ReSkill adapts the structure of\n  [Anthropic's skill creator](https://github.com/anthropics/skills/blob/main/skills/skill-creator/SKILL.md)\n  into an RL feedback loop for analyzing rollout experience and proposing skill\n  updates during training.\n- **Skill versioning and sampling**: ReSkill tracks skill versions, loads active\n  skills, samples/testing skill bundles, and supports skill-policy\n  co-evolution over training.\n\n## ⚙️ Installation\n\n```bash\ngit clone https://github.com/amazon-science/reskill.git\ncd reskill\ngit submodule update --init --recursive verl\npip install -e .\n```\n\nInstall only the benchmark and backend extras you need:\n\n```bash\npip install -e \".[\u003cenv\u003e,vllm]\"\n```\n\nValidated stack pins are recorded under `requirements/`.\n\nThe current benchmark extras are `alfworld`, `search`, and `scienceworld`.\nAdditional environment support will be added over time.\n\n## 🚀 Usage\n\nPrepare data for an environment:\n\n```bash\npython scripts/data_prep/prepare_\u003cenv\u003e.py --output_dir data/\u003cenv\u003e\n```\n\nRun training:\n\n```bash\npython scripts/train.py --config-name \u003cenv\u003e\n```\n\nConcrete configs live under `configs/`, and cluster launch examples live under\n`scripts/launch/`.\n\n## 🛠️ Customize ReSkill\n\nReSkill is designed so both sides of the co-evolution loop can be customized.\n\n- **Policy side**: customize the environment, rollout format, action projection,\n  rewards, group rollout settings, and backend profiles.\n- **Skill side**: customize skill-generation prompts, trigger behavior, active\n  skill budgets, version testing/sampling, and skill library persistence.\n\n## 📢 Release Note\n\n\u003e This codebase is under active restructuring and testing as we work toward a stable release. Thank you for your patience and interest!\n\n## 🗺️ Roadmap\n\n- Track newer veRL releases.\n- Add SGLang rollout backend support.\n- Add backend config profiles for vLLM and SGLang.\n- Expand validated environment examples.\n\n## 🙏 Acknowledgements\n\nWe thank the contributors to [veRL](https://github.com/volcengine/verl),\n[verl-agent](https://github.com/langfengq/verl-agent), and\n[Anthropic Skill Creator](https://github.com/anthropics/skills/blob/main/skills/skill-creator/SKILL.md)\nfor their open-source foundations and inspiration, which ReSkill builds upon.\n\n## 📄 License\n\nApache 2.0\n\n## 📚 Citation\n\nIf you find this work helpful, please kindly consider citing our paper and\nstarring the repository.\n\n```bibtex\n@article{he2026reskill,\n  title={ReSkill: Reconciling Skill Creation with Policy Optimization in Agentic RL},\n  author={He, Zelin and Lin, Haotian and Han, Boran and Zhu, Wei and Fang, Haoyang and Wang, Bernie and Zhu, Xuan and Li, Runze and Reimherr, Matthew},\n  journal={arXiv preprint arXiv:2606.01619},\n  year={2026}\n}\n```\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Famazon-science%2Freskill","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Famazon-science%2Freskill","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Famazon-science%2Freskill/lists"}