{"id":"arxiv:2601.21557","title":"Meta Context Engineering via Agentic Skill Evolution","date":"2026-01-29","section":"Agents, open-endedness, AGI","quality":0.035459,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/","source_url":"https://arxiv.org/abs/2601.21557","text_available":true,"authors":["Haoran Ye","Xuning He","Vincent Arak","Haonan Dong"],"outgoing":[{"source":"arxiv:2601.21557","target":"arxiv:2407.21783","evidence":"id","id":"arxiv:2407.21783","title":"The Llama 3 Herd of Models","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/de3cebb4c28bc5d41a843279f219dca14476284991ea5c7c0aeab7068a40e306/"},{"source":"arxiv:2601.21557","target":"arxiv:2505.22954","evidence":"id","id":"arxiv:2505.22954","title":"Darwin Godel Machine: Open-Ended Evolution of Self-Improving Agents","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/c29f716871f80314868b5647692de387f26e12166b93bd362915fd272b6e081c/"},{"source":"arxiv:2601.21557","target":"arxiv:2507.18074","evidence":"id","id":"arxiv:2507.18074","title":"AlphaGo Moment for Model Architecture Discovery","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1c22da54910a26f1f81b276f5cae3a81e22f7db024f7d1a5fa89ccd4f291ffa8/"},{"source":"arxiv:2601.21557","target":"arxiv:2507.19457","evidence":"id","id":"arxiv:2507.19457","title":"GEPA: Reflective Prompt Evolution Can Outperform Reinforcement Learning","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/c1ef72d96fa3292590a6a3d112ab509840081b59aaa771cf42b8b0e8ef62ba08/"},{"source":"arxiv:2601.21557","target":"arxiv:2512.24601","evidence":"id","id":"arxiv:2512.24601","title":"Recursive Language Models","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/ac41983c934a3432ce7b6531e367b9e639bddbed93d4c6513781e76407d9a8bd/"},{"source":"arxiv:2601.21557","target":"doi:10.1038/s41586-023-06924-6","evidence":"title","id":"doi:10.1038/s41586-023-06924-6","title":"Mathematical discoveries from program search with large language models","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/70a6b427d42fd28c9d538648e148a97db2fb2614c7764f3231e53f4bebdce887/"},{"source":"arxiv:2601.21557","target":"url:sakana.ai/shinka-evolve","evidence":"id","id":"url:sakana.ai/shinka-evolve","title":"ShinkaEvolve: Towards Open-Ended And Sample-Efficient Program Evolution","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/a13eace17f3407cb1634b579b2738f15f1e035897a8f1842305198c6d2bf35e1/"}],"incoming":[{"source":"arxiv:2603.19461","target":"arxiv:2601.21557","evidence":"id","id":"arxiv:2603.19461","title":"Hyperagents","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/17eec8a15debc212b5530f117dd8ba8bad339590729cbbc4f75a47755ebcb544/"},{"source":"arxiv:2603.28052","target":"arxiv:2601.21557","evidence":"id","id":"arxiv:2603.28052","title":"Meta-Harness: End-to-End Optimization of Model Harnesses","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/8bb0b87384dfdd42c419cf26cc7e52f1f0d89f63afa5d5968e246b73a44c99cb/"},{"source":"arxiv:2607.07847","target":"arxiv:2601.21557","evidence":"id","id":"arxiv:2607.07847","title":"When Does Continual Learning Require Learning","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/86e88c6cf21a66879eab5c156913fc85d14eb69c394c78c33c0e81d144e528eb/"},{"source":"url:lilianweng.github.io/posts/2026-07-04-harness","target":"arxiv:2601.21557","evidence":"id","id":"url:lilianweng.github.io/posts/2026-07-04-harness","title":"Harnesses for models (Lilian Weng, 2026)","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/c1ea249334ba128a9fff468e25fa4f8d13d4cccd13661ee130bec020ebb6463b/"}],"text_parts":[{"start":0,"end":15999,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-1.html"},{"start":15999,"end":31847,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-2.html"},{"start":31847,"end":47660,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-3.html"},{"start":47660,"end":62573,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-4.html"}],"contents":[{"title":"1 Introduction","offset":0,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-1.html#text"},{"title":"2 Background and Related Work","offset":6557,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-1.html#text"},{"title":"2.1 Agentic Context Engineering","offset":6591,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-1.html#text"},{"title":"2.2 Evolutionary Computation with LLMs","offset":10234,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-1.html#text"},{"title":"3 Meta Context Engineering","offset":12405,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-1.html#text"},{"title":"3.1 Problem Formulation","offset":13048,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-1.html#text"},{"title":"3.2 Meta-Level: Agentic Skill Evolution","offset":16628,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-2.html#text"},{"title":"3.3 Base-Level: Fully Agentic Context Optimization","offset":19588,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-2.html#text"},{"title":"3.4 Algorithmic Orchestration","offset":21036,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-2.html#text"},{"title":"4 Experiments","offset":23102,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-2.html#text"},{"title":"4.1 Experimental Setup","offset":23120,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-2.html#text"},{"title":"Tasks and Datasets.","offset":23148,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-2.html#text"},{"title":"Baselines.","offset":24632,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-2.html#text"},{"title":"Models.","offset":25820,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-2.html#text"},{"title":"MCE Instantiation.","offset":26600,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-2.html#text"},{"title":"Evaluation Settings.","offset":27058,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-2.html#text"},{"title":"4.2 Main Results","offset":27735,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-2.html#text"},{"title":"4.2.1 Context Performance","offset":27757,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-2.html#text"},{"title":"4.2.2 Context Adaptability, Efficiency, and Transferability","offset":30198,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-2.html#text"},{"title":"4.2.3 Training Efficiency","offset":32794,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-3.html#text"},{"title":"4.3 Ablation Studies","offset":33795,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-3.html#text"},{"title":"Ablating the Bi-level Design ( Table 3 ).","offset":34176,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-3.html#text"},{"title":"Ruling Out Model Confounds ( Table 4 ).","offset":35275,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-3.html#text"},{"title":"5 Conclusion and Discussion","offset":36209,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-3.html#text"},{"title":"Appendix A Experimental Details","offset":39712,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-3.html#text"},{"title":"A.1 FiNER","offset":39928,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-3.html#text"},{"title":"A.2 USPTO-50k","offset":41214,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-3.html#text"},{"title":"A.3 Symptom2Disease","offset":42039,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-3.html#text"},{"title":"A.4 LawBench","offset":42949,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-3.html#text"},{"title":"A.5 Aegis2","offset":43904,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-3.html#text"},{"title":"Appendix B Prompts","offset":45101,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-3.html#text"},{"title":"Appendix C Learned Skills","offset":45124,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-3.html#text"},{"title":"Appendix D Case Studies and Analysis","offset":45864,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-3.html#text"},{"title":"D.1 FiNER","offset":46081,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-3.html#text"},{"title":"D.1.1 Task Characteristics and Baseline Limitations","offset":46096,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-3.html#text"},{"title":"D.1.2 Skill Evolution","offset":47187,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-3.html#text"},{"title":"D.1.3 Context Evolution","offset":47630,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-3.html#text"},{"title":"D.2 USPTO50k","offset":48390,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-4.html#text"},{"title":"D.2.1 Task Characteristics and Baseline Limitations","offset":48408,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-4.html#text"},{"title":"D.2.2 Skill Evolution","offset":49086,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-4.html#text"},{"title":"D.2.3 Context Evolution","offset":49488,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-4.html#text"},{"title":"D.3 Symptom2Disease","offset":50785,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-4.html#text"},{"title":"D.3.1 Task Characteristics and Baseline Limitations","offset":50810,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-4.html#text"},{"title":"D.3.2 Skill Evolution","offset":52647,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-4.html#text"},{"title":"D.3.3 Context Evolution","offset":53418,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-4.html#text"},{"title":"D.4 LawBench","offset":55127,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-4.html#text"},{"title":"D.4.1 Task Characteristics and Baseline Limitations","offset":55145,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-4.html#text"},{"title":"D.4.2 Skill Evolution","offset":57594,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-4.html#text"},{"title":"D.4.3 Context Evolution","offset":58470,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-4.html#text"},{"title":"D.5 Aegis2.0","offset":59438,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-4.html#text"},{"title":"D.5.1 Task Characteristics and Baseline Limitations","offset":59456,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-4.html#text"},{"title":"D.5.2 Skill Evolution","offset":60040,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-4.html#text"},{"title":"D.5.3 Context Evolution","offset":60588,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-4.html#text"},{"title":"Appendix E Utility Functions","offset":61431,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-4.html#text"},{"title":"Instructions for reporting errors","offset":61609,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/part-4.html#text"}],"text_code_points":62573,"source_markdown":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/bedaeecadf58846f4728c50c03396fc96ba18429ecf14052b4c5e177394a246e/source.md"}