{"repo":"metaevo-ai/meta-context-engineering","free":true,"listed":false,"github":"https://github.com/metaevo-ai/meta-context-engineering","clone":"git clone https://github.com/metaevo-ai/meta-context-engineering.git","description":"[ICML 2026] Meta Context Engineering via Agentic Skill Evolution","language":"Python","stars":157,"topics":["agent-skills","agents","claude","context-engineering","evolutionary-computation","large-language-models","skills"],"license":"MIT","category":"ai-agents","readme_excerpt":"[ICML 2026] Meta Context Engineering via Agentic Skill Evolution Superseding Static Harnesses with Learnable Skills for Context Optimization This repository accompanies the paper Meta Context Engineering via Agentic Skill Evolution. Meta Context Engineering (MCE) is a bi-level agentic framework that co-evolves context engineering skills and context artifacts , replacing rigid CE heuristics with learnable skills that automatically discover optimal context representations and optimization procedures. Key Results MCE achieves consistent improvements across five diverse domains (finance, chemistry, medicine, law, AI safety): Setting Metric MCE Best Baseline Improvement --------- -------- ----- --------------- ------------- Offline Avg. Relative Gain vs Base 89.1% 70.7% (ACE) +18.4% Online Avg. Relative Gain vs Base 74.1% 41.1% (ACE) +33.0% Efficiency gains: - 13.6× faster training than ACE - 4.8× fewer rollouts required - Dynamic context length : 1.5K to 86K tokens based on task needs Reproduce experiments: See mce-artifact for code and data used in our paper. Why MCE? Current context engineering methods are fundamentally limited by manually crafted harnesses , for example: - Prompt rewriting (GEPA) favors brevity → fails on tasks requiring detailed knowledge - Additive curation (ACE) favors verbosity, structuring context as rigid itemized lists → causes context bloat and lacks structural expressiveness - Manually crafted agentic harnesses restrict optimization to narrow, intuiti","default_branch":null,"files":null,"tree":[],"storefront":"/r/metaevo-ai","claimed":false,"request_supported":{"post":"https://gitbuyer.com/r/metaevo-ai/meta-context-engineering/request-supported","requests":0},"note":"indexed from public GitHub; nothing is for sale on this page. Clone it from GitHub. Paid listings live at /search."}