Agent-Native Primitives and HEELO Framework Improve Real-World Task Completion
arXiv — cs.MA metadata · Haibo Jin, Peng Kuang, Xiaopeng Yuan, Lucheng Fu, Jerry Wang, Xucheng Yu, Yaoning Yu, Dehao Wu, Runguo Li, Colin Su, Jingxuan Sun, Xuyang Liu, Haojing Luo, Haohan Wang · Published on Readzer
Summary
The paper introduces agent-native primitives and the HEELO framework to address the 'last mile' gap in real-world task execution. On Terminal-Bench 4.0, HEELO with Dev-Primitives raises resolution from 37.3% (GPT-5.6 Sol with Codex) to 55.5% on the same backbone. On 46 real-world tasks, Claude Opus 5.5 with Claude Code and GPT-6.1 Sol with Codex each complete 9 tasks alone, but 39 and 35 respectively when connected to HEELO via MCP server, with 1.9x and 2.3x less time and 1.55x lower cost per completed task.
This is a reviewed summary. Read the complete article at the original source.
Why it matters by industry
- Artificial intelligence
For AI model developers, this research provides a reusable framework (HEELO) and primitive libraries that significantly boost agent performance on real-world benchmarks, offering a practical approach to close the gap between resource access and application.