This repository serves as a comprehensive survey and curated collection of academic papers on "Code as Agent Harness," a paradigm shift where code acts as an executable, verifiable, and stateful foundation for AI agents. It targets researchers and engineers seeking to understand and leverage code for agentic reasoning, action, and environment modeling. The primary benefit is a structured, organized overview of a rapidly advancing field, facilitating research and development in executable, stateful AI systems.

How It Works

The project organizes research papers into three interconnected layers: Harness Interface, Harness Mechanisms, and Scaling the Harness. This structure categorizes how code functions as the fundamental interface for agents (reasoning, acting, environment modeling), the internal processes that manage code execution (planning, memory, tool use, debugging), and strategies for deploying code-centric systems at scale, particularly in multi-agent scenarios. This layered approach provides a systematic framework for understanding the evolution and application of code-centric AI agents.

Highlighted Details

• Accompanies the survey paper "Code as Agent Harness: Toward Executable, Verifiable, and Stateful Agent Systems."
• Organizes papers across three core layers: Harness Interface, Harness Mechanisms, and Scaling the Harness.
• Covers diverse application domains including coding assistants, GUI/OS automation, scientific discovery, and embodied intelligence.
• Actively welcomes contributions and updates from the research community.

Maintenance & Community

The repository is actively maintained and welcomes paper suggestions, pull requests, and collaborations. Contact information for the authors is provided (xuyingn2@illinois.edu, kt42@illinois.edu, twei10@illinois.edu, zihaoli5@illinois.edu, bei4@illinois.edu). The project is committed to ongoing updates with recent work in code-centric agentic systems.

Licensing & Compatibility

This repository is released under the MIT License. This permissive license allows for broad use, modification, and distribution, including for commercial purposes, with minimal restrictions.

Limitations & Caveats

As a curated list of research papers, this repository does not provide executable code or a development framework. Its value lies in its comprehensive organization and indexing of existing academic work in the "Code as Agent Harness" domain.