What is codeasagentharnessth?

This repository is a Thai-language translation of an academic research paper titled "Code as Agent Harness: Toward Executable, Verifiable, and Stateful Agent Systems," originally published on arXiv in May 2026. The translation covers pages 1 through 66 of the original PDF, including the full text, figures, and tables. It is version 1.0.0 of the translation. The paper itself is about a concept in AI research: the idea that instead of giving an AI agent a list of instructions in plain language, you give it actual computer code as its control structure. This approach aims to make AI agents more reliable, easier to verify, and better at maintaining context across multiple steps. In other words, the code becomes the blueprint that tells the agent what to do and how to track what it has done. The translation was produced by a contributor named Tuengdin using Google Gemini as a translation aid, followed by manual review and formatting. Technical terms, proper names, acronyms, citations, URLs, and code identifiers are kept in English even within the Thai text. The translation is described as unofficial and for educational and knowledge-sharing purposes only, not an authorized version endorsed by the original authors or Google. If you find translation errors or technical terminology issues, the repository invites you to open a GitHub Issue to suggest corrections. The translated content is released under the Creative Commons Attribution 4.0 International license, while Tuengdin retains copyright over the human-created portions of the translation under applicable law.