The latest commit of little-fridge/viewcraft (c1996d9eb75d3f9b92b03e598ac7ee4faded5108, October 14, 2026) is a change that makes the paper's arXiv link and official citation information public. ViewCraft is a framework that refines both the field of view of multimodal agents and the procedure for reading it, and according to the README the repository contains 10 reusable task skills together with paper resources.
The preferred-citation entry in CITATION.cff has been filled in. The paper title is 'VIEWCRAFT: Evolving What Multimodal Agents See and How They Read It', and 10 authors are listed (Binxiao Xu, Ruichuan An, Bocheng Zou, Weihua Du, Xinyu Wei, Hangyu Guo, YiFan Zhang, Renye Yan, Chengqi Duan, Hang Hua), together with the venue arXiv, the date October 2026, and the DOI 10.48550/arXiv.2610.15915. The Paper badge in the README and the BibTeX example were also changed from the PDF inside the repository to arXiv:2610.15915, and primaryClass cs.CV was added to the BibTeX.
Changes on the website side were included as well. The project page is deployed from a separate public repository, the Paper link points to the arXiv address, and the wording '2026 PREPRINT · arXiv submission pending' has disappeared. The conspicuous point is the fact that the code repository and the task skill library have been made public. The previous document described them as non-public, but in this commit the description was changed to public. The size of the change is 5 files and 102 lines (70 added, 32 deleted).
However, this commit is the handling of the paper's appearance and the organization of the disclosure scope, not a performance evaluation. The site merely states that benchmark results are included, but no figures are contained in this commit, so the actual performance must be checked in the original arXiv:2610.15915 paper. The code is distributed under the MIT license, and external datasets retain their own respective conditions.