Agentic Publication Protocol: An Attempt to Modernize Scientific Publication

A new proposal called the Agentic Publication Protocol (APP) aims to move scientific publishing beyond static PDFs by packaging papers with code, data, and executable instructions that an AI agent can use to explain findings and reproduce results [1]. The protocol, described in a paper submitted to arXiv on June 15, 2026, treats a version-controlled repository as the publication object itself [1]. It relies on an agent-facing instruction file, named AGENTS.md, and optional skills to define a "paper agent" capable of explaining the work, reproducing key results when feasible, and supporting follow-up research [2]. The authors argue that much of scientific progress depends on tacit know-how — running code, reproducing figures, interpreting edge cases, and avoiding failed paths — that static manuscripts cannot convey [2]. The APP arrives amid broader efforts to make computational research more interactive. In 2021, Hugging Face launched Spaces, a platform that has since hosted over 12,000 open-source machine learning demos built by the community [10]. arXiv later integrated Hugging Face Spaces through its arXivLabs framework, adding a Demo tab to paper abstract pages so readers can try models directly in a browser without writing code [10]. Authors can link a Space to their paper by including the paper’s URL in the Space README file or by associating the Space with a model already linked to the paper on the Hugging Face Hub [11]. The APP extends this interactivity further by embedding an agent that can actively assist a reader. The paper outlines design principles for the protocol, agent skills useful for publishing under it, and development tools for evaluating and improving both the protocol and the associated skills [2]. It also includes a broader discussion of how scientific research may change in what the authors call the agent era [2]. A separate scoping review of generative systems for quantum circuit and code generation, published on arXiv in early 2026, found that while all reviewed systems addressed syntactic validity and most addressed semantics, none reported end-to-end evaluation on quantum hardware [9]. That gap between generated artifacts and practical deployment highlights the kind of reproducibility challenge the APP is designed to address by bundling code, data, and environment specifications directly with a paper [1][9]. The APP paper was submitted to the arXiv digital libraries category and is available as an HTML experimental version alongside the traditional PDF [1]. The authors have not yet released a public implementation of the protocol’s development tools.