Agentic Publication Protocol

Abstract

Scientific publication is still organized primarily around static manuscripts, even though much of scientific progress depends on tacit know-how: how to run code, reproduce figures, interpret edge cases, choose useful follow-up directions, and avoid failed paths. Large language model agents create an opportunity to publish not only knowledge, but also operational know-how in a form that future readers and researchers can directly use. This paper outlines the Agentic Publication Protocol (APP), a lightweight repository format for packaging a paper together with code, data, environment information, reproducibility instructions, and an agent-facing instruction file. APP treats a version-controlled repository as the publication object and uses AGENTS.md and optional skills to define a paper agent that can explain the work, reproduce key results when possible, and support follow-up research. We describe the design principles and details of the protocol, the agent skills useful for publishing papers under it, and development tools for evaluating and improving the protocol and associated skills. We close with a broader discussion of the future of scientific research in the agent era.

The Protocol

An APP publication is a version-controlled repository, fixed by a release tag, that bundles the paper with its code, data, environment, and an agent-facing AGENTS.md. A clear split keeps the authoritative ground truth separate from optional auxiliary material.

Structure of an APP publication repository — **Figure 1.** Structure of an APP publication: the paper and human-facing documentation, agent-facing instructions, and the optional code, data, environment, supplementary material, and skills that make the work usable as an agent-readable publication object.

Publishing Workflow

Optional skills guide an author from a working paper repository to a validated, tagged release. The publish-paper metaskill decomposes the process into focused steps: reproduce results, prepare staging, define the paper agent, validate, and release.

Overall APP publication workflow — **Figure 2.** The publication workflow supported by APP skills. *Top:* from a working paper repository to a validated APP release. *Bottom:* the `publish-paper` metaskill and its step-specific skills.

The publish-paper metaskill and its step skills — **Figure 2.** The publication workflow supported by APP skills. *Top:* from a working paper repository to a validated APP release. *Bottom:* the `publish-paper` metaskill and its step-specific skills.

Evaluation

In the controlled compare-app benchmark, a neutral evaluator scores an APP paper agent against a general repository-aware agent on the same reader questions. On the eleven public papers the APP agent wins every paper, averaging 9.25 versus 8.50 overall, with its largest margins on grounding and honesty.

Average compare-app aspect scores over 11 public papers — **Figure 3.** Average `compare-app` aspect scores over the eleven public papers, for accuracy, informativeness, grounding, and honesty (neutral evaluator: Codex, `gpt-5.5`, reasoning effort `xhigh`).

Outlook

Beyond a single paper, agentic publication may change the shape of the research network itself: paper agents that not only explain finished work, but also propose and pursue follow-up directions through interaction.

Schematic of the research network under agentification — **Figure 4.** A schematic of how agentification may change the research network. Each node is a research work and directed links point forward in time.

Talk to This Paper

This paper, which proposes APP, is also an APP publication itself. It ships with an AI agent. Clone the repository and open it in an AI coding agent: it loads AGENTS.md, explains the work, reproduces key results when possible, and helps you build on it — using the included paper, code, and data as ground truth.

git clone https://github.com/XiaoliangQi/agentic-publication-protocol-dev.app
cd agentic-publication-protocol-dev.app

Contributions

A lightweight repository-and-release format that publishes a paper together with its code, data, environment, reproduction instructions, and an agent-facing AGENTS.md.
A clean separation between the stable publication object — the repository and its tagged release — and the optional skills used to create, validate, and release it.
A controlled compare-app evaluation in which the APP paper agent outperforms a general repository-aware agent on all eleven public papers (mean 9.25 vs. 8.50), with its clearest gains in grounding and honesty.

Citation

@article{lu2026agenticpublicationprotocol,
  title={Agentic Publication Protocol: An Attempt to Modernize Scientific Publication},
  author={Lu, Sirui and Qi, Xiao-Liang},
  year={2026},
  url={https://github.com/XiaoliangQi/agentic-publication-protocol-dev.app}
}