Publish a paper

The working repo is the folder that contains the materials you want to publish. It may also contain raw materials, notes, or other context that helps explain the paper.

You need an AI coding agent that can read APP skills and operate in your working repo. To complete publication, the required Git and GitHub operations must also be completed, such as creating commits, tags, releases, and release assets. You can either allow the agent to perform these operations or run them manually when the skill instructs you to do so. If you want the agent to perform them, you may need to set up GitHub authentication.

Open your working repo in an AI coding agent with this plugin installed, then invoke the publish-paper skill:

# Claude Code
/publish-paper

# Codex
$publish-paper

The publish-paper skill walks through the full workflow in order. It can span multiple sessions, and it may pause for author decisions between steps.

You can also call each step directly if you only need part of the workflow, want to resume from a known checkpoint, or are debugging a publication candidate. The step skills should be used in this order:

1. reproduce-results — reproduce/check existing paper results before staging, including figures, tables, experiments, and analytic derivations.
2. prepare-staging — organize the author-approved paper, code, data, environment notes, reproduction reports, and supplementary materials into publication-staging/.
3. define-paper-agent — draft and revise AGENTS.md, CLAUDE.md, and reader-facing documentation for the staged paper agent.
4. validate-publication — check APP structure, paths, privacy, consistency, reproduction status, and reader-agent usability.
5. release-outcome — perform the final review/freeze and either publish the validated release or record a developer-sandbox outcome.

publish-paper is still the recommended entry point because it keeps these steps coordinated and asks for author approval at the right checkpoints.