yokoten doctor

Live: https://doctor.yokoten.app Paste a public GitHub repo. yokoten doctor shows what an AI understands from it: how to build it, how to test it, and what not to touch. If that looks right, you save it to your repo as AGENTS.md, and the page checks that it's there. The repo's code never runs. Some of its files go to Muse Spark Open, on Baseten, to write that text. A repo that already has AGENTS.md gets no new text. The page just says so. After a check, you can also get a review report: the risks an AI agent should know before it changes your project, each linked to a line of code. About one to two minutes. The whole review run is saved, and anyone can check it with one command. Who it helps: anyone who lets an AI agent work in their repo, and open source maintainers who want agents to follow their rules. It is free, with no account, for public repos. Built during the event, in this repo: the page, the server, the Muse part, the public record and the review report (the server job, the agent that calls Muse, its record and the page section). Pre-existing: yokoten, its logo and its site style. yokoten is my open source tool. It is an npm dependency, and no yokoten code is copied here. It started before the event, and it kept changing during it, in its own repo. This page uses yokoten 0.2.0, released on 2026-10-30. That version holds two things made during the event that this page uses: `doctor --files`, and the `how` field in its JSON. This repo holds none of that work, and I do not submit it as this entry. cool-workflow, which plans the review report, is also my open source tool (BSD-2-Clause), made before the event. The page installs it from npm; no cool-workflow code is copied here.