yokoten · 横展

How it works

What changes on day one

Your agent tries to…Without yokotenWith yokoten
force-push, or tag a releaseWithout yokotenit happensWith yokotenstopped before it runs, and told whycodetest
“fix” a failing test by editing the testWithout yokotenthe test passes, the bug staysWith yokotennot allowed in a fixcodetest
commit your email, a server IP, your cityWithout yokotenit goes publicWith yokotenthe commit stops; the report names the file and line, never the textcodetest
open a pull requestWithout yokotenas you, with all your rightsWith yokotenas a bot; nothing merges without youcodetest
work in another projectWithout yokotenstarts from zeroWith yokotensees what that project is missing, and offers to set it upcodetest

The road

Software teams work in a loop: plan, build, test, review, release. yokoten takes you along it one stage at a time. The yokoten repo does every stage itself, so you can see it working.

  1. doctor: stage 1 · safety net

    Safety netyou are here

    You don’t have to work out where you are. doctor reads your project’s files and says it, for example stage 1 of 5, safety net. A new project starts at stage 1. The order is advice, not a gate.

    You learn
    rules your agents read, and checks that stop them when rules aren’t enough
    See it here
    AGENTS.md

  2. doctor: stage 2 · issues and pull requests

    Issues and pull requests

    You learn
    every change starts as an issue and merges only when CI passes; labels, milestones, and a board keep track

  3. doctor: stage 3 · release

    Release

    You learn
    commit messages that say what changed and why, a changelog, a version number for every release
    See it here
    CHANGELOG.md

  4. doctor: stage 4 · a team of agents

    A team of agents

    You learn
    a bot opens pull requests, workers take tasks, a reviewer checks them; only you merge
    See it here
    the guide

  5. doctor: stage 5 · going public

    Going public

    You learn
    scan every past commit for private data, turn on security settings, protect main
    See it here
    security policy

What it can’t do

Honest limits, from the guide:

  • The guards read command text. A command put together in pieces gets past them. They catch mistakes and drift; they aren’t a lock. The lock is on the server: branch protection, code owners, and a bot that can’t release or change CI.
  • Server settings aren’t checked yet. doctor lists the three steps for the owner. Checking them through the API is planned.
  • Claude Code first. The guard also runs in Codex and Gemini CLI, but nobody has yet seen it block a real session there (#30). Muse Code support is planned.

The full list in the guide