mesa-ai-cto

Check your machine, then scaffold a production-grade project from a plan produced by The AI CTO.

doctor

npx mesa-ai-cto doctor

  Runtime
    ✓ Node.js           v22.18.0
    ✓ npm               10.9.3
  Git
    ✓ git               git version 2.50.0
    ✓ user.name         Gaurav Madan
    ✗ user.email        not set
  Editor
    ✓ VS Code           detected
    ✗ `code` on PATH    not on PATH
  GitHub
    ✓ GitHub CLI        github.com
  Vercel
    ✗ Vercel CLI        not installed

  Fix this before you start:
    git config --global user.email "[email protected]"

  Not needed yet, but you will want:
    In VS Code: Command Palette → "Shell Command: Install 'code' command in PATH"
    npm install -g vercel

  Not ready yet — see above.

Blocking problems are listed apart from ones you will only need later, because "one thing to fix" sitting above "ready to build" does not answer the only question this command exists to answer.

Three results, not two. A tool can be missing, installed but not working, or impossible to reach right now — a gh auth status that times out on hotel wifi has not told us your account is broken. Each leads to different advice, so each is reported differently, and a result we could not determine never fails the run.

For coding agents

doctor is designed to be run by Claude Code, Codex or any agent driving a founder's setup — so it does not just say what is wrong, it says what to do about it, on this operating system, and whether the agent may do it unattended.

npx mesa-ai-cto doctor --json

JSON goes to stdout and nothing else does, so it pipes cleanly. Human output goes to stderr. Exit codes: 0 ready, 1 a required check failed, 2 a usage error.

Read nextActions. It is the failures only, already ordered with blocking ones first, so an agent that repairs one item and re-runs has removed the reason the build could not start.

{
  "ok": false,
  "nextActions": [
    { "id": "git.installed", "blocking": true, "autoRunnable": true,
      "command": "winget install --id Git.Git -e --accept-package-agreements" },

    { "id": "git.email", "blocking": true, "autoRunnable": false, "needsInput": true,
      "commandTemplate": "git config --global user.email \"<user.email>\"" },

    { "id": "editor.code", "blocking": false, "autoRunnable": false,
      "manual": "Open VS Code, press the Command Palette shortcut, and run \"Shell Command: Install 'code' command in PATH\"" }
  ]
}

The loop:

1. Run doctor --json. If ok is true, stop.
2. For each entry in nextActions where autoRunnable is true — run command.
3. Where needsInput is true — ask the founder for the value, then run commandTemplate with the placeholder replaced. Do not invent one: a repository whose history says [email protected] is worse than one with no commits, and rewriting author history later is painful.
4. Where only manual is present — relay it and wait. There is no command; guessing one (symlinking code by hand, for instance) creates something the next update overwrites.
5. Re-run doctor --json. Repeat at most a few times, then report what is left.

Only act on entries where blocking is true if the goal is to get building. The rest are needed for deploying later and can wait.

Platform. Commands are chosen for the machine doctor ran on — brew on macOS, winget on Windows. On Linux, where package managers differ too much between distributions to guess safely, there is no command: the entry carries a url instead, and guessing apt would be wrong often enough to matter.

Coming next

npx mesa-ai-cto <name> --tier<N> --token=<t>   scaffold a project
npx mesa-ai-cto add <module>                    add a module guide

You get that command, with its token, from your plan in the AI CTO dashboard.

Licence

Proprietary. See LICENSE.

The guides, checklists and boilerplate are not contained in this package. They are delivered only to a holder of a valid install token.