Auto-fix and build repair

Sometimes a change introduces an error that stops the app from building or running. When that happens, Vaulter tries to fix it for you before you ever have to deal with it.

Three layers of self-repair

Vaulter watches for problems in three different ways, and each one repairs itself:

• Build repair. After the agent generates code, Vaulter validates it. If validation finds a real problem, a repair pass runs in the background and fixes it before the preview reveals. You never see the error, just a working app.
• Runtime error repair. If your running preview throws an error, an unhandled failure, or a crash in the browser, Vaulter catches it, gathers what it caught, and sends the agent a targeted fix request. Repair attempts are measured and spaced out, so it never thrashes or burns your tokens chasing the same error.
• Blank-screen repair. After every preview loads, Vaulter checks that the app actually rendered. If the screen came up blank or broken, that on its own triggers a repair, once per build.

This also applies when you open a project that was left in a broken state: Vaulter notices on load and tries to get it running again, so you can pick up where you left off.

What you will see

• Most of the time, nothing at all. The build works, or Vaulter quietly fixes it, and the preview updates.
• While a repair runs, a soft loading cover sits over the preview. It drops on its own when a rebuild succeeds.
• If the repair budget runs out, the cover becomes an explicit "Preview failed to build" message with a Retry button.

When auto-fix cannot resolve it

If a problem persists past the automatic attempts, describe in chat what you were trying to do and what went wrong. The agent can read the build logs and your app's health directly, so it usually finds the cause quickly with a little direction. If you are truly stuck, see Troubleshooting and support.

Frequently asked

Does auto-fix use my tokens?

No, never. Automatic repairs are on us, always, even when they involve real AI work. Your allowance is only spent on the changes you ask for. See Tokens and usage.

Why do I see short repair messages in chat?

So the history is honest: when a repair runs, it appears as a brief entry rather than happening invisibly. You can see what was fixed without wading through detail.

Auto-fix keeps trying but the error comes back

Repair attempts are deliberately spaced and budgeted, so a stubborn error eventually needs you: tell the agent what you were doing when it first broke, and include what the screen shows. The reproduction is usually the missing piece.

Does auto-fix change things I did not ask for?

It repairs the specific breakage, aiming for the smallest fix that makes the app run again. It does not redesign or refactor; it gets you back to working.