Why I'm building Check My App

I want a coding agent to take a ticket, build the requested feature, deploy it, and receive an independent check of the result. If the implementation doesn't meet the request, the findings should return directly to the coding agent for another attempt.

Check My App is the part I'm building to provide that check.

The builder loop

The workflow I want is:

Ticket → coding agent → deployment → Check My App
→ findings → coding agent → new deployment → another check

The checker gets the original ticket and the deployed application. It checks the behavior requested in that ticket, rather than auditing everything it can find. The original requirements remain the reference across attempts.

I want a separate checker because it approaches the result from a different position: it uses the application instead of reasoning from the code it just wrote. Its job is to report what it observed, with evidence the builder can act on. A second agent isn't automatically right; it needs to demonstrate the mismatch.

How this differs from self-healing

My self-healing loop repairs reported failures in existing behavior. The builder loop checks whether newly requested behavior has been implemented. Both need feedback that reaches the next agent without a mandatory handoff to me.

Check My App already performs browser checks through MCP. The ticket-specific builder loop described here is the integration I want to complete. I choose what to build; I want the implementation and checking to continue without my approval between attempts.