Set up an AI Vibe coding agent orchestration system (Claude Code + Cursor, with automated QA gate)

I want someone to build me a working AI development pipeline and then hand me the keys. I am not hiring you to write my application code - the AI agents do that. I am hiring you to wire up the harness that runs them, prove it works, and document it so I operate it alone. WHAT I WANT WHEN YOU ARE DONE I file a ticket describing what I want in ordinary language. Everything after that is automatic: 1. The ticket routes to the right coding agent - Claude Code for my Laravel/MySQL backend, Cursor for my Flutter Android app 2. When a ticket needs both, the two agents hand off to each other and stay coordinated on the same ticket. No copying and pasting between tools by me. 3. The result gets built into a real APK and run through automated screen tests 4. A separate reviewing agent inspects the code changes, logs, and test screenshots and returns a pass or fail 5. I get exactly one notification when all of that agrees the work is ready, and I approve or reject 6. I hear from you or the system otherwise only if a genuine clarification is needed or the automation itself is broken That is the whole ask. One request in, one approval request out. MY SITUATION Two existing repositories (Flutter app, Laravel backend) plus a control repository holding about 160 open tickets, already organized. I have a substantial GitHub Actions implementation of the above that gets close but is not reliable, and I have transcripts and documentation of exactly where it fails. I am not attached to that implementation. If you think a commercial platform gets me there faster and more reliably, say so and set that up instead. Budget and vendor choice are open. I care about the outcome, not the route. I do have hard requirements on limits: bounded retry attempts, per-job timeouts, and account-level spending caps. I have had a runaway agent loop burn resources overnight and I will not accept a design that can do that again. SCOPE AND ACCEPTANCE This is a one-time setup engagement with a defined finish line, not an ongoing development contract. Milestone 1 - Assessment. Review my existing package, transcripts, and repositories. Tell me what you would keep, replace, or buy. Written recommendation. Milestone 2 - Build. Stand up the pipeline on whatever you recommended. Milestone 3 - Pilot and acceptance. Five real tickets run one at a time, including one that should fail the automated review and one written deliberately vaguely. The pilot passes only if I never have to repair the automation by hand. Milestone 4 - Handover. Written operating instructions, a one-page runbook for what to do when a run fails, and a walkthrough call. After this I run it myself. WHAT YOU NEED - Deep GitHub Actions experience, or deep experience with a comparable agent orchestration platform - Continuous integration and continuous deployment for Android: signing, build identity, artifact handling - Automated user-interface testing - Maestro preferred, Appium or similar acceptable - Enough Laravel and Flutter familiarity to read the repositories and run the existing tests. You are not writing features in them. - Real hands-on experience driving AI coding agents (Claude Code, Cursor, Codex, or similar) from automation rather than from a chat window WHAT THIS IS NOT Not application feature work. Not a build-from-scratch product. Not a long-term retainer. Please skip this if your portfolio is mostly websites and new product builds. HOW WE WILL WORK All access read-only to start. Credentials rotated before and after. The pipeline runs against synthetic test data only - no live client records. One ticket at a time until it proves stable. My time zone is US Central.