#062Guide3 min
Audit an AI-built app before going live
Organize separate checks for deployed pages, source code and database permissions.
Hackaiz team · updated 21/11/2026

01Map the app you are about to release
Before publishing an AI-built application, write down what you will actually deliver: the live domain, the GitHub repository, the database and any agent components. A working preview proves that a user journey can succeed; it does not establish that private records stay private or that a secret was kept out of the browser bundle. Use this inventory to choose distinct inspection targets.
Hackaiz supports repository, live website, Supabase, PostgreSQL and MySQL analysis paths. They answer different questions. Repository checks can examine secrets, dependencies and authentication-related code; a live-site check sees deployed behavior; a database check examines accounts and permissions. Connecting one of these does not silently connect or cover the others.
02Authorize the deployed domain first
Use /connect, whose English heading is “Scan your platform”, to prepare the target. A website or API must have verified domain ownership before analysis. The supported proofs are a DNS TXT record named _hackaiz, an HTTPS file at /.well-known/hackaiz-verification.txt, or an instruction to the project's AI to create that file. Use the exact value and domain shown in the verification panel.
Treat the preview domain and the production domain as separate deployment details. A successful check of one address is not a statement about another environment with different configuration. If the file proof cannot be published at the required address, arrange the DNS proof with the person responsible for the domain rather than trying to bypass authorization.
03Select source and data access deliberately
Connect the GitHub repository that contains the release candidate, not an unrelated prototype. For a direct PostgreSQL or MySQL connection, the database form asks you to confirm ownership and recommends a dedicated read-only account. The live database inspection asks about accounts, rights and server settings; it does not read application table contents or write to them.
Choose checks according to the release risks you identified. For example, a customer portal needs particular attention to record access, while an app with an assistant also needs scrutiny of its instructions and tools. Selected analyses use credits. Unticked checks are not run or billed, so leaving database checks off means you must not describe that database as checked.
04Make findings part of the release decision
Read Danger, To fix and To watch as prompts for investigation, not a replacement for understanding the affected feature. A leaked credential needs different handling from a missing browser header. Identify the affected component, the evidence and the potential impact on the data or actions available in your app. Record any checks that were missing, incomplete or unable to run alongside the findings.
For a suspected access problem, have the developer explain which server-side rule separates users and how that rule was tested. A hidden button in the interface is not the same thing as a server refusing an unauthorized request. Do not interpret a report with no findings as certification or as proof that every possible business workflow was tested.
05Review corrections before shipping
A Hackaiz GitHub fix is a draft pull request that a human reviews and approves; it is never automatically merged. Compare the proposed change with the intended behavior, run your project's tests and check that legitimate users can still complete important tasks. Release readiness includes both restricting unauthorized access and preserving the service you meant to deliver.
After the approved changes reach the intended environment, run the relevant verification again and note the revision and deployment concerned. Rechecking uses credits, so choose the scope that answers whether the original issue is addressed. Optional monitoring concerns connected systems only; it is follow-up after connection, not an automatic watch over every application an AI has created for you.
Frequently asked questions
Take the next step
Was this article helpful?