Overskill has a built-in Security Review you can run on your app before you publish it. It asks the AI to audit your app for the most common security mistakes so you can fix them before real users (and their data) ever touch it. It uses credits, like any message you send the AI.
How to run it
- Open your app in the editor and click Publish.
- In the publish panel, find Security Review and click Run Scan.
- The publish panel closes and a security-audit request is sent to the AI in your chat. The AI reads through your app and replies with what it found, ranked by severity (Critical, High, Medium, Low).
- Review the findings and fix anything important. For most issues you can just describe the fix to the AI (
remove the hardcoded API key and load it from an environment variable
) and it'll handle it.
What it looks for
The review focuses on the things that actually bite real apps:
- Exposed secrets — API keys, tokens, or passwords accidentally left in your code.
- Unsafe HTML / cross-site scripting (XSS) — places where untrusted content could run as code in a visitor's browser.
- Authentication & token handling — whether sign-in tokens are stored and checked safely.
- Input validation — forms and data that aren't checked before they're used.
- Data-exposure risks — sensitive information that could leak through the page or logs.
A few things to know
- The review is on-demand — you run it when you want, as part of publishing. It isn't a continuous, always-on scanner.
- It's a strong first line of defense, not a guarantee. No automated review catches everything, so treat a clean result as
the common mistakes are handled,
notnothing can ever go wrong.
Keep good habits: never hardcode secrets, validate input, and don't expose data you wouldn't want public. - Publishing is up to you. A finding, even a critical one, doesn't stop Publish, so fix what matters before you go live.
The short version
Before you publish: open the publish panel, click Run Scan under Security Review, fix anything critical, then ship with confidence.
Does the review use credits?
Yes. Run Scan sends the audit request to the AI as a chat message, so it uses credits like any other message. Asking the AI to fix what it found is another message and uses credits too.
A clean review is not proof that your app is secure. Test its access rules with accounts that should and should not have access.