Free tool · Runs in your browser
Paste the text that describes your agent — system prompt, configuration, tool manifest, or the code that wires it together — and the scan reviews it against the controls a security reviewer expects in a production deployment. Nothing is uploaded; the review is executed locally on this page.
0 characters · minimum 40
What this scan does and does not do
The scan is a static review of the text you provide. It detects the patterns and omissions that most often show up in AI deployments that fail a security review: credentials in configuration, an agent with shell access, no isolation boundary, an unbounded tool surface, no approval gate, no audit trail, and content that can be used to override instructions.
It does not run your code, connect to your systems, or probe a live endpoint, and a clean result is not a substitute for a penetration test. Treat it as the pre-flight check it is: fast, private, and useful for finding the expensive gaps before a customer or an auditor does.
