Please do not open a public issue for a security problem. Report it privately, either:

• through GitHub's private vulnerability reporting, or
• by email to security@stuntdouble.io.

Include the version, what an attacker can do, and steps or code to reproduce it. We will acknowledge your report within three working days, keep you updated as we investigate, and credit you in the advisory unless you would rather we did not.

Fixes land in the latest published release. While stuntdouble is below 1.0, only the latest release gets fixes.

In scope: the published stuntdouble package, including:

• Studio and the viewer: reaching them from anything but their own 127.0.0.1 address and key, a write without Studio's own Origin, or a page value rendered as anything but text.
• Keys and tokens in ~/.stuntdouble/credentials.json reaching the page, a log, the project or the model.
• The browser reaching a local or private address other than the app under test, or a page getting past the third-party guard (payments, sign-ups, passwords, card numbers) when safety is third_party.
• The MCP server calling a model, starting an agent or reading an agent's credentials.

The browser guards come from @stuntkit/browser-toolset and read the DOM the page shows. They are a strong default rather than a sandbox, so a report is most useful when it shows a realistic page getting past them. A bypass in the guards themselves can also go to StuntKit.