Forge Dialog is pre-1.0. Fixes land on the most recent minor line; older 0.x lines are not
patched.
| Version | Supported |
|---|---|
| 0.7.x | Yes |
| < 0.7 | No |
Please report privately rather than opening a public issue.
Use GitHub's private vulnerability reporting: go to the repository's Security tab and choose Report a vulnerability. That opens a draft advisory visible only to you and the maintainers.
Helpful things to include: the affected version and entry point, a minimal reproduction, the impact you believe it has, and any suggested fix.
Expect an acknowledgement within a few days. You will be kept updated while a fix is prepared, and credited in the advisory unless you prefer otherwise.
Forge Dialog renders UI in the browser and ships no server component, so most reports will concern how untrusted content reaches the DOM.
In scope
- Content passed through documented APIs escaping into the DOM as markup when it should be text.
- Anything letting a dialog break out of its accessibility or focus containment in a way that enables clickjacking or input capture.
- Supply-chain issues in what the package publishes: unexpected files in the tarball, or a build artifact that does not match this source.
Out of scope
unsafeHtml. It is documented as trusted-input-only and performs no sanitization by design; passing user input to it is a bug in the calling application. Usehtmltogether withsanitizeHtml, which refuses to render until a sanitizer is supplied.- Vulnerabilities in an application's own sanitizer implementation.
- Findings that require an attacker to already execute script on the page.
contentandmessagestrings are rendered as text, never parsed as markup.npm run test:cspfails the build ifevalornew Functionappears indist/, so the package stays usable under a strict Content Security Policy.- Releases publish with npm provenance and attach a CycloneDX SBOM.
- Development dependencies are kept free of known advisories;
npm auditis expected to be clean.