Skip to content

Security: fernandogarzaaa/experience-validation-engine

Security

SECURITY.md

Security policy

Reporting a vulnerability

Report privately through GitHub security advisories. Please do not open a public issue for a vulnerability.

Include what you would need to reproduce it yourself: version, adapter, and a minimal case. You should get an initial response within a week.

Supported versions

The latest published minor release receives security fixes. EVE is pre-1.0, so older minors are not patched — upgrade first, then report if the issue persists.

Scope notes

A few things about EVE's design are worth knowing before reporting:

  • EVE drives a real browser against whatever URL you point it at. It is a testing tool, and it executes the target application's JavaScript in a normal browser context. Point it at applications you trust, or run it in the container image where that context is isolated.
  • The perception script runs inside the target page. It reads rendered content and returns it. Treat any application EVE inspects as capable of influencing what appears in a report.
  • Reports can contain whatever was on screen. If a persona types into a form, or the app displays real data, it lands in report.md, the JSON output, and any captured screenshots. Do not publish reports from runs against production data without reading them first.
  • The MCP server speaks stdio and executes locally. It exposes EVE's capabilities — including navigating to URLs — to whatever agent it is wired into. Configure it the way you would any local tool server.

Automated scanning

CI runs CodeQL on every push and pull request plus weekly, and fails the build on high or critical advisories in production dependencies. Dev-tooling advisories are reported without blocking, since they gate on upstream release schedules and do not ship to users. Dependabot opens weekly dependency PRs.

There aren't any published security advisories