Lightpanda

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.90). The skill explicitly instructs agents to navigate and extract content from arbitrary external URLs (see SKILL.md: MCP tools like goto/markdown, CLI fetch --dump examples, and CDP page.goto usage), so untrusted third-party web pages can be read and used to drive interactions (click/fill/evaluate) and therefore could inject instructions indirectly.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The install script fetches and installs a remote binary at runtime from URLs like https://github.com/lightpanda-io/browser/releases/download/nightly/lightpanda-x86_64-linux (and the corresponding nightly releases) and also queries https://api.github.com/repos/lightpanda-io/browser/releases/tags/nightly for checksums, which downloads and then executes remote code as a required dependency.