ego-browser
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPRIVILEGE_ESCALATION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The macOS installation script (
scripts/install.sh) downloads theego liteapplication DMG from the vendor's CDN athttps://cdn.ego.app. This is a necessary component for the skill's functionality. - [COMMAND_EXECUTION]: The skill relies on the
Bashtool to executeego-browser nodejscommands. These commands utilize heredocs to pass JavaScript directly to the browser automation runtime, enabling arbitrary browser control within the Chromium environment. - [PRIVILEGE_ESCALATION]: The installation process for macOS (
scripts/install.sh) usessudoto perform administrative tasks, including moving the application to the/Applicationsfolder, removing Gatekeeper quarantine attributes (com.apple.quarantine), and executing system package installers. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to scrape and interact with arbitrary web content, which exposes the agent to instructions embedded in third-party websites.
- Ingestion points: Web data enters the agent context through
page.snapshot(), accessibility tree reads, and targeted DOM extraction in tools likelearnings/google/tools/search-extract.jsandlearnings/x-com/tools/timeline.js. - Boundary markers: No explicit boundary markers or instructions to ignore embedded commands are present in the provided extraction scripts.
- Capability inventory: The skill possesses the ability to execute shell commands via the
Bashtool, perform network requests viafetch, and write to the filesystem throughscreenshotandscreencastfeatures. - Sanitization: The provided scripts extract raw text (e.g., via
innerText) from the DOM and return it to the agent without filtering or sanitizing potential prompt injection vectors.
Audit Metadata