browser
Pass
Audited by Gen Agent Trust Hub on Oct 7, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [COMMAND_EXECUTION]: The
scripts/start.cjsscript useschild_process.spawnto launch the Chrome browser process based on the operating system. - Evidence:
const chrome = spawn(chromePath, args, { detached: true, stdio: 'ignore' });inscripts/start.cjs. - [INDIRECT_PROMPT_INJECTION]: The skill navigates to external websites and processes their content (DOM elements, text, evaluation results), creating a surface for instructions embedded in web pages to influence agent behavior.
- Ingestion points:
nav.cjs(navigation to external URLs),eval.cjs(executing JS on pages), andpick.cjs(inspecting DOM elements). - Boundary markers: None identified; the agent is not explicitly instructed to ignore or sanitize content found on processed web pages.
- Capability inventory: The agent can execute arbitrary JavaScript in the browser (
eval.cjs), navigate to new URLs (nav.cjs), and save screenshots to the local file system (screenshot.cjs). - Sanitization: No sanitization or filtering of retrieved web content is performed before returning it to the agent context.
- [DYNAMIC_EXECUTION]: The
scripts/eval.cjsscript accepts arbitrary JavaScript code as a command-line argument and executes it within the browser context using theRuntime.evaluatemethod of the Chrome DevTools Protocol. - Evidence:
ws.send(JSON.stringify({ id: 1, method: 'Runtime.evaluate', params: { expression: code, ... } }));inscripts/eval.cjs. - [DATA_EXPOSURE]: The skill manages sensitive browser data and screenshots locally.
- Persistent Profiles:
scripts/start.cjssupports a--profileflag that stores cookies, authentication tokens, and browser history in~/.chrome-debug-profile, which could be accessed by subsequent browser sessions. - Temporary Files:
scripts/screenshot.cjswrites browser screenshots to the system's temporary directory usingos.tmpdir().
Audit Metadata