agent-browser

Pass

Audited by Gen Agent Trust Hub on May 14, 2026

Risk Level: SAFE
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill recommends installing the agent-browser CLI from NPM and fetches additional skill definitions using the tool. These resources originate from a well-known service and are central to the skill purpose.
  • [COMMAND_EXECUTION]: Shell commands are used to verify the environment, install the CLI tool, and launch Google Chrome with specific debugging flags. This is standard for browser automation tasks.
  • [PROMPT_INJECTION]: The skill is exposed to indirect prompt injection through browser automation. Evidence Chain: 1. Ingestion points: Page title, URL, and HTML snapshot (SKILL.md). 2. Boundary markers: Explicit Safety Rules section with instructions to ignore embedded commands. 3. Capability inventory: Browser automation (navigation, clicking, typing) and file_write (SKILL.md). 4. Sanitization: Explicit instructions to treat page content as untrusted evidence rather than instructions.
  • [DATA_EXFILTRATION]: Network activity is localized to communication with the Chrome DevTools Protocol (CDP) on 127.0.0.1. No suspicious external data transfers were identified.
Audit Metadata
Risk Level
SAFE
Analyzed
May 14, 2026, 08:55 PM
Security Audit — agent-trust-hub — agent-browser