agent-browser

Warn

Audited by Gen Agent Trust Hub on Jul 8, 2026

Risk Level: MEDIUMREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill documentation explicitly describes the capability to inject and execute custom JavaScript within the browser's console to extract data or simulate user behavior.
  • [COMMAND_EXECUTION]: The installation process involves executing code from a remote registry via the shell command npx clawhub@latest install agent-browser.
  • [DATA_EXFILTRATION]: The skill can monitor and intercept network requests (XHR/Fetch), access browser session data (cookies, localStorage), and export screenshots, videos, or PDFs, which provides a path for harvesting sensitive information.
  • [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it processes untrusted data from websites while possessing high-privilege capabilities:
  • Ingestion points: Web content accessed via tools such as navigate, snapshot, and network request monitoring described in SKILL.md.
  • Boundary markers: There are no explicit delimiters or instructions provided to the agent to distinguish between its instructions and embedded commands within the processed web pages.
  • Capability inventory: Includes direct element interaction (clicking, typing), custom script execution (console injection), and session persistence.
  • Sanitization: There is no evidence of sanitization or validation of the ingested web content before the agent performs actions based on that data.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jul 8, 2026, 04:47 PM
Security Audit — agent-trust-hub — agent-browser