dynamic-content-extraction

Pass

Audited by Gen Agent Trust Hub on Aug 14, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONDATA_EXFILTRATIONREMOTE_CODE_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill contains meta-instructions that attempt to influence the agent's internal tool-selection logic by prioritizing runtime tool catalog checks over static notes and referencing a 'vault' for web tooling. It is also susceptible to indirect prompt injection: 1. Ingestion points: Extracts data from external websites such as Booking.com, Marriott, and Amazon (SKILL.md). 2. Boundary markers: Absent. 3. Capability inventory: Possesses browser automation tools capable of network requests (fetch), script execution (browser_console), and low-level browser control (browser_cdp). 4. Sanitization: Absent.
  • [COMMAND_EXECUTION]: The skill relies on executing arbitrary JavaScript snippets through browser_console and browser_cdp to interact with the DOM and perform extraction tasks.
  • [DATA_EXFILTRATION]: Provides instructions to access and potentially exfiltrate sensitive session data using Network.getAllCookies and to reproduce internal API requests using fetch() within the browser console environment.
  • [REMOTE_CODE_EXECUTION]: The workflow relies on dynamic code generation and execution, providing specific JavaScript snippets for runtime injection via Runtime.evaluate to handle complex web framework rendering.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 14, 2026, 07:56 AM
Security Audit — agent-trust-hub — dynamic-content-extraction