adversarial-audit

Pass

Audited by Gen Agent Trust Hub on Apr 16, 2026

Risk Level: SAFEDATA_EXFILTRATIONPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection attack surface because it processes untrusted data (local code and external web content) while possessing significant capabilities. \n
  • Ingestion points: Local codebase files (via exploration agents) and external web application content (via browser verification agent). \n
  • Boundary markers: Absent. The agent is not instructed to disregard instructions embedded in the data it audits. \n
  • Capability inventory: Browser automation (tabs_create_mcp, tabs_context_mcp), task management (TaskCreate, TaskUpdate), and file system writes (writing reports). \n
  • Sanitization: No sanitization or validation of the ingested code or HTML content is performed.\n- [DATA_EXFILTRATION]: The skill's verification phase involves navigating to external, user-provided URLs using a browser. This constitutes a network operation to non-whitelisted domains. While intended for legitimate testing, this capability could be abused to exfiltrate local data if the agent's logic is manipulated.\n- [COMMAND_EXECUTION]: The skill utilizes browser automation and task management tools to perform its auditing tasks, which involve interacting with the execution environment and remote web services.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 16, 2026, 01:41 AM