testing
Fail
Audited by Gen Agent Trust Hub on Feb 16, 2026
Risk Level: CRITICALREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- REMOTE_CODE_EXECUTION (CRITICAL): The file
scripts/with-server.pyusessubprocess.Popen(shell=True)to run commands provided via the--serverargument. This allows an attacker to execute arbitrary shell commands if they can influence the agent's input. - PROMPT_INJECTION (HIGH): The skill uses Playwright to capture data from potentially untrusted web pages in
examples/console-logging.py. This data is then processed by the agent, creating a vector for indirect prompt injection where a website could control the agent's actions. - DATA_EXFILTRATION (MEDIUM): The use of
file://URLs inexamples/static-html-automation.pyenables the browser to access the local filesystem. If the path is manipulated, it could be used to expose sensitive system files to the agent or external outputs.
Recommendations
- AI detected serious security threats
Audit Metadata