oracle
Fail
Audited by Gen Agent Trust Hub on Mar 15, 2026
Risk Level: HIGHDATA_EXFILTRATIONREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONCREDENTIALS_UNSAFE
Full Analysis
- [DATA_EXFILTRATION]: The script
scripts/refresh_cookies.pydirectly accesses and decrypts the Chrome browser's cookie database (~/Library/Application Support/Google/Chrome/*/Cookies). It specifically targets cookies related to 'chatgpt' and stores them in a local file~/.oracle/cookies.json. - [CREDENTIALS_UNSAFE]: The script uses
security find-generic-passwordto programmatically retrieve the 'Chrome Safe Storage' key from the macOS Keychain. While this requires a user prompt, it facilitates the decryption of sensitive session tokens. - [REMOTE_CODE_EXECUTION]: The script
scripts/run_oracle.pyusesnpx -y @steipete/oracleto download and execute an external Node.js package at runtime. This allows for the execution of unverified remote code with the same privileges as the agent. - [COMMAND_EXECUTION]: The skill uses
subprocess.check_outputandsubprocess.callto execute system commands (security,npx) and manage external processes.
Recommendations
- AI detected serious security threats
Audit Metadata