windows-credential-harvesting

Fail

Audited by Gen Agent Trust Hub on Jul 6, 2026

Risk Level: HIGHCREDENTIALS_UNSAFECOMMAND_EXECUTIONDATA_EXFILTRATIONREMOTE_CODE_EXECUTION
Full Analysis
  • [CREDENTIALS_UNSAFE]: The skill searches for and extracts stored credentials from various sources including AWS (~/.aws/credentials), Azure, and GCP configuration files, as well as browser databases (Chrome, Edge, Firefox). It also targets PowerShell history files and registry keys containing autologon passwords.
  • [COMMAND_EXECUTION]: Executes administrative and diagnostic commands to harvest secrets, such as cmdkey /list, netsh wlan show profile for cleartext WiFi keys, and vssadmin to manipulate volume shadow copies for SAM/SYSTEM hive access.
  • [DATA_EXFILTRATION]: Explicitly instructs the user to exfiltrate DPAPI master key files and credential blobs to an external 'attacker machine' for offline processing if local decryption fails.
  • [REMOTE_CODE_EXECUTION]: References and provides usage instructions for several third-party offensive security tools like mimikatz, SharpDPAPI, and SharpChrome to perform privileged credential extraction.
  • [PROMPT_INJECTION]: The skill presents an indirect prompt injection surface as it ingests untrusted data from log files, transcripts, and sticky notes which could contain malicious instructions. Evidence: The skill reads from ConsoleHost_history.txt and PowerShell_transcript*. Capabilities: The skill has extensive shell command execution capabilities via cmd.exe and PowerShell which could be abused if instructions are found in these files.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Jul 6, 2026, 09:15 AM
Security Audit — agent-trust-hub — windows-credential-harvesting