The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/agent.py uses subprocess.check_output to execute system-level PowerShell commands (Get-ADUser, Get-WinEvent) to perform environment enumeration and hunt for service account anomalies.
[DATA_EXFILTRATION]: The skill queries sensitive environmental metadata from Active Directory, including account names (SamAccountName), administrative privilege status (AdminCount), group memberships (MemberOf), and historical logon information containing source IP addresses. This data is returned to the agent's context and represents a high-value data exposure surface.
[DYNAMIC_EXECUTION]: The skill employs subprocess to invoke external shells (PowerShell) for performing system queries, including commands constructed with runtime arguments derived from CLI input.
[INDIRECT_PROMPT_INJECTION]: The scripts/process.py script identifies a vulnerability surface by ingesting and parsing external JSON and CSV files.
Ingestion points: The parse_logs function in scripts/process.py reads potentially untrusted data from external log files.
Boundary markers: None identified; untrusted data from log fields like CommandLine and Task_Content is extracted and interpolated into findings without delimiters or warnings against following embedded instructions.
Capability inventory: The skill set includes capabilities for Active Directory enumeration and system command execution through agent.py.
Sanitization: Extracted command-line data is truncated but not sanitized or escaped to prevent the agent from interpreting embedded instructions as legitimate commands.

detecting-service-account-abuse