The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/agent.py utilizes subprocess.check_output to execute external security tools including SharpHound.exe and bloodhound-python. While these tools are standard for Active Directory auditing, the execution of shell commands with arguments (such as domain names and credentials) passed from the agent context represents a capability that must be monitored for argument injection.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data exported from Active Directory environments (JSON and ZIP files).
Ingestion points: Data is ingested through the analyze_bloodhound_json function in scripts/agent.py and the import_zip method in scripts/process.py.
Boundary markers: No explicit boundary markers or instructions to ignore embedded content were found when processing the data for analysis or reporting.
Capability inventory: The skill has the capability to execute subprocesses (in agent.py) and write files to the local system (in both agent.py and process.py).
Sanitization: The agent.py script performs basic output truncation (500 characters), but otherwise, the properties extracted from AD objects are processed without significant sanitization, which could allow maliciously crafted AD object properties to influence the agent's output or logic.

exploiting-active-directory-with-bloodhound