The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/agent.py utilizes the subprocess module to run system utilities including powershell, tasklist, and ps. These calls are implemented using list-based arguments without shell=True, which is a secure practice, and are used solely to gather diagnostic system information.
[EXTERNAL_DOWNLOADS]: Instructions in SKILL.md and references/api-reference.md facilitate the acquisition of threat intelligence feeds from well-known and trusted security organizations, such as CISA (cisa.gov) and abuse.ch (feodotracker.abuse.ch, threatfox.abuse.ch). These downloads are used to update local Indicators of Compromise (IOC) lists.
[INDIRECT_PROMPT_INJECTION]: The skill acts as an ingestion point for external, untrusted data by analyzing Zeek network logs and Windows Event Logs. 1. Ingestion points: Data is read from log files and system command outputs in scripts/agent.py and scripts/process.py. 2. Boundary markers: The tools rely on standard structured formats (TSV, CSV, and JSON) to delineate data fields, reducing the risk of injection attacks. 3. Capability inventory: The skill possesses capabilities for file system auditing, system process monitoring, and reading security logs. 4. Sanitization: Input data is parsed into structured dictionaries and objects, with specific fields (like network ports) being cast to integers to ensure data integrity.

detecting-ransomware-precursors-in-network