The Agent Skills Directory

[COMMAND_EXECUTION]: The automation scripts scripts/agent.py and scripts/process.py execute system-level commands (PowerShell, iptables, netsh) by interpolating variables directly into command strings without adequate shell escaping.
Evidence: scripts/agent.py uses f-strings in disable_ad_account and block_ip_firewall to build PowerShell commands. scripts/process.py uses similar patterns for firewall rules and system discovery tools like netstat and ps.
[EXTERNAL_DOWNLOADS]: The skill documentation and scripts recommend installing multiple third-party Python packages for API interaction and directory services.
Evidence: requests, ldap3, python-dateutil, pyyaml, and crowdstrike-falconpy are listed as requirements.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it ingests and acts upon data from external security tools (Splunk and CrowdStrike) where an attacker might influence content.
Ingestion points: SIEM search results and EDR detection metadata processed in scripts/process.py.
Boundary markers: No delimiters or instruction-ignore markers are used when processing external data.
Capability inventory: The skill has high-privilege capabilities including modifying firewall rules, disabling user accounts, and collecting system-wide process/network information.
Sanitization: Input variables such as hostnames and usernames are subjected to minimal sanitization before being used in command execution logic.

containing-active-security-breach