The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/agent.py uses subprocess.run to execute security-sensitive operations including firewall modifications (iptables) and Active Directory management (powershell).
[COMMAND_EXECUTION]: In scripts/agent.py, functions such as disable_ad_account and reset_ad_password use f-strings to construct PowerShell commands. This approach is vulnerable to command injection if input parameters like usernames or domain controllers contain malicious characters.
[COMMAND_EXECUTION]: The sinkhole_domain function in scripts/agent.py appends unvalidated domain strings directly to the system /etc/hosts file, which could allow an attacker to manipulate local DNS resolution.
[DATA_EXFILTRATION]: The skill implements a collect_volatile_evidence function that gathers sensitive system state information (e.g., process lists, network connections, and open files). While intended for forensics, this capability exposes high-value system metadata.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through its processing of incident-related data.
Ingestion points: Input parameters are passed from the agent to scripts/agent.py via command-line arguments.
Boundary markers: The skill does not implement delimiters or warnings to ignore instructions embedded in the processed data.
Capability inventory: The skill possesses the ability to execute shell and PowerShell commands and perform file-write operations.
Sanitization: There is no evidence of validation or sanitization for parameters used in command construction or file writes.

containing-active-breach