The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/agent.py executes shell commands via subprocess.run(shell=True) inside the find_suid_binaries function. The evidence_root path variable is interpolated directly into the command string without sanitization, allowing for arbitrary command execution if a malicious path string is provided (e.g., via command-line arguments).
[DATA_EXFILTRATION]: The skill is designed to collect and read highly sensitive files including /etc/shadow, /etc/sudoers, and user .ssh/authorized_keys. While this is the intended forensic purpose, the systematic collection and exposure of these system secrets represents a significant data exposure risk.
[PROMPT_INJECTION]: The skill has an indirect prompt injection surface. 1. Ingestion points: It reads untrusted data including .bash_history and system logs from a potentially compromised target system. 2. Boundary markers: No delimiters or instructions are used to isolate this data from the agent's logic. 3. Capability inventory: The skill uses subprocess.run and broad file system operations in SKILL.md and scripts/agent.py. 4. Sanitization: Log content is displayed and processed without escaping or validation, potentially allowing malicious instructions in logs to influence the agent.

analyzing-linux-system-artifacts