The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/agent.py executes system-level commands to gather diagnostic information. Specifically, it uses subprocess.run to list crontabs for all users on the system and to check package ownership of files using dpkg. This behavior is consistent with its stated purpose as a threat-hunting tool.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from the local filesystem which could theoretically contain malicious instructions targeting the AI agent if the agent processes the generated report.
Ingestion points: The script reads the contents of user shell profiles (e.g., .bashrc), crontab files, and systemd service units in scripts/agent.py to identify suspicious entries.
Boundary markers: No boundary markers are used when outputting the contents of these files into the final report.
Capability inventory: The skill has the ability to read arbitrary files and execute system commands via subprocess for auditing purposes.
Sanitization: The script does not sanitize the text content of the files it reads; it only performs regex-based risk assessment to flag suspicious patterns.
[SAFE]: Static analysis detections for reverse shell patterns in scripts/agent.py are evaluated as benign in this context. These patterns are stored within a list of signatures (SUSPICIOUS_PATTERNS) for the express purpose of identifying malicious activity on the host system as part of the tool's intended scanning logic, rather than being part of an attack payload within the skill itself.

analyzing-persistence-mechanisms-in-linux