The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/agent.py script executes the ausearch system utility using the subprocess module to query audit logs. This is a controlled execution using a list of arguments, which prevents shell-based command injection.
[DATA_EXFILTRATION]: The skill is designed to read and monitor sensitive system logs and files, such as /var/log/audit/audit.log and /etc/shadow, which is essential for its stated purpose of intrusion detection. There is no evidence of these findings being transmitted to external or unauthorized destinations.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted system log data which could potentially contain attacker-influenced content.
Ingestion points: System audit logs read from /var/log/audit/audit.log by the Python agent script.
Boundary markers: The log parsing logic separates fields but does not include explicit instructions for the AI to ignore instructions potentially embedded within the log strings.
Capability inventory: The skill can execute specific local binaries and write data to the local file system.
Sanitization: Field values from the logs are extracted via regular expressions but are not further sanitized for potential prompt injection patterns.

analyzing-linux-audit-logs-for-intrusion