The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides instructions for querying system and application logs using tools like journalctl. If an attacker writes malicious instructions to the logs, the agent could be influenced while processing them.\n
Ingestion points: System logs via journalctl and application logs in /var/log/.\n
Boundary markers: No explicit delimiters or instructions are provided to separate log data from agent commands.\n
Capability inventory: The skill provides a shell script with curl capabilities and a Python class for logging.\n
Sanitization: While structured logging (JSON) is used, no specific sanitization logic is implemented to remove potential instructions from the log data.\n- [DATA_EXFILTRATION]: The verify-log-integrity.sh script uses curl to send data to an external URL defined by an environment variable (ALERT_WEBHOOK_URL). This establishes a potential network egress point for data.

audit-logging