The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.run to execute various system commands, including docker inspect, docker ps, docker exec, docker logs, and lsof. These commands are used to gather health information, network configuration, and process status for the firewall containers. The commands are invoked using list-based arguments rather than shell strings, which mitigates standard command injection risks.
[INDIRECT_PROMPT_INJECTION]: The tool parses Squid access logs which contain external data like URLs and user agents. This represents a potential surface where malicious instructions could be embedded in logs to influence an agent's subsequent reasoning.
Ingestion points: Squid logs are read from /tmp directories or directly from Docker containers in scripts/parse-squid-logs.py and scripts/test-domain.py.
Boundary markers: The skill does not currently use specific boundary markers or delimiters when presenting log data to the agent.
Capability inventory: The skill has the ability to read local files, execute Docker management commands, and perform network connectivity checks.
Sanitization: Log lines are parsed using regular expressions to extract specific fields (timestamps, IPs, domains), providing a level of structural validation, though they are not explicitly sanitized for instruction-like content.

awf-debug-tools