The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes a custom protocol runtime (scripts/_skill_protocol_runtime.py) to execute a local classification script (scripts/route_engineering_work.py).
The runtime uses subprocess.run with shell=False, ensuring that user-provided input strings are treated as literal arguments and cannot trigger shell-based command injection.
The runtime includes an explicit security check to ensure that any executed Python scripts are located within the authorized skill directory.
[DATA_EXFILTRATION]: No network operations or access to sensitive local data (e.g., credentials, SSH keys) were detected. The skill logic is confined to processing the user's request text.
[PROMPT_INJECTION]: The skill instructions and associated routing policy focus on classification and specifically forbid the agent from performing state-changing actions like committing code or creating pull requests, mitigating the impact of potential adversarial inputs.
[DYNAMIC_EXECUTION]: The skill uses importlib.import_module for internal component loading and dynamic command construction for local script routing. These patterns are used for framework configuration and do not involve executing untrusted remote code.

route-engineering-work