The Agent Skills Directory

[SAFE]: No malicious patterns or security risks were identified. The skill is configured as a script-executor and is not user-invocable, which limits the potential attack surface.
[COMMAND_EXECUTION]: The skill invokes the local architecture resolution tool using subprocess.run to classify verification tasks. This call is constructed as a structured list of arguments and executed without a shell, which prevents shell command injection.
[SAFE]: The skill handles project-specific metadata and task files (e.g., execution.toon, TASK-*.json, marshal.json) to drive plan execution. All data processing is contained within the local environment, and no external network communication or data exfiltration was detected.
[SAFE]: Potential indirect inputs from task files are handled safely using shlex.split for command parsing before analysis by local tools. There are no dangerous dynamic execution patterns such as eval() or exec() used in the skill scripts.

manage-execution-manifest