The Agent Skills Directory

[COMMAND_EXECUTION]: The skill interacts with the host environment to run standard development tools like flutter, uvicorn, and firebase. These operations are performed using subprocess.run with shell=False, which prevents command injection vulnerabilities.
[DYNAMIC_EXECUTION]: The skill includes functionality to execute related scripts from a sibling skill directory (project-engineering). This execution is handled securely using the local Python interpreter and absolute script paths.
[INDIRECT_PROMPT_INJECTION]: The skill processes project metadata from files like pubspec.yaml and PRODUCT.md to detect the project stack. While this is an ingestion of potentially untrusted data, the skill only uses this information for stack detection and does not interpolate it into shell commands or unsafe prompts.

flutter-engineering