The Agent Skills Directory

[SAFE]: The skill is a standard technical assistant designed to guide users through the CUDA-Q platform. No malicious content or intent was identified.
[PROMPT_INJECTION]: No behavioral overrides, safety bypasses, or instruction-ignoring patterns were found. The use of $ARGUMENTS is limited to functional routing within the guide.
[DATA_EXFILTRATION]: No commands access sensitive file paths (e.g., .ssh, .env, .aws) or perform unauthorized network operations. References to credentials correctly instruct the user to manage them via environment variables per provider documentation.
[EXTERNAL_DOWNLOADS]: The skill references official NVIDIA documentation (nvidia.github.io) and standard package installation via pip. These are recognized vendor resources and do not represent a security risk.
[REMOTE_CODE_EXECUTION]: Code snippets provided in the guide (e.g., installation scripts and Bell state kernels) are intended for user execution to set up and verify the environment. There are no instructions for the agent to execute untrusted remote code.
[COMMAND_EXECUTION]: While the skill has access to the Bash tool, it does not use it for dangerous operations. The bash commands listed in the documentation are educational examples for the user.
[CREDENTIALS_UNSAFE]: No hardcoded API keys or secrets were found. The skill follows best practices by advising users to use environment variables for hardware provider access.

cudaq-guide