The Agent Skills Directory

[REMOTE_CODE_EXECUTION] (LOW): The skill uses 'curl https://sdk.cloud.google.com | bash' to install the Google Cloud SDK. While this is a critical execution pattern, the source is a trusted domain (google.com), justifying a downgrade to LOW per the analysis framework.
[COMMAND_EXECUTION] (MEDIUM): The instructions include 'sudo apt-get install pass' for dependency management. Privilege escalation is generally a HIGH severity finding, but it is downgraded to MEDIUM here as it is necessary for the skill's primary purpose.
[COMMAND_EXECUTION] (MEDIUM): Multiple examples promote the use of the '--yolo' flag for automated execution of AI-generated prompts. This bypasses human review for terminal commands, which is a significant risk factor if the model interprets untrusted input.
[EXTERNAL_DOWNLOADS] (LOW): Software and API validation requests are directed to 'sdk.cloud.google.com' and 'generativelanguage.googleapis.com', which are trusted sources.
[PROMPT_INJECTION] (LOW): The skill describes CI/CD and Docker workflows where the AI analyzes untrusted repository content. This defines an Indirect Prompt Injection surface (Category 8), though no specific exploits are present in the provided documentation.

gemini-auth