The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute local CLI commands (gbrain google setup, gbrain waiting, etc.) to manage Google integrations. These appear to be part of the primary purpose of the tool and are executed within a local 'gbrain' environment.
[CREDENTIALS_SAFE]: The skill explicitly warns against putting secrets in command-line arguments or chat. It recommends saving Google Cloud JSON secret files to local disk with restricted permissions (0600) and passing the file path or using environment variables, which aligns with security best practices.
[DATA_EXPOSURE]: The skill involves reading Gmail, Calendar, and Contact data. The instructions note that commitment extraction sends text to a configured chat provider (AI backend) for processing, but provides an opt-out mechanism (gbrain config set loops.extraction_enabled false) and advises informing the user of this once during setup.
[INDIRECT_PROMPT_INJECTION]: As the skill ingests external data (Gmail threads), it is technically susceptible to indirect prompt injection. However, the skill primarily uses structured commands and local tools to manage this data, and does not exhibit high-privilege automated actions that would be triggered by email content without user oversight.

google-loops