The Agent Skills Directory

[PROMPT_INJECTION]: The skill includes robust defenses against instruction overriding. In rules.md and examples.md, it explicitly instructs the AI to adhere to 'Invariants', stating that user instructions cannot override core safety principles like the 'Alert-only' posture.
[DATA_EXFILTRATION]: No unauthorized data exfiltration was detected. The skill manages sensitive credentials (OAuth tokens, API keys) using environment variables and GitHub Secrets, which is the recommended practice for this type of automation. Network communication is limited to official Google API and Slack Webhook endpoints.
[EXTERNAL_DOWNLOADS]: The skill specifies well-known, legitimate Python dependencies in its requirements.txt (such as google-api-python-client and requests). These are sourced from official registries.
[COMMAND_EXECUTION]: No dangerous or unauthorized shell command execution patterns were found. The skill operates as a standard Python automation.
[SAFE]: The skill's architecture is read-only regarding Google Ads, and its write access to Google Sheets is restricted to maintaining its own configuration and state tabs.

budget-guardian