The Agent Skills Directory

[SAFE]: The skill was audited across all threat categories and no malicious patterns were detected. It demonstrates high-quality implementation and attention to security best practices.
[CREDENTIALS_UNSAFE]: The skill handles sensitive Google Cloud service account keys but manages them securely. It implements a local setup flow that stores the key in the user's home directory (~/.config/firebase) and explicitly sets restrictive file permissions (0600) to ensure the secret is not accessible by other local users.
[DATA_EXFILTRATION]: All network operations are directed to official Google API endpoints (googleapis.com) for Firebase, Firestore, and Identity Toolkit services. No data is sent to unauthorized or unknown third-party domains.
[PROMPT_INJECTION]: The instructions are instructional and transparent, lacking any attempts to override safety filters or bypass system guidelines.
[DATA_EXPOSURE]: The skill handles PII (emails and phone numbers of Firebase Auth users). It correctly addresses this by requiring explicit user confirmation before executing the script that retrieves this data.
[REMOTE_CODE_EXECUTION]: The skill uses only the Ruby standard library and does not download or execute any external scripts, packages, or gems, significantly reducing the supply chain attack surface.

firebase