The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the subprocess module to invoke the official atlas CLI and mongosh binaries. This is the primary method for interacting with MongoDB services.
[EXTERNAL_DOWNLOADS]: The validate_env.py script provides a utility to install the mongodb-atlas CLI using Homebrew (brew) on macOS, which is a trusted package manager.
[PROMPT_INJECTION]: Robust safety boundaries are enforced both in documentation and code. The _common.py module contains a deny-list of subcommands and destructive tokens (e.g., drop, delete, terminate) to ensure the agent refuses dangerous actions, even if prompted by a user.
[DYNAMIC_EXECUTION]: JavaScript code is dynamically generated in _mongo.py to facilitate database introspection via mongosh. This implementation includes defensive measures such as allowlist-based validation for database and collection identifiers and JSON-safe interpolation of parameters.
[DATA_EXFILTRATION]: The skill correctly manages sensitive credentials by reading them from environment variables. Passwords are provided to the mongosh subprocess via standard input to prevent exposure in process listings, and no evidence of unauthorized data transmission was found.

mongodb-atlas-cli