The Agent Skills Directory

[Command Execution]: The skill instructs the agent to use the mda CLI tool for project initialization (mda init), local development (mda dev), and deployment (mda deploy). These are standard operations for the intended developer workflow.
[External Package Installation]: The documentation references the installation of the managed-deepagents package via uv (Python) and npm (Node.js). These resources originate from the vendor and are necessary for the skill's functionality.
[Credential Management Best Practices]: The skill includes strong security guidance for the agent, specifically instructing it not to hardcode credentials, to use .env files correctly, and to avoid echoing secrets to the terminal or user replies.
[Indirect Prompt Injection Surface]: The skill involves scaffolding agents based on user descriptions and supports shared durable memory. The documentation acknowledges this attack surface by explicitly warning that memory content should be treated as untrusted input and should not grant authority or bypass safety approvals.
[Human-in-the-loop Controls]: The skill advocates for the use of interrupt_on for sensitive tool calls and provides guidance on setting file permissions, which are effective mitigations against unauthorized automated actions.

managed-deep-agents