The Agent Skills Directory

[External Downloads]: The skill installs software and libraries directly from a repository hosted on GitHub. These downloads originate from a well-known organization and are used to install the primary logic of the tool itself.
[Command Execution]: The exec and manage capabilities allow for the execution of shell commands within created environments. This is a core feature for testing software deployments in isolated contexts. The skill includes a --confirmed flag requirement for model-generated plans, which serves as a human-in-the-loop safeguard before execution.
[Indirect Prompt Injection]: The manage capability represents an ingestion point where natural language instructions are converted into environment actions.
Ingestion points: User-provided requests in words passed to the manage command (referenced in SKILL.md).
Boundary markers: Not explicitly defined in the prompt instructions, though the tool separates the planning phase from execution.
Capability inventory: Includes the ability to execute commands (exec), copy files (file-push, file-pull), and modify environment states.
Sanitization: The tool implements a --confirmed flag, requiring the user to review and authorize the generated plan before any actions are taken, reducing the risk of unintended behavior.

amplifier-digital-twin-universe