The Agent Skills Directory

[Indirect Prompt Injection] (HIGH): The skill is highly vulnerable to indirect injection attacks because it combines high-privilege write/execute tools with external data processing.
Ingestion points: The skill reads and processes user requests and external deployment-related files (via Read and Grep tools) which may contain malicious instructions.
Boundary markers: Absent. There are no delimiters or instructions provided to help the agent distinguish between its core logic and data-embedded instructions.
Capability inventory: The skill allows the use of Bash (arbitrary command execution) as well as Write and Edit (filesystem modification).
Sanitization: None detected. No mechanisms are described to validate or escape external content before it is processed by the high-privilege tools.
[Command Execution] (MEDIUM): The skill explicitly requests the Bash tool. In the context of ML deployment (which often involves handling environment variables and secrets), providing unrestricted shell access to an agent processing untrusted project files is a significant security risk.

azure-ml-deployer