The Agent Skills Directory

[SAFE]: The skill utilizes the official azure-containerregistry and azure-identity packages, which are trusted libraries for interacting with Microsoft Azure services. All network activity is directed at official Azure domains such as azurecr.io and mcr.microsoft.com.
[INDIRECT_PROMPT_INJECTION]: The skill processes metadata and binary blobs from an external source (Azure Container Registry) and has capabilities to write files and delete resources. This represents a standard management attack surface; however, the skill provides specific guidance on obtaining authorization and performing dry-runs before destructive actions.
Ingestion points: Registry repository names, tag properties, and manifest metadata are read from the Azure API (SKILL.md).
Boundary markers: Not explicitly defined; uses standard SDK return objects.
Capability inventory: Includes file system writing (open("layer.tar.gz", "wb")) and manifest/repository deletion (SKILL.md).
Sanitization: Standard SDK validation of API responses.
[EXTERNAL_DOWNLOADS]: The skill instructs the user to install azure-containerregistry via pip. This is a well-known, official library from Microsoft and is considered safe to use.

azure-containerregistry-py