The Agent Skills Directory

Indirect Prompt Injection Surface: The skill facilitates the retrieval of data from external Azure Storage resources (Blobs, File Shares, etc.), which creates a potential surface for indirect prompt injection. If an agent retrieves and processes untrusted content without adequate boundary markers, it might inadvertently execute instructions embedded within that data.
Ingestion points: The skill includes tools such as storage_blob_get and CLI commands like az storage blob download (documented in SKILL.md) that allow the agent to ingest external content into its context.
Capability inventory: The skill provides significant capabilities, including the ability to write or modify data via storage_blob_put and az storage blob upload, which could be misused if a prompt injection occurs.
Boundary markers: The instructions do not currently include explicit delimiters or 'ignore embedded instructions' warnings when the agent is prompted to read and process blob content.
Sanitization: There are no specified patterns for sanitizing, escaping, or validating the content retrieved from external storage before it is interpolated into the agent's reasoning process.

azure-storage