The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill guides users to install official, well-known libraries from Microsoft, including msgraph-sdk, azure-identity, and msal. These are standard tools for interacting with Microsoft's cloud ecosystem.
[COMMAND_EXECUTION]: The documentation mentions using standard CLI tools like azure-cli and brew for local authentication setup. This is a legitimate workflow for developers to authenticate their local environment.
[DATA_EXFILTRATION]: While the skill facilitates the reading and writing of data across Microsoft 365 (e.g., sending emails, uploading files), all network operations target official Microsoft endpoints (graph.microsoft.com and login.microsoftonline.com). The code is designed for legitimate data management within the user's authorized tenant.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface because it enables the agent to ingest untrusted data from external sources such as email bodies and file contents.
Ingestion points: External data enters the context through message listing (list_messages.py) and file reading (onedrive_operations.py) operations.
Boundary markers: The provided scripts do not explicitly implement boundary markers or instructions to ignore embedded commands in the processed data.
Capability inventory: The skill has significant capabilities, including sending emails (send_email.py), messaging via Teams (send_message.py), and modifying files (onedrive_operations.py).
Sanitization: The sample code demonstrates data retrieval but does not include specific sanitization or validation logic for the content of emails or files. Users should be aware that processing untrusted data could influence agent behavior.

microsoft-365