The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes shell and PowerShell launcher scripts (scripts/run.sh, scripts/run.ps1) to execute a pre-bundled CLI binary named msgraph. This binary provides the core logic for searching the local documentation indexes and performing API calls.
[DATA_EXFILTRATION]: The skill is designed to manage and transmit sensitive data (OAuth tokens, Entra ID secrets, and Graph API resources) to official Microsoft endpoints (graph.microsoft.com and login.microsoftonline.com). While it handles high-value credentials, its operations are constrained to these trusted domains and include automatic cleanup of session-scoped token caches.
[PROMPT_INJECTION]: The skill possesses a data ingestion surface through msgraph graph-call, which retrieves data from the Microsoft Graph API. Because this data is external and potentially controlled by third parties (e.g., email content or directory fields), there is a risk of the agent interpreting embedded instructions. This is classified as an indirect prompt injection risk.
[CREDENTIALS_UNSAFE]: The documentation and scripts reference environment variables for authentication (e.g., MSGRAPH_CLIENT_SECRET, MSGRAPH_FEDERATED_TOKEN_FILE). These are used as standard mechanisms for providing credentials to the tool and do not represent hardcoded secrets within the skill itself.

msgraph