The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The scripts/Install-MsGraphRunner.ps1 script downloads a ZIP archive from an external GitHub repository (https://github.com/merill/msgraph/releases/latest/download/msgraph.zip), extracts its contents, and identifies a runner script for execution. This script is subsequently executed by scripts/Invoke-GraphGateway.ps1 to perform Microsoft Graph operations.\n- [COMMAND_EXECUTION]: The script scripts/Invoke-GraphGateway.ps1 uses the PowerShell call operator to execute a command defined by the RunnerPath parameter or environment variables. This allows for the execution of arbitrary local or downloaded tools with arguments derived from user and agent input.\n- [EXTERNAL_DOWNLOADS]: The skill is configured to download a third-party utility from GitHub to act as its primary execution substrate for Microsoft Graph requests.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from Microsoft 365 sources (e.g., Outlook emails, OneDrive files) which may contain malicious instructions designed to manipulate the AI agent.\n
Ingestion points: Data retrieved from Microsoft Graph via the Invoke-GraphGateway.ps1 script and associated assets.\n
Boundary markers: The skill relies on high-level routing logic but lacks technical sanitization or boundary markers to prevent the agent from following instructions embedded in external data.\n
Capability inventory: The skill possesses extensive capabilities including reading/writing emails and events, and executing local scripts.\n
Sanitization: There is no evidence of sanitization, filtering, or escaping of data fetched from external Microsoft Graph endpoints.

microsoft-graph-gateway