teams-desktop-automation
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes PowerShell scripts to interact with the Windows UI Automation API. This allows the agent to locate the Microsoft Teams process, identify the message composer, and set draft content. Safety mechanisms are in place, such as forbidding the use of
SendKeysor button invocation, ensuring the final send action remains with the human user. - [DYNAMIC_EXECUTION]: The script
Set-TeamsDesktopDraft.ps1utilizes theAdd-Typecmdlet to compile a small, static C# snippet at runtime. This snippet interfaces with the Win32PrintWindowAPI to capture screenshots of the Teams window for user verification. This is a standard method for achieving functional UI capture not natively available in PowerShell. - [INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it reads existing text from the Teams UI composer.
- Ingestion points: Data is read from the
ValuePattern.Current.Valueproperty of the Teams 'Edit' control (composer) inSet-TeamsDesktopDraft.ps1. - Boundary markers: The skill uses strict window title matching (
RecipientDisplayName | Microsoft | ExpectedAccount | Microsoft Teams) to ensure it only interacts with the intended chat. The safety contract explicitly requires a manual user 'send' action. - Capability inventory: The skill can open URI schemes (
msteams:/l/chat), interact with UI elements (read/write), and save screenshots to the local file system. - Sanitization: Recipient and account emails are validated against a regular expression before use. The skill also performs a comparison between the existing draft and the new message, refusing to overwrite if they differ.
Audit Metadata