M365 Declarative Agent Debugging Skill
Pass
Audited by Gen Agent Trust Hub on Mar 10, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [PROMPT_INJECTION]: The skill outlines a 'Memory via OneDrive' mechanism that involves processing data from external shared folders, creating a surface for indirect prompt injection. 1. Ingestion points: Shared OneDrive folder links provided by the user in chat (SKILL.md). 2. Boundary markers: No delimiters or instructions to ignore embedded commands are specified for the ingested content. 3. Capability inventory: Includes 'OneDriveAndSharePoint' for file access and 'CodeInterpreter' for logic execution (SKILL.md). 4. Sanitization: No content validation or sanitization steps are documented for the files read from OneDrive.
- [COMMAND_EXECUTION]: The skill provides instructions for executing developer-oriented commands. Evidence: Instructions include 'npx teamsapp validate' and PowerShell scripts to verify icon transparency. Context: These commands are part of the official Microsoft Teams Toolkit workflow and are appropriate for the skill's debugging purpose.
- [EXTERNAL_DOWNLOADS]: The skill references external tools and documentation. Evidence: Uses 'npx' to execute the 'teamsapp' package and links to official Microsoft documentation. Context: These resources are provided by Microsoft, a trusted organization, and are handled as safe references in this context.
Audit Metadata