m365-mail
Pass
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [Indirect Prompt Injection] (LOW): The skill allows an agent to read untrusted content from emails, creating a risk for indirect prompt injection. 1. Ingestion points: Email content retrieved via 'read', 'inbox', and 'search' commands. 2. Boundary markers: No delimiters or ignore instructions are mentioned. 3. Capability inventory: The skill can send, delete, and move emails, which an attacker could exploit through injected instructions. 4. Sanitization: Verification is impossible as script source code is not provided.
- [External Downloads] (LOW): The setup instructions require the installation of the 'msal' and 'requests' Python packages.
- [Data Exfiltration] (LOW): The tool communicates with the Microsoft Graph API (a non-whitelisted domain) to process email data.
- [Command Execution] (LOW): The documentation requires making a local script 'm365mail.py' executable and running it, which is standard for CLI-based agent skills.
Audit Metadata