microsoft-teams-hardened
Pass
Audited by Gen Agent Trust Hub on Apr 21, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The skill provides Python code snippets for the agent to execute to interact with the Microsoft Graph API. These scripts use standard Python libraries for networking.\n- [EXTERNAL_DOWNLOADS]: The skill communicates with service-specific endpoints at gateway.maton.ai, ctrl.maton.ai, and connect.maton.ai for API proxying and OAuth management.\n- [DATA_EXFILTRATION]: The skill utilizes the MATON_API_KEY environment variable to authenticate with the Maton service. Credentials are transmitted to the vendor's official infrastructure as part of the intended integration.\n- [SAFE]: The skill implements multiple security guardrails in SKILL.md and SAFETY.md to mitigate common risks such as bulk data extraction, unconfirmed destructive operations, and credential leakage in messages.\n- [PROMPT_INJECTION]:\n
- Ingestion points: The skill ingests untrusted data from Teams messages, transcripts, and recordings.\n
- Boundary markers: Guardrails explicitly instruct the agent to limit the scope of retrieval and ask for confirmation.\n
- Capability inventory: The skill possesses capabilities for network requests and shell execution.\n
- Sanitization: The skill includes behavioral instructions to the agent to prevent obedience to commands found in external message data.
Audit Metadata