outlook-web
Pass
Audited by Gen Agent Trust Hub on Sep 25, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill enforces a strict read-only architecture using platform-level action policies with a 'default: deny' stance, effectively preventing write operations like sending or deleting emails.- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data from external sources (emails and Teams messages).
- Ingestion points: Untrusted data enters the agent context via accessibility snapshots in the search, read, and summary subcommands.
- Boundary markers: The skill proactively enables AGENT_BROWSER_CONTENT_BOUNDARIES=1 to wrap and isolate page content from instructions.
- Capability inventory: Capabilities are restricted to a managed agent-browser subprocess with no unmanaged network or file-writing permissions.
- Sanitization: The skill utilizes content boundary markers to safely distinguish and extract data from the browser's output.
Audit Metadata