inbox-providers
Pass
Audited by Gen Agent Trust Hub on Sep 25, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill describes architectures where AI agents ingest untrusted data from inbound emails, which is an inherent attack surface for indirect prompt injection.
- Ingestion points: Processes inbound email content (text/HTML) via provider APIs or webhooks as described in the "Setting up an agent inbox" and "Safety architecture" sections of
SKILL.md. - Boundary markers: The skill explicitly recommends defense-in-depth measures, including Layer 2 (inbound scanning for injection patterns) and Layer 3 (content isolation) to delimit untrusted content.
- Capability inventory: Agents configured via this skill would have the capability to send, receive, and search emails, and potentially act on their contents.
- Sanitization: The skill advocates for infrastructure-level sanitization, including Unicode normalization, stripping hidden text/HTML, and utilizing canary tokens to detect data exfiltration.
Audit Metadata