inbound-processing

Pass

Audited by Gen Agent Trust Hub on Sep 25, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill defines processes for ingesting untrusted data from external email sources, which represents an indirect prompt injection surface area.
  • Ingestion points: Webhook payloads from email providers (Postmark, SendGrid, Mailgun) and storage buckets (AWS SES).
  • Boundary markers: The documentation explicitly recommends implementing detection weights for suspicious patterns and using sanitization layers to delimit untrusted content.
  • Capability inventory: The skill itself implements no active capabilities (such as file system writes or network operations) as it is purely informational.
  • Sanitization: Provides high-quality guidance on HTML sanitization, character normalization to UTF-8, and strict allowlisting of tags and attributes to mitigate XSS and injection risks.
  • [PROMPT_INJECTION]: The skill contains several keywords associated with prompt injection, such as 'ignore previous instructions', 'system:', and 'override your rules'. These strings are used as examples within a tutorial section that teaches agents how to detect and filter malicious inbound emails, rather than being instructions directed at the agent itself.
  • [EXTERNAL_DOWNLOADS]: The documentation references several reputable libraries for email parsing and sanitization (e.g., mailparser, Talon, email_reply_parser). These references are standard technical documentation and do not involve automated execution or installation during skill loading.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 25, 2026, 05:20 PM