skills/dfinity/icskills/smtp-gateway/Gen Agent Trust Hub

smtp-gateway

Pass

Audited by Gen Agent Trust Hub on Nov 19, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill provides various icp (DFINITY's ICP CLI) and dig commands for local testing, deployment, and DNS verification. These are standard development operations for the platform.
  • [INDIRECT_PROMPT_INJECTION]: The skill's primary purpose is to process external, untrusted data (inbound email) which is a known vector for indirect prompt injection if the content is later processed by an LLM.
  • Ingestion points: The smtp_request and smtp_request_validate methods in both Rust and Motoko examples receive raw email headers and body content.
  • Boundary markers: The skill explicitly instructs users to not trust the sender's address (Pitfall 8) and provides guidance on DKIM verification to establish identity.
  • Capability inventory: The canister stores data in stable memory and is capable of performing https-outcalls (e.g., for DNS-over-HTTPS as mentioned in Pitfall 8).
  • Sanitization: The provided examples include input bounding (e.g., MAX_BODY_BYTES, MAX_HEADERS) and specific checks for authorized callers (principals).
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 19, 2026, 02:13 AM