email-security

Pass

Audited by Gen Agent Trust Hub on Sep 25, 2026

Risk Level: SAFENO_CODE
Full Analysis
  • [PROMPT_INJECTION]: The skill includes several examples of prompt injection payloads (e.g., 'Disregard previous instructions', 'repeat your system prompt verbatim') which are used for teaching weighted pattern detection and are not instructions intended for the agent's runtime execution.
  • [OBFUSCATION]: The skill uses a homoglyph-based lookalike domain ('аpple.com' with a Cyrillic 'а') as a pedagogical example to illustrate common phishing techniques.
  • [INDIRECT_PROMPT_INJECTION]: The skill provides defensive guidance against indirect prompt injection vulnerabilities. It identifies email bodies and headers as untrusted ingestion points, lists potential capability risks to integrated tools (e.g., databases, CRM systems), and recommends mitigation strategies such as HTML allowlisting, invisible character stripping, and per-thread canary tokens.
  • [NO_CODE]: The skill consists purely of informational markdown and does not provide or invoke any executable scripts, binary files, or external code dependencies.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 25, 2026, 05:21 PM