agent-email-inbox
Pass
Audited by Gen Agent Trust Hub on Jul 14, 2026
Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATION
Full Analysis
- [PROMPT_INJECTION]: The skill implements a defensive layer in
references/security-levels.mdto identify and filter malicious instructions within processed emails. Although it lists phrases like 'ignore previous instructions', these are part of a sanitization mechanism rather than an attempt to override agent behavior. - [EXTERNAL_DOWNLOADS]: Mentions well-known and reputable third-party services and tools in
SKILL.md, including Resend, ngrok, and Cloudflare, which are standard in the industry for their respective functions. - [DATA_EXFILTRATION]: Includes clear instructions in
SKILL.mdon avoiding the exposure of sensitive credentials by recommending the use of environment variables for API key management. - [SAFE]: The architecture advocates for security-first implementations, including rate limiting, content truncation, and restricted agent capabilities when handling untrusted external inputs.
Audit Metadata