agentmail

Pass

Audited by Gen Agent Trust Hub on Mar 14, 2026

Risk Level: SAFEPROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection via incoming emails.
  • Ingestion points: scripts/check_inbox.py (polling) and references/WEBHOOKS.md (webhook handlers) ingest message content from external senders.
  • Boundary markers: The provided code scripts do not include boundary markers or delimiters for untrusted content. The author addresses this risk in SKILL.md by suggesting the use of allowlists and separate review sessions.
  • Capability inventory: The skill can send emails (scripts/send_email.py) and attach local files to outgoing messages. These capabilities could be exploited if an agent follows instructions embedded in a malicious incoming email.
  • Sanitization: No automated sanitization or filtering of incoming email content is implemented within the provided Python scripts.
  • [DATA_EXFILTRATION]: The skill transmits email data and attachments to the vendor's API endpoint (api.agentmail.to). While this is the intended functionality, it involves sending potentially sensitive user-provided content to an external service.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 14, 2026, 06:02 PM