agentmail

Pass

Audited by Gen Agent Trust Hub on Jul 14, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: Fetches a platform-specific binary from the official AgentMail GitHub releases during package installation (npm/scripts/postinstall.js).
  • [COMMAND_EXECUTION]: Instructs the agent to install and execute a CLI tool (agentmail-cli) and executes a binary wrapper (npm/bin/agentmail) using execFileSync to run the downloaded AgentMail core logic.
  • [PROMPT_INJECTION]: The CLI tool includes a feature (pkg/cmd/flagoptions.go) that allows reading arbitrary local files and including their contents in API requests (as body or attachments) by prefixing a parameter value with the @ character. This creates a surface for indirect prompt injection where a malicious actor could trick the agent into exfiltrating sensitive local files (e.g., SSH keys, credentials) via email.
  • Ingestion points: Agent instructions or user-provided data processed by the agent in SKILL.md.
  • Boundary markers: Absent.
  • Capability inventory: The agentmail tool has the capability to read filesystem contents and transmit data via network requests to the AgentMail API.
  • Sanitization: Absent; the tool treats any string starting with @ as a potential file path to be read without further validation.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 14, 2026, 11:24 PM
Security Audit — agent-trust-hub — agentmail