The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it is designed to ingest and process text from external emails, which could contain malicious instructions for the agent.
Ingestion points: Incoming messages are retrieved in scripts/check_inbox.py and received via webhooks as described in references/WEBHOOKS.md.
Boundary markers: The provided Python scripts do not implement delimiters or boundary markers for email content, though SKILL.md recommends them as a defense layer.
Capability inventory: The agent has the capability to send emails (scripts/send_email.py), read local files for attachments, and interact with external APIs like GitHub (references/EXAMPLES.md).
Sanitization: No input sanitization is performed within the provided Python scripts; however, the author provides a detailed TypeScript remediation example in SKILL.md for filtering senders.
[EXTERNAL_DOWNLOADS]: The skill requires several standard third-party libraries for full functionality.
Evidence: The documentation recommends installing agentmail, python-dotenv, flask, ngrok, pdfplumber, and requests via package managers.
[COMMAND_EXECUTION]: The skill includes scripts that perform network operations and file system access.
Evidence: scripts/send_email.py reads files from local paths to create attachments and communicates with api.agentmail.to. references/EXAMPLES.md demonstrates workflows involving temporary file creation and interaction with the GitHub API.

agentmail