The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Fetches and executes the agentmail-mcp package from the NPM registry via npx during setup.
[EXTERNAL_DOWNLOADS]: Requires the installation of the mcp package from the PyPI registry.
[PROMPT_INJECTION]: Identified an indirect prompt injection surface as the agent processes untrusted data from incoming emails.
Ingestion points: Untrusted content enters the agent's context through the get_thread and list_threads tools in SKILL.md.
Boundary markers: Absent; there are no instructions for the agent to distinguish between its own logic and instructions embedded in incoming emails.
Capability inventory: The agent has permissions to send_message, reply_to_message, create_inbox, and delete_inbox across SKILL.md.
Sanitization: No sanitization or validation of incoming email content is implemented or instructed.

agentmail