The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process email and mailbox-related content, which serves as a significant attack surface for indirect prompt injection. Malicious instructions embedded in received emails or attachments could attempt to override the agent's instructions or abuse downstream tools. The skill includes extensive documentation and instructions to mitigate this risk.
Ingestion points: The skill fetches external data via list_agent_messages and the chat_with_mailbox_agent tool, which interacts with the mailbox assistant.
Boundary markers: The instructions mandate the separation of authenticated user instructions from mailbox data and recommend using prompt-based boundaries to define allowed and prohibited actions.
Capability inventory: The downstream assistant has access to mailbox read/write tools, email scheduling, and third-party integrations via Composio and PayBox.
Sanitization: The security guidelines require stripping active HTML, ANSI/OSC escapes, and bidirectional controls, as well as enforcing character limits on ingested messages.
[EXTERNAL_DOWNLOADS]: The skill configuration defines a connection to a remote Model Context Protocol (MCP) server that provides the operational logic for the mailbox agent.
Source: https://console.mermail.app/mcp using streamable_http transport.
Verification: The endpoint belongs to the official Mermail application console, which is the primary service integrated by the author.

mermail-mail-agent