The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external sources that could contain malicious instructions.
Ingestion points: The skill processes data from email content and webhook deliveries via tools like list_webhook_deliveries and get_webhook (specified in SKILL.md and references/webhooks.md).
Boundary markers: Instructions mandate resolving exact resources via stable IDs and obtaining explicit approval for administrative changes.
Capability inventory: The skill has high-privilege administrative capabilities, including member role management, mailbox provisioning, and webhook configuration.
Sanitization: Guidelines explicitly forbid the agent from repeating signing secrets or full destination URLs in conversation logs to prevent credential leakage.
[EXTERNAL_DOWNLOADS]: The skill configures a connection to a remote Model Context Protocol (MCP) server.
The agent configuration in agents/openai.yaml specifies a dependency on an MCP server hosted at https://console.mermail.app/mcp.
[COMMAND_EXECUTION]: The skill leverages specific administrative tools to perform workspace tasks.
It utilizes authenticated tools for mailbox creation (create_mailbox), workspace membership management (update_member_role), and domain verification (verify_email_domain).
It implements a 'destructive action' workflow requiring a matching single-use token from prepare_destructive_action for high-risk operations.

mermail-administer-workspace