The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Directs the user to connect to an external MCP server at https://rube.app/mcp. This is a tool provider service (Composio) used for the skill's primary functionality.- [PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection because it ingests and processes untrusted content from Zendesk.- Ingestion points: The agent reads support ticket subjects, descriptions, and comments from Zendesk via tools like ZENDESK_GET_ZENDESK_TICKET_BY_ID and ZENDESK_LIST_ZENDESK_TICKETS.- Boundary markers: The instructions do not define clear delimiters or specific instructions to the model to ignore potential commands embedded in the ticket data.- Capability inventory: The skill possesses significant write capabilities, such as ZENDESK_REPLY_ZENDESK_TICKET, ZENDESK_UPDATE_ZENDESK_TICKET, and ZENDESK_DELETE_ZENDESK_TICKET, which could be manipulated by a successful injection.- Sanitization: There are no mentions of sanitizing or filtering the ticket content before it is processed by the agent.

zendesk-automation