The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from emails, creating a risk that malicious instructions within an email could manipulate the agent's behavior.\n
Ingestion points: Tools like GMAIL_FETCH_EMAILS and GMAIL_FETCH_MESSAGE_BY_MESSAGE_ID ingest external data from the user's inbox into the agent's context.\n
Boundary markers: The instructions lack specific delimiters or system prompts to distinguish between email content and valid instructions, which could lead to instruction confusion.\n
Capability inventory: The agent is granted high-privilege capabilities including GMAIL_SEND_EMAIL, GMAIL_REPLY_TO_THREAD, and GMAIL_BATCH_MODIFY_MESSAGES.\n
Sanitization: There is no mechanism described for sanitizing or validating ingested email data before it is acted upon.\n- [EXTERNAL_DOWNLOADS]: The skill requires a connection to a remote Model Context Protocol (MCP) server at https://rube.app/mcp.

gmail-automation