The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted customer emails, which creates a surface for indirect prompt injection attacks. \n
Ingestion points: The agent uses tools like get_email, get_email_context, and get_thread to read customer-supplied message bodies and attachments (documented in references/tools.md). \n
Boundary markers: Robust boundaries are implemented, requiring all customer requests to be matched against an "owner-verified order record" before fulfillment. The skill explicitly states that customer assertions cannot fill verified identity or payment fields (references/workflows.md). \n
Capability inventory: The agent has the ability to send emails (reply_to_email), save drafts (save_draft), and initiate x402 payments via PayBox, all of which are gated by mandatory owner authorization (references/security.md). \n
Sanitization: The skill enforces a 10,000 character limit per message and strictly requires a scan_status: clean metadata flag before interpreting any content (references/security.md).\n- [EXTERNAL_DOWNLOADS]: The skill references and interacts with external well-known services for data retrieval. \n
The agent fetches market data and protocol information from official CoinMarketCap (CMC) endpoints, including https://mcp.coinmarketcap.com/mcp and https://pro-api.coinmarketcap.com (references/cmc-research.md). \n
It utilizes the Mermail platform's official MCP server at https://console.mermail.app/mcp for workspace and mailbox management (agents/openai.yaml).

mermail-research-agent