The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from an external catalog API (xstock.mermail.app) and user-provided text which could theoretically contain malicious instructions.
Ingestion points: Data enters the agent via GET /api/v1/products and GET /api/v1/categories endpoints.
Boundary markers: The skill implements a strict intake limit of 10,000 normalized characters for external content and explicitly warns that catalog/provider output is untrusted.
Capability inventory: The skill can perform portfolio reads (paybox_get_portfolio) and initiate token swaps (paybox_request_swap).
Sanitization: Instructions require validating evidence hashes, identity fingerprints, and provenance source types for all catalog items. It also prohibits inferring categories from names or tickers and mandates decimal-string arithmetic to prevent floating-point errors in financial amounts.
[EXTERNAL_DOWNLOADS]: The skill fetches product information and verification data from the vendor's catalog at https://xstock.mermail.app and uses an MCP server at https://console.mermail.app/mcp. These network operations are limited to the vendor's infrastructure for the primary purpose of product discovery.
[COMMAND_EXECUTION]: The skill executes financial transactions via the paybox_request_swap tool. Security is maintained through a mandatory human-in-the-loop workflow where the PayBox MCP App handles the actual quote review, fee disclosure, and user signing. The skill is strictly prohibited from bypassing these steps or using alternate purchase paths like transfers or arbitrary plugins.

mermail-xstocks-desk