The Agent Skills Directory

[COMMAND_EXECUTION]: The skill defines a large set of tools (51 total) that perform HTTP operations against official Swiggy API endpoints at mcp.swiggy.com. These tools are scoped to legitimate service functions such as searching for restaurants, managing carts, and placing orders.
[PROMPT_INJECTION]: The instructions include 'Non-negotiable rules' designed to constrain agent behavior for user safety. These rules mandate human-in-the-loop confirmation for tools that spend money (place_food_order, checkout, book_table) or destroy state (delete_address, clear_cart). No malicious override or bypass patterns were detected.
[DATA_EXFILTRATION]: The skill accesses delivery addresses and phone numbers as part of its core functionality. It implements explicit privacy protections, such as stripping coordinates from address results in the get_addresses tool. The report_error tool uses a transparent mailto: flow for user support, which is a standard and user-initiated practice for debugging.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest dynamic content from restaurant menus and product listings. The author mitigates this risk by requiring the agent to refresh cart state frequently and ensuring that critical operations always stop for user verification before proceeding.
[CREDENTIALS_SAFE]: The skill utilizes a modern OAuth 2.1 + PKCE flow for authentication. It explicitly forbids passing session credentials as tool arguments, instead relying on standard bearer tokens within the connection layer.

swiggy-mcp