The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from food.com through search and scraping tools, creating a surface for potential indirect prompt injection.
Ingestion points: Data returned from the food_com__post_nlp_search tool and various unbrowse.* scraping and browsing tools described in SKILL.md.
Boundary markers: The instructions include a directive to "Report only what a succeeded result says," providing a basic logical filter for the agent.
Capability inventory: The skill possesses the ability to search, scrape, and perform actions (like login autofill) on the food.com domain.
Sanitization: There are no explicit instructions for the agent to sanitize or escape the content fetched from the external website.
[EXTERNAL_DOWNLOADS]: The skill connects to a remote MCP server hosted at https://unbrowse.ai/mcp/food.com to fetch tool definitions and execute requests. This is a vendor-managed resource consistent with the skill's primary function.
[COMMAND_EXECUTION]: The documentation provides manual setup instructions for users, including claude mcp add and npx skills add to register the MCP server and skill. These are standard administrative commands for the agent environment.

food