peristyle-grocery-cart

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 0.90). The prompt explicitly describes obtaining a Kroger api_key and instructs the agent to save and include it as an Authorization: Bearer pk_… header (and shows raw examples), which requires the LLM/agent to handle or emit secret tokens verbatim (even though an MCP backchannel is recommended as a safer option).

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.75). The required workflow ingests outsider-authored free text from the recipe index (e.g., recipe titles/ingredient text and source_url/creator attribution) returned by runtime API calls like POST /v1/recipes/search / GET /v1/recipes, where the underlying recipe content is authored by third parties (not the operating user) and is then placed into the agent’s LLM context for confirmation and display.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The skill requires calling the Peristyle Grocery Cart API at runtime (default http://localhost:8001 and its /v1/* endpoints), and the JSON responses (recipes, matches, suggested products, UPCs) directly control the agent's prompts, confirmations, and cart-add actions, making this a required external runtime dependency that influences instructions.