embedded-agents-messages-api

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

• Third-party content exposure detected (medium risk: 0.30). The required workflow defines a Next.js API route (templates/route.ts) that reads untrusted user-submitted prompt and history text from HTTP POST request bodies and passes them directly into the LLM agent runner.

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

• Potentially malicious external URL detected (medium risk: 0.30). The configuration files (.mcp.json) invoke the perplexity-mcp package via npx -y perplexity-mcp, establishing a third-party runtime package dependency with common/reputable unofficial provenance.

MEDIUM W009: Direct money access capability detected (payment gateways, crypto, banking).

• Direct money access detected (medium risk: 0.30). The skill documents building embedded Claude agents in a Next.js application, including reading and changing caller records via app-provided tools (e.g., invoices). However, it does not perform or establish direct financial operations such as transfers, charges, or trades; financial interactions mentioned are indirect (e.g., listing invoices or building usage/billing prototypes). Therefore, it falls under indirect financial scope or general app building.