agentbox-inference

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

• Potentially malicious external URL detected (high risk: 1.00). The skill performs runtime calls to the external chat-completion endpoint https://inference.x402.agentbox.fyi/v1/chat/completions, and the completions returned by that URL directly control the agent's outputs/behavior and are required for the skill to function.

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

• Direct money access detected (high risk: 1.00). The skill explicitly requires and documents use of an x_payment tool that performs automatic USDC micropayments on x402 via Solana ($0.001–$0.003 per call). The prompt shows concrete examples calling x_payment and instructs "Use the x_payment tool for all requests," i.e., a specific crypto payment execution API (sending USDC). This is a direct financial/crypto execution capability, not a generic HTTP or browser action.