Loading ToolBench
Multi-tenant payment gateway for API monetization with autonomous agent settlement on Base L2. Implements HTTP 402 protocol with USDC micropayments and programmatic fee routing.
Single tool with critical security and design flaws. Tool name 'route_x402_payment' lacks action verb clarity (should be 'execute_payment' or 'send_payment'). Description is adequate (97 chars) but does not warn of irreversible consequences or mention the 1% split behavior clearly. Input schema present but parameters lack descriptions in the actual ListToolsRequestSchema response, only Zod schema has descriptions, which LLMs cannot see. No output schema documented. Tool exposes private keys as parameters (CRITICAL SECURITY VIOLATION). No error handling guidance. Composition is monolithic: payment routing, USDC transfer, and proxy settlement bundled into one tool.
Executes an automatic 1% split utility payment on Base L2 and settles access verification via the X402 Cloudflare proxy.
Private key exposed as tool parameter. Credentials must never appear in tool parameters, use server-side secret injection via environment variables or vault. Agent traces log every parameter; secrets leak into logs and prompt history.
Input schema in ListToolsRequestSchema lacks parameter descriptions. Only the Zod schema has descriptions; the JSON Schema returned to the LLM has bare type declarations. LLMs cannot infer that 'privateKey' must be 64-char hex or that 'amountUSD' is a string representation, not a number.
No output schema documented. LLM cannot plan downstream actions or extract transaction hash, settlement status, or error details. Incomplete response contract.
Tool name 'route_x402_payment' is vague. Does it route, execute, send, or settle? Should start with clear action verb: 'execute_payment', 'send_payment', or 'settle_payment'. Current name does not convey irreversible side effects.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 41 | 2026-07-28+ | v2 |
No error handling guidance. Tool can fail (invalid key, insufficient balance, network error, proxy rejection) but provides no recovery hints. LLM receives raw errors with no actionable next steps.