Loading ToolBench
An AI-powered phone calling service that uses Lightning Network payments (L402 protocol) for authentication and billing. Allows users to make automated calls with AI agents to perform tasks like confirmations and inquiries.
Three tools with complete input schemas and descriptions, but significant gaps in output documentation and error handling. Tool names follow verb_noun convention (place_call, poll_call_status, confirm_call_with_answers). Descriptions are present (95-180 chars) and explain WHAT the tools do, but lack guidance on WHEN to use them or error recovery paths. Parameters are typed and mostly described, but missing constraints (e.g., E.164 format validation, duration bounds, enum for action). No documented output schemas, LLMs cannot plan downstream calls or extract required fields. Error handling is absent; no guidance on retryability, user-fixable vs fatal errors, or recovery steps. L402 authentication is a domain-specific concern but not surfaced in tool descriptions.
Submits clarification answers to an AI agent that requested confirmation during a call. Used when call state is pending_confirm.
Initiates an AI-powered phone call with L402 Lightning Network authentication. Requires payment via Lightning invoice before call execution. Supports task instructions, duration, language selection, and optional refund address.
Polls the status of an ongoing or completed AI phone call using its call_id.
No output schemas documented. LLMs cannot infer response structure (fields, types, pagination) or plan downstream calls. Examples show responses like {state, call_id, session_id, questions} but this is not formally specified.
Missing error handling guidance. No documentation of error cases (invalid phone, payment failure, call timeout, confirmation rejection), retryability, or recovery steps. LLMs cannot self-correct on failures.
Parameter constraints missing. phone_number lacks E.164 format validation hint; duration_minutes unbounded (should be 1-60 or similar); action enum exists but not enforced in description; language lacks enum of supported values.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 50 | 2026-07-28+ | v2 |
L402 Lightning authentication is a prerequisite but not surfaced in tool descriptions. LLMs cannot infer that place_call requires out-of-band payment or that a 402 response triggers a multi-step flow.
confirm_call_with_answers has vague 'answers' parameter (type: object, no schema). LLMs cannot know what keys/structure to provide. Should document expected answer format or reference questions from pending_confirm state.