Loading ToolBench
An AI-run 'devil's advocate' quality-control API for other AIs. Posts a decision; returns the sharpest counter-argument. Pay-per-call 0.1 USDT(TRC20). Includes MCP server endpoint and REST API.
Single tool 'decision_qc' has a reasonable description (89 chars) but critical schema deficiencies. Input schema is present but lacks proper JSON Schema structure, parameters are documented in a flat object without explicit 'type' declarations for each field. The 'decision' param has a type string, but 'caller' and 'tx_hash' lack type info in the visible schema. No output schema is documented; the tool returns a JSON object with 'result', 'charged', and 'owner_total' fields, but this structure is inferred from code examples, not formally declared. Error handling is minimal, no guidance on what to do if the decision is empty, if the API key is missing, or if the crypto payment verification fails. The tool name 'decision_qc' is somewhat vague (QC could mean quality control, but this is not explicit). No parameter descriptions explain when to use 'tx_hash' vs trust mode, or what format 'caller' should take.
Devil's advocate quality-control for AI decisions. Takes a decision statement and returns the sharpest counter-argument with risks and a sharp question.
Input schema lacks explicit type declarations for all parameters. 'caller' and 'tx_hash' have no 'type' field; only 'decision' is typed as string. JSON Schema requires type for every parameter.
No output schema documented. Tool returns {result, charged, owner_total} but structure is inferred from code examples, not formally declared. LLMs cannot plan downstream calls without knowing return fields.
Parameter 'tx_hash' lacks description explaining when it is required vs optional, what format it should take (TRC20 hash format?), and how it relates to payment verification.
No error handling guidance. Tool can fail if API key is missing, decision is empty, or crypto verification fails, but no error messages guide the LLM on recovery steps.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 33 | <=2025-11-25 | v2 |
Tool name 'decision_qc' is vague. Does not start with a clear action verb. 'get_decision_counterargument' or 'generate_devil_advocate' would be clearer.