Loading ToolBench
Analisa prints de WhatsApp para detectar golpes. Use APENAS a ferramenta analisar_fraude.
Single tool with critical definition gaps. The tool 'analisar_fraude' has a basic description and minimal input schema. Schema is present but trivial (only one string parameter with generic description). No output schema documented. No error handling guidance. No parameter constraints or validation hints. The description is adequate in length (144 chars, within 10-1024 range) but lacks actionable context for LLM selection (does not explain WHEN to use it vs alternatives, what prerequisites exist, what the return structure contains). Schema shows only a basic string type with minimal description ('Texto extraido do print'). No enum constraints, no format specifications, no validation rules. Output structure is inferred from code (returns dict with 'status' and 'analise' keys) but not formally documented in tool schema. Parameter description does not explain the expected format, structure, or constraints of the text input.
Analisa print de WhatsApp para detectar fraudes. Extraia o texto da imagem e passe para esta ferramenta. Chama o prompt de fraude da OpenAI.
No output schema documented. Tool returns a dict with 'status' and 'analise' fields, but this structure is not formally declared in the tool definition. LLMs cannot plan downstream operations or validate responses without knowing the output shape.
Parameter description is generic and lacks validation context. 'Texto extraido do print' does not explain format requirements, expected length, character encoding, or constraints. Should specify: minimum/maximum length, expected structure, character restrictions, and what constitutes valid input.
Tool description does not answer critical questions: WHEN to use this tool instead of alternatives? What are the prerequisites? What does it return and in what format? Without this context, LLMs struggle with tool selection and output interpretation.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 42 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 29 | - | v1 |
No error handling guidance. Tool calls external OpenAI API but provides no guidance on handling failures: API timeouts, invalid prompt IDs, malformed responses, authentication errors. LLM cannot recover from failures without explicit error messages and recovery steps.
Tool name 'analisar_fraude' does not start with an action verb in English. While Portuguese naming is acceptable, the tool should use a clearer verb: 'detect_fraud', 'analyze_fraud', or 'check_fraud' (in English) or 'analisar_fraude' with consistent English-verb pattern if using Portuguese. Current name is acceptable but not optimized for multilingual LLM understanding.