Loading ToolBench
Adaptador MCP do NOMOS Browser Runtime — traduz tools/call em POST da API v1. Sem lógica de navegador.
Critical gaps across all dimensions. Tool descriptions are in Portuguese and extremely brief (5-15 chars), far below the 10-1024 char baseline and insufficient for LLM selection. Input schemas visible but parameters lack type definitions and descriptions, only high-level object types with minimal guidance. No output schemas documented. Tool names use 'browser.*' prefix but lack clear verb-noun structure (e.g., 'browser.goto' vs 'navigate_to_url'). No error handling guidance, no parameter constraints (enums, ranges), no security considerations. The 'verification' parameter appears across multiple tools but is never explained. The 'browser.task' tool accepts only 'objective' string with no guidance on format or expected behavior. No evidence of idempotency, pagination, or composition patterns. This is a thin adapter layer that translates MCP calls to HTTP POST without LLM-optimized definitions.
Marca um checkbox
Clica em um elemento
Faz download de arquivo
Extrai dados da página
Preenche um campo de formulário
Foca em um elemento
Navega para uma URL
Descriptions are critically short (5-15 chars, all in Portuguese), far below 10-1024 char baseline. Examples: 'Navega para uma URL' (19 chars), 'Clica em um elemento' (20 chars). LLMs cannot determine when to select these tools or what they do beyond a literal translation.
Input parameters lack type definitions and descriptions. 'target' appears in 10+ tools as type 'object' with only 'Descritor de alvo' (5 chars), no guidance on structure, required fields, or how to construct it. 'verification' parameter is unexplained across all tools that use it.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 36 | 2026-07-28+ | v2 |
Passa o mouse sobre um elemento
Envia sequência de teclado
Monitora requisições de rede
Observa mudanças na página
Pressiona uma tecla
Captura uma screenshot da página
Rola a página
Seleciona uma opção em um dropdown
Executa uma tarefa de navegação
Digita texto em um elemento
Desmarca um checkbox
Faz upload de arquivo
Aguarda uma condição
No output schemas documented. LLMs cannot plan downstream tool calls or extract required data (e.g., does browser.extract return JSON, structured object, or string?). No pagination, no field documentation, no chaining IDs.
Tool naming lacks clear verb-noun structure. 'browser.goto', 'browser.click', 'browser.type' are acceptable but 'browser.task' is vague, does it execute a predefined task, create one, or run an arbitrary objective? No distinction between read-only and write operations in names.
No error handling guidance. Tools declare WRITE risk but provide no recovery hints, retryability classification, or actionable error messages. No validation rules for parameters (e.g., what makes a valid 'scope' for screenshot? What are valid 'direction' values for scroll?).