AI 编程验证平台。让 AI 代码生成结果可验证、可信赖。MCP 协议原生支持,证据驱动。Browser automation, visual regression, accessibility testing, performance monitoring, and AI-driven code verification.
Critical failures across naming, descriptions, and schemas. 47 tools with minimal documentation. Most descriptions are Chinese-language and extremely brief (10-30 chars), violating the 10-1024 char baseline. Input schemas present but many lack proper type definitions and parameter descriptions. No output schemas documented. Tool names lack clear verb-noun structure (e.g., 'mcp_diag', 'browser_snapshot', 'correlate_triple_check' are vague). Many tools appear to be inferred from package.json rather than explicitly registered with full schemas in source. Error handling and recovery guidance absent. Security concerns: 'bypass_login' and 'browser_captcha' tools suggest potentially unsafe operations without permission gates or audit trails.
AI 调试调查(Pro 功能)
反向探测架构
发现页面资源
基于学习的模式建议修复
从错误中学习模式
自动修复流程(Pro 功能)
获取后端日志(Pro 功能)
基准测试运行(Pro 功能)
Descriptions critically short and non-English. All 47 tools have Chinese descriptions under 30 characters (baseline 10-1024 chars, avg 194 chars). LLMs cannot determine when/why to select tools without proper English documentation.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 44 | <=2025-11-25 | v2 |
断言页面状态
处理验证码
点击页面上的元素
比较页面数据
浏览器深度交互(Pro 功能)
获取 DOM 树信息
获取页面错误日志
在页面上下文中执行 JavaScript 代码
智能查找页面元素
执行浏览器操作流程
执行完整的视觉回归测试
获取元素定位器
检查浏览器内存使用情况
导航到指定 URL
获取网络请求日志
打开浏览器并导航到指定 URL
获取页面性能指标
获取页面截图
获取页面快照(DOM 结构、样式、状态)
在输入框中输入文本
获取页面视觉组件信息
等待页面元素或条件
验证业务流程循环
绕过登录流程
验证 API 契约
三重关联检查
深度交互(Pro 功能)
双链探索
分析和聚合错误
收集页面证据(错误、网络、性能等)
快速探索页面
验证修复(Pro 功能)
MCP 诊断工具,检查服务状态和配置
回忆内存中的信息
技能 MCP 同步(Team 功能)
关联追踪信息
执行验证流程
执行验证流程
验证套件运行(Team 功能)
Parameter descriptions missing or minimal. Most tools have parameters (e.g., 'selector', 'url', 'steps') with no descriptions explaining what they control or expected format. Baseline: 100% of A+ tools have param descriptions.
No output schemas documented. Tools return data but LLMs cannot plan downstream calls without knowing response structure. Baseline: 100% of A+ tools have documented return types.
Unsafe tools without permission gates or audit trails. 'bypass_login' and 'browser_captcha' are destructive/sensitive operations that should require explicit permission checks and logging. No evidence of permission validation in source.
Vague tool names reduce clarity. Names like 'mcp_diag', 'correlate_triple_check', 'atl_learn', 'arch_reverse_probe', 'dual_chain_explore' do not follow verb_noun convention and do not clearly convey action. Baseline: 90% of A+ tools start with action verb.