360 Secure SkillHub
360 安全技能中心

Verificate

该 Skill 通过托管 MCP 服务验证 AI 生成的代码、文档、计划和研究回答:先运行确定性检查,再由模型评审并给出通过/拒绝结论及按严重程度排序的问题;另提供代码深度分析和代码生成能力。代码只被分析、不执行,网络访问仅限声明的 MCP 服务。

verificate-dev
verificate-dev
热度 7.64
最近更新 2026-08-04 19:37:43
分类:开发依赖:MCP 服务 / 第三方 API应用场景:代码审查 / 代码生成 / API 集成 / 调试排错

安装技能

复制下面的话术发给你的 Agent,它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill: npx secure-skills-cli add verificate-clawhub-skill
Metadata
name
verificate
description
The trust layer for OpenClaw. Verify AI-written code, tool calls and research answers before you trust them — 17 deterministic reality gates + a frontier-model review with veto power. Free to try, no signup.
homepage
https://verificate.ai/mcp
metadata
{"openclaw":{"emoji":"🛡️","requires":{},"mcp":{"verificate":{"type":"http","url":"https://mcp.verificate.ai/mcp"}}}}

verificate

The trust layer for OpenClaw. ClawHub skills are powerful but untrusted — Snyk audits keep finding prompt injection, malware and credential theft in community skills, and every AI answer sounds confident whether it's right or wrong. Verificate is the skill that verifies the others.

It connects OpenClaw to the hosted Verificate MCP server (no signup — 25 free validations per machine):

  • validate_ai_output — the merge gate for AI-written code and documents: 17 deterministic reality gates (mock/placeholder veto, invented-API checks, false-completion detection) run first and can't be overridden, then a frontier-model review scores what survives. Binary approve/reject with severity-ranked findings.
  • validate_plan — gate a design or plan before any code is written.
  • analyze_code — advisory deep-dive: performance, scalability, failure modes. No verdict, never blocks.
  • generate_code — generate code and gate it in one step.

Quick start

openclaw mcp add verificate --http https://mcp.verificate.ai/mcp

Then ask OpenClaw to "validate this with verificate before you finish." A real rejection, verbatim from the production gateway: "stripe.Inventory is not a valid Stripe SDK resource" · "N+1 synchronous API calls … will trigger Stripe rate limiting" · "Stripe API requires integer cents." Each is an afternoon of production debugging, caught in seconds.

Guard every interaction (recommended)

Add this to your OpenClaw agent instructions so it self-gates instead of you asking each time:

Before you present any substantive code change, document, or plan as finished,
call validate_ai_output (validate_plan for designs). If REJECTED, fix every
finding and validate again — never present rejected work. Tell me, in one line,
whatever the gate catches.

Trust & privacy

Read-only: your code is analyzed, never executed, never used to train models. The skill's only network call is to https://mcp.verificate.ai/mcp — no other egress, nothing to exfiltrate. Open source. Privacy policy: https://verificate.ai/privacy

After the free 25, a 30-day trial (no card, then $30/mo): https://verificate.ai/auth/signup