Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add instructions/validpilot/ai-verify-mcp/agents-mdgit clone --depth 1 https://github.com/validpilot/ai-verify-mcpWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.01993 | $0.01993 |
| Opus 5 | $0.00996 | $0.00996 |
| Sonnet 5 | $0.00399 | $0.00399 |
| Haiku 4.5 | $0.00199 | $0.00199 |
Grade A, and why
ai-verify-mcp AGENTS.md scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 190 lines — stays where its author put it; the contents beside it link to each section on GitHub.
AI 代理指南
本文档为在 ai-verify-mcp 项目上工作的 AI 编程代理(Claude、Cursor、Trae、Copilot 等)提供指导。
项目概述
ai-verify-mcp 是一个 MCP(模型上下文协议)服务器,为 AI 代理提供 136 个浏览器验证工具(v1.9.3)。它使 AI 代理能够通过标准化的 MCP 接口执行端到端的 Web 验证、调试和自动化修复。
核心技术栈:Node.js + Playwright + @modelcontextprotocol/sdk
架构
server.js # 主 MCP 服务器(所有工具处理器位于 switch 语句中)
tools/ # 136 个 JSON schema 文件(每个工具一个)
handlers/ # 工具处理器实现(browser/security/validation/exploration 等)
engines/ # Playwright / Chrome 适配器引擎
core/ # 核心工具(产物、配置、安全、脱敏、报告)
hands/ # 高级操作器(browser_operator、evidence_collector、verification_runner)
brain/ # AI 逻辑(error_aggregator)
rules/ # 验证规则(suggested-rules.json)
docs/ # 用户文档
examples/ # 演示示例
bin/ # CLI 入口点
test/ # 单元测试(npm test 运行)
test-reports/ # 测试报告(含 v1.8.0 全量测试报告)
核心约定
-
工具 schema 与实现必须完全匹配 —
tools/*.json中的每个工具在server.js的 switch 语句中都必须有对应的处理器。参数、类型和描述必须保持一致。 -
安全第一 — 切勿在工具输出中暴露 API 密钥、令牌、密码或 Cookie。使用
core/redaction.js中的脱敏工具。browser_eval工具存在已知安全风险,需要仔细进行输入验证。 -
敏感数据脱敏 — 所有可能包含敏感数据的工具输出(Cookie、存储、带认证头的网络响应)在返回给 MCP 客户端之前必须经过脱敏处理。
-
错误处理 — 禁止空 catch 块。始终记录错误。HTTP 服务器未经认证不得暴露于公共网络。
-
会话管理 — 浏览器会话必须能够完全关闭,以防止资源泄漏。池操作必须处理竞态条件以确保线程安全。
-
日志数组边界控制 — 日志数组必须有大小限制,以防止内存泄漏。
-
Schema 命名 — 使用
inputSchema(驼峰命名),而非input_schema(下划线命名),以与 MCP SDK 约定保持一致。 -
CLI 参数 — 对于敏感配置,优先使用环境变量而非命令行参数。
代码风格
- 所有 JS 文件顶部使用
'use strict'; - 2 空格缩进
- 字符串使用单引号
- 行尾不使用分号? — 不,使用分号(参见现有代码)
- 优先使用
const>let>var - 所有异步操作使用 async/await
添加新工具
- 创建
tools/<tool_name>.json,包含完整的 JSON schema(name、description、inputSchema) - 将工具对象添加到
server.js的tools数组中 - 在工具处理器 switch 语句中添加
case '<tool_name>':代码块 - 更新 README.md 和 docs/USER-MANUAL.md 中的工具数量
- 如适用,添加测试或验证脚本
- 更新 CHANGELOG.md
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 2d ago First seen · 190 lines · 1,993 tokens per session scan A 9293e4b51ec5
ai-verify-mcp AGENTS.md is an instructions file published in the GitHub repository validpilot/ai-verify-mcp (0 stars, last pushed 24d ago), licensed MIT. It adds 1,993 tokens to every session, about $0.0100 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other instructions, from other repositories
mcpscore AGENTS.md
Instructions for mcp-box/mcpscore, covering agent instructions for mcpscore and gotchas (learned the hard way).
Genesis-World-MCP AGENTS.md
AGENTS.md instructions for Gnidreve/Genesis-World-MCP, covering agents.md — cas genesisworld mcp, what this is, standing orders — read first, design pillars and source layout & conventions.
youtube-studio-mcp AGENTS.md
AGENTS.md instructions for hygef-v4/youtube-studio-mcp: When the user requests software development, software architecture, new features, or modifications, ALWAYS read and follow CLAUDE.md to execute the AI-DLC workflow.
atelier-mcp copilot-instructions.md
Copilot instructions for anshrajore/atelier-mcp, covering run npm run sync-adapters to update, universal ui/ux rules, universal backend architecture rules and framework presets (available in skills/atelier/presets/).
ssd-ai update-docs-on-code-change.instructions.md
Automatically update README.md and documentation files when application code changes require documentation updates.
ssd-ai AGENTS.md
Instructions for ssdeanx/ssd-ai, covering visualization and src/ structure diagram.