2026-03-03 06:32:50 网络安全文章来源：ZONE.CI 全球网 0 阅读模式

文章总结： NeuroSploitv3是一个AI驱动的自动化渗透测试平台，将自主代理与100种漏洞类型检测、独立Kali容器沙箱、误报加固和漏洞利用链技术相结合。其核心特点包括三流并行测试架构、抗幻觉验证流程、WAF检测绕过及多LLM提供商支持，旨在提升安全评估的自动化程度与结果可靠性。 综合评分： 78 文章分类： 渗透测试,AI安全,安全工具,漏洞分析,解决方案

cover_image

NeuroSploit 一人工智能自动化渗透测试平台

网络安全民工

2026年2月24日 20:58 天津

以下文章来源于蓝星安全，作者蓝星安全

蓝星安全 .

感谢关注蓝星安全，专注分享网络安全、数据安全、人工智能安全等知识，在星球可以获取最新的标准规范、培训课件、解决方案、攻防工具、白皮书、指导手册等，星球资料每天都会进行更新

人工智能驱动的自主渗透测试平台

NeuroSploit v3 是一个高级安全评估平台，它将 AI 驱动的自主代理与 100 种漏洞类型、每次扫描隔离的 Kali Linux 容器、误报加固、漏洞利用链以及具有实时监控功能的现代 React Web 界面相结合。

平台特点

涵盖 10 个类别、100 种漏洞类型，并提供 AI 驱动的测试提示
自主代理– 三流并行渗透测试（侦察 + 初级测试员 + 工具运行器）
每个扫描都使用独立的 Kali 容器——每次扫描都在其自身隔离的 Docker 容器中运行。
抗幻觉流程——阴性对照、执行证明、置信度评分
漏洞利用链引擎– 自动将发现的漏洞关联起来（SSRF->内部漏洞，SQLi->数据库特定漏洞等）
WAF检测与绕过– 16种WAF特征码，12种绕过技术
智能策略调整——死端检测、收益递减、优先级重新计算
多提供商 LLM – Claude、GPT、Gemini、Ollama、LMStudio、OpenRouter
实时仪表盘– 基于 WebSocket 的实时扫描进度、结果和报告
沙箱仪表盘– 实时监控正在运行的 Kali 容器、工具和健康状况

快速入门

选项 1：Docker（推荐）

#&nbsp;Clone repositorygit clone https://github.com/CyberSecurityUP/NeuroSploit/cd&nbsp;NeuroSploitv2#&nbsp;Copy environment file and add your API keyscp .env.example .env nano .env &nbsp;#&nbsp;Add ANTHROPIC_API_KEY, OPENAI_API_KEY, or GEMINI_API_KEY#&nbsp;Build the Kali sandbox image (first time only, ~5 min)./scripts/build-kali.sh#&nbsp;Start backenduvicorn backend.main:app --host 0.0.0.0 --port 8000

选项 2：手动设置

#&nbsp;Backendpip install -r requirements.txt uvicorn backend.main:app --host 0.0.0.0 --port 8000 --reload#&nbsp;Frontend (new terminal)cd&nbsp;frontend npm install npm run dev

构建 Kali 沙箱镜像

#&nbsp;Normal build (uses Docker cache)./scripts/build-kali.sh#&nbsp;Full rebuild (no cache)./scripts/build-kali.sh --fresh#&nbsp;Build + run health check./scripts/build-kali.sh --test#&nbsp;Or via docker-composedocker compose -f docker/docker-compose.kali.yml build

可通过http://localhost:8000（生产版本）或http://localhost:5173（开发模式）访问 Web 界面。

平台结构

NeuroSploitv3/
├── backend/ &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; # FastAPI Backend
│ &nbsp; ├── api/v1/ &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;# REST API (13 routers)
│ &nbsp; │ &nbsp; ├── scans.py &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; # Scan CRUD + pause/resume/stop
│ &nbsp; │ &nbsp; ├── agent.py &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; # AI Agent control
│ &nbsp; │ &nbsp; ├── agent_tasks.py &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; # Scan task tracking
│ &nbsp; │ &nbsp; ├── dashboard.py &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; # Stats + activity feed
│ &nbsp; │ &nbsp; ├── reports.py &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; # Report generation (HTML/PDF/JSON)
│ &nbsp; │ &nbsp; ├── scheduler.py &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; # Cron/interval scheduling
│ &nbsp; │ &nbsp; ├── vuln_lab.py &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;# Per-type vulnerability lab
│ &nbsp; │ &nbsp; ├── terminal.py &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;# Terminal agent (10 endpoints)
│ &nbsp; │ &nbsp; ├── sandbox.py &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; # Sandbox container monitoring
│ &nbsp; │ &nbsp; ├── targets.py &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; # Target validation
│ &nbsp; │ &nbsp; ├── prompts.py &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; # Preset prompts
│ &nbsp; │ &nbsp; ├── vulnerabilities.py &nbsp; &nbsp; &nbsp; # Vulnerability management
│ &nbsp; │ &nbsp; └── settings.py &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;# Runtime settings
│ &nbsp; ├── core/
│ &nbsp; │ &nbsp; ├── autonomous_agent.py &nbsp; &nbsp; &nbsp;# Main AI agent (~7000 lines)
│ &nbsp; │ &nbsp; ├── vuln_engine/ &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; # 100-type vulnerability engine
│ &nbsp; │ &nbsp; │ &nbsp; ├── registry.py &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;# 100 VULNERABILITY_INFO entries
│ &nbsp; │ &nbsp; │ &nbsp; ├── payload_generator.py # 526 payloads across 95 libraries
│ &nbsp; │ &nbsp; │ &nbsp; ├── ai_prompts.py &nbsp; &nbsp; &nbsp; &nbsp;# Per-vuln AI decision prompts
│ &nbsp; │ &nbsp; │ &nbsp; ├── system_prompts.py &nbsp; &nbsp;# 12 anti-hallucination prompts
│ &nbsp; │ &nbsp; │ &nbsp; └── testers/ &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; # 10 category tester modules
│ &nbsp; │ &nbsp; ├── validation/ &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;# False-positive hardening
│ &nbsp; │ &nbsp; │ &nbsp; ├── negative_control.py &nbsp;# Benign request control engine
│ &nbsp; │ &nbsp; │ &nbsp; ├── proof_of_execution.py # Per-type proof checks (25+ methods)
│ &nbsp; │ &nbsp; │ &nbsp; ├── confidence_scorer.py # Numeric 0-100 scoring
│ &nbsp; │ &nbsp; │ &nbsp; └── validation_judge.py &nbsp;# Sole authority for finding approval
│ &nbsp; │ &nbsp; ├── request_engine.py &nbsp; &nbsp; &nbsp; &nbsp;# Retry, rate limit, circuit breaker
│ &nbsp; │ &nbsp; ├── waf_detector.py &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;# 16 WAF signatures + bypass
│ &nbsp; │ &nbsp; ├── strategy_adapter.py &nbsp; &nbsp; &nbsp;# Mid-scan strategy adaptation
│ &nbsp; │ &nbsp; ├── chain_engine.py &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;# 10 exploit chain rules
│ &nbsp; │ &nbsp; ├── auth_manager.py &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;# Multi-user auth management
│ &nbsp; │ &nbsp; ├── xss_context_analyzer.py &nbsp;# 8-context XSS analysis
│ &nbsp; │ &nbsp; ├── poc_generator.py &nbsp; &nbsp; &nbsp; &nbsp; # 20+ per-type PoC generators
│ &nbsp; │ &nbsp; ├── execution_history.py &nbsp; &nbsp; # Cross-scan learning
│ &nbsp; │ &nbsp; ├── access_control_learner.py # Adaptive BOLA/BFLA/IDOR learning
│ &nbsp; │ &nbsp; ├── response_verifier.py &nbsp; &nbsp; # 4-signal response verification
│ &nbsp; │ &nbsp; ├── agent_memory.py &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;# Bounded dedup agent memory
│ &nbsp; │ &nbsp; └── report_engine/ &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; # OHVR report generator
│ &nbsp; ├── models/ &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;# SQLAlchemy ORM models
│ &nbsp; ├── db/ &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;# Database layer
│ &nbsp; ├── config.py &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;# Pydantic settings
│ &nbsp; └── main.py &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;# FastAPI app entry
│
├── core/ &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;# Shared core modules
│ &nbsp; ├── llm_manager.py &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; # Multi-provider LLM routing
│ &nbsp; ├── sandbox_manager.py &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; # BaseSandbox ABC + legacy shared sandbox
│ &nbsp; ├── kali_sandbox.py &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;# Per-scan Kali container manager
│ &nbsp; ├── container_pool.py &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;# Global container pool coordinator
│ &nbsp; ├── tool_registry.py &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; # 56 tool install recipes for Kali
│ &nbsp; ├── mcp_server.py &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;# MCP server (12 tools, stdio)
│ &nbsp; ├── scheduler.py &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; # APScheduler scan scheduling
│ &nbsp; └── browser_validator.py &nbsp; &nbsp; &nbsp; &nbsp; # Playwright browser validation
│
├── frontend/ &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;# React + TypeScript Frontend
│ &nbsp; ├── src/
│ &nbsp; │ &nbsp; ├── pages/
│ &nbsp; │ &nbsp; │ &nbsp; ├── HomePage.tsx &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; # Dashboard with stats
│ &nbsp; │ &nbsp; │ &nbsp; ├── AutoPentestPage.tsx &nbsp; &nbsp; &nbsp;# 3-stream auto pentest
│ &nbsp; │ &nbsp; │ &nbsp; ├── VulnLabPage.tsx &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;# Per-type vulnerability lab
│ &nbsp; │ &nbsp; │ &nbsp; ├── TerminalAgentPage.tsx &nbsp; &nbsp;# AI terminal chat
│ &nbsp; │ &nbsp; │ &nbsp; ├── SandboxDashboardPage.tsx # Container monitoring
│ &nbsp; │ &nbsp; │ &nbsp; ├── ScanDetailsPage.tsx &nbsp; &nbsp; &nbsp;# Findings + validation
│ &nbsp; │ &nbsp; │ &nbsp; ├── SchedulerPage.tsx &nbsp; &nbsp; &nbsp; &nbsp;# Cron/interval scheduling
│ &nbsp; │ &nbsp; │ &nbsp; ├── SettingsPage.tsx &nbsp; &nbsp; &nbsp; &nbsp; # Configuration
│ &nbsp; │ &nbsp; │ &nbsp; └── ReportsPage.tsx &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;# Report management
│ &nbsp; │ &nbsp; ├── components/ &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;# Reusable UI components
│ &nbsp; │ &nbsp; ├── services/api.ts &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;# API client layer
│ &nbsp; │ &nbsp; └── types/index.ts &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; # TypeScript interfaces
│ &nbsp; └── package.json
│
├── docker/
│ &nbsp; ├── Dockerfile.kali &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;# Multi-stage Kali sandbox (11 Go tools)
│ &nbsp; ├── Dockerfile.sandbox &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; # Legacy Debian sandbox
│ &nbsp; ├── Dockerfile.backend &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; # Backend container
│ &nbsp; ├── Dockerfile.frontend &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;# Frontend container
│ &nbsp; ├── docker-compose.kali.yml &nbsp; &nbsp; &nbsp;# Kali sandbox build
│ &nbsp; └── docker-compose.sandbox.yml &nbsp; # Legacy sandbox
│
├── config/config.json &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; # Profiles, tools, sandbox, MCP
├── data/
│ &nbsp; ├── vuln_knowledge_base.json &nbsp; &nbsp; # 100 vuln type definitions
│ &nbsp; ├── execution_history.json &nbsp; &nbsp; &nbsp; # Cross-scan learning data
│ &nbsp; └── access_control_learning.json # BOLA/BFLA adaptive data
│
├── scripts/
│ &nbsp; └── build-kali.sh &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; # Build/rebuild Kali image
├── tools/
│ &nbsp; └── benchmark_runner.py &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;# 104 CTF challenges
├── agents/base_agent.py &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; # BaseAgent class
├── neurosploit.py &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; # CLI entry point
└── requirements.txt

自主代理

AI代理（autonomous_agent.py）自主地协调整个渗透测试。

并行架构

&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; ┌─────────────────────┐
&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; │ &nbsp; Auto Pentest &nbsp; &nbsp; &nbsp;│
&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; │ &nbsp; Target URL(s) &nbsp; &nbsp; │
&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; └────────┬────────────┘
&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;│
&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; ┌──────────────┼──────────────┐
&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; ▼ &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;▼ &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;▼
&nbsp; &nbsp;┌──────────────┐ ┌──────────────┐ ┌──────────────┐
&nbsp; &nbsp;│ &nbsp;Stream 1 &nbsp; &nbsp;│ │ &nbsp;Stream 2 &nbsp; &nbsp;│ │ &nbsp;Stream 3 &nbsp; &nbsp;│
&nbsp; &nbsp;│ &nbsp;Recon &nbsp; &nbsp; &nbsp; │ │ &nbsp;Junior Test │ │ &nbsp;Tool Runner │
&nbsp; &nbsp;│ &nbsp;─────────── │ │ &nbsp;─────────── │ │ &nbsp;─────────── │
&nbsp; &nbsp;│ &nbsp;Crawl pages │ │ &nbsp;Test target │ │ &nbsp;Nuclei scan │
&nbsp; &nbsp;│ &nbsp;Find params │ │ &nbsp;AI-priority │ │ &nbsp;Naabu ports │
&nbsp; &nbsp;│ &nbsp;Tech detect │ │ &nbsp;3 payloads &nbsp;│ │ &nbsp;AI decides &nbsp;│
&nbsp; &nbsp;│ &nbsp;WAF detect &nbsp;│ │ &nbsp;per endpoint│ │ &nbsp;extra tools │
&nbsp; &nbsp;└──────┬───────┘ └──────┬───────┘ └──────┬───────┘
&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; │ &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;│ &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;│
&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; └────────────────┼────────────────┘
&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;▼
&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; ┌─────────────────────┐
&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; │ &nbsp;Deep Analysis &nbsp; &nbsp; &nbsp;│
&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; │ &nbsp;100 vuln types &nbsp; &nbsp; │
&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; │ &nbsp;Full payload sets &nbsp;│
&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; │ &nbsp;Chain exploitation │
&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; └─────────┬───────────┘
&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; ▼
&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; ┌─────────────────────┐
&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; │ &nbsp;Report Generation &nbsp;│
&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; │ &nbsp;AI executive brief │
&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; │ &nbsp;PoC code per find &nbsp;│
&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; └─────────────────────┘

代理自主模块

扫描功能

暂停/恢复/停止（带检查点）
人工验证– 确认或拒绝人工智能的发现
已确认结果的屏幕截图（剧作家）
交叉扫描学习——历史成功率影响未来优先事项
CVE 测试– 正则表达式检测 + AI 生成的有效载荷

100 种漏洞类型

类别

| 类别 | 类型 | 示例 | | — | — | — | | 注射 | 38 | XSS（反射型/存储型/DOM）、SQL注入、NoSQL注入、命令注入、SSTI、LDAP、XPath、CRLF、标头注入、日志注入、GraphQL注入 | | 检查 | 21 | 安全标头、CORS、点击劫持、信息泄露、调试端点、错误泄露、源代码暴露 | | 人工智能驱动 | 41 | BOLA、BFLA、IDOR、竞态条件、业务逻辑、JWT 操纵、OAuth 缺陷、原型污染、WebSocket 劫持、缓存投毒、HTTP 请求走私 | | 验证 | 8 | 身份验证绕过、会话固定、凭据填充、密码重置漏洞、多因素身份验证绕过、默认凭据 | | 授权 | 6 | BOLA、BFLA、IDOR、权限提升、强制浏览、功能级访问控制 | | 文件访问 | 5 | LFI、RFI、路径遍历、文件上传、XXE | | 请求伪造 | 4 | SSRF、CSRF、云元数据、DNS 重绑定 | | 客户端 | 8 | CORS、点击劫持、开放重定向、DOM 覆盖、原型污染、PostMessage、CSS 注入 | | 基础设施 | 6 | SSL/TLS、HTTP 方法、子域名劫持、主机头、CNAME 劫持 | | 云/供应 | 4 | 云元数据、S3存储桶配置错误、依赖关系混乱、第三方脚本 |

有效载荷引擎

95 个库中共有526 个有效载荷
73 个 XSS 存储有效载荷+ 5 个特定于上下文的集合
针对不同类型人工智能决策的提示，并带有防幻觉指令
WAF自适应有效载荷转换（12种技术）

Kali 沙箱系统

每次扫描都在其独立的 Kali Linux Docker 容器中运行，提供以下功能：

完全隔离——同时进行的扫描之间互不干扰
按需工具– 56 种工具，仅在需要时安装
自动清理– 扫描完成后销毁容器
资源限制– 每个容器的内存（2GB）和 CPU（2 核）限制

预装工具（28）

| 类别 | 工具 | | — | — | | 扫描仪 | 核、naabu、httpx、nmap、nikto、masscan、whatweb | | 发现 | subfinder、katana、dnsx、uncover、ffuf、gobuster、waybackurls | | 开发 | dalfox，sqlmap | | 系统 | curl、wget、git、python3、pip3、go、jq、dig、whois、openssl、netcat、bash |

按需工具（还有 28 种）

首次请求时自动安装在容器内：

APT：wpscan、dirb、hydra、john、hashcat、testssl、sslscan、enum4linux、dnsrecon、amass、medusa、crackmapexec 等。
Go：gau、gitleaks、anew、httprobe
Pip：dirsearch、wfuzz、arjun、wafw00f、sslyze、commix、trufflehog、退休

容器池

ContainerPool (global coordinator, max 5 concurrent)
&nbsp; ├── KaliSandbox(scan_id="abc") → docker: neurosploit-abc
&nbsp; ├── KaliSandbox(scan_id="def") → docker: neurosploit-def
&nbsp; └── KaliSandbox(scan_id="ghi") → docker: neurosploit-ghi

TTL强制执行– 容器在60分钟后自动销毁
孤立容器清理– 服务器启动时移除过时的容器
优雅回退– 如果 Docker 不可用，则回退到共享容器

抗幻觉与验证

NeuroSploit 使用多层验证流程来消除误报：

验证流程

Finding Candidate
&nbsp; &nbsp; │
&nbsp; &nbsp; ▼
┌─────────────────────┐
│ Negative Controls &nbsp; &nbsp;│ &nbsp;Send benign/empty requests as controls
│ Same behavior = FP &nbsp; │ &nbsp;-60 confidence if same response
└─────────┬───────────┘
&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; ▼
┌─────────────────────┐
│ Proof of Execution &nbsp; │ &nbsp;25+ per-vuln-type proof methods
│ XSS: context check &nbsp; │ &nbsp;SSRF: metadata markers
│ SQLi: DB errors &nbsp; &nbsp; &nbsp; │ &nbsp;BOLA: data comparison
└─────────┬───────────┘
&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; ▼
┌─────────────────────┐
│ AI Interpretation &nbsp; &nbsp;│ &nbsp;LLM with anti-hallucination prompts
│ Per-type system msgs │ &nbsp;12 composable prompt templates
└─────────┬───────────┘
&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; ▼
┌─────────────────────┐
│ Confidence Scorer &nbsp; &nbsp;│ &nbsp;0-100 numeric score
│ ≥90 = confirmed &nbsp; &nbsp; &nbsp;│ &nbsp;+proof, +impact, +controls
│ ≥60 = likely &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;│ &nbsp;-baseline_only, -same_behavior
│ <60 = rejected &nbsp; &nbsp; &nbsp; &nbsp;│ &nbsp;Breakdown visible in UI
└─────────┬───────────┘
&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; ▼
┌─────────────────────┐
│ Validation Judge &nbsp; &nbsp; │ &nbsp;Final verdict authority
│ approve / reject &nbsp; &nbsp; │ &nbsp;Records for adaptive learning
└─────────────────────┘

抗幻觉系统提示

12 个可组合的提示，适用于 7 个任务场景：

anti_hallucination– 核心诚实准则
proof_of_execution需要确凿证据
negative_controls– 与良性请求进行比较
anti_severity_inflation– 准确的严重程度评级
access_control_intelligence– BOLA/BFLA 数据比较方法

访问控制自适应学习

记录 BOLA/BFLA/IDOR 各领域的 TP/FP 结果
9 种默认响应模式，6 种已知的 FP 模式（WSO2、Keycloak 等）
历史假阳性率会影响未来的置信度评分

Web 图形用户界面

沙盒仪表盘

对每次扫描的 Kali 容器进行实时监控：

池统计信息– 活动/最大容器数、Docker 状态、TTL
容量条– 可视化利用率指示器
每个容器的卡片– 名称、扫描链接、运行时间、已安装工具、状态
操作– 健康检查、销毁（需确认）、清理过期/孤立文件
5 秒自动轮询，实现实时更新

API 参考

基本 URL

http://localhost:8000/api/v1

终点

扫描

| 方法 | 端点 | 描述 | | — | — | — | | POST | /scans | 创建新扫描 | | GET | /scans | 列出所有扫描结果 | | GET | /scans/{id} | 获取扫描详情 | | POST | /scans/{id}/start | 开始扫描 | | POST | /scans/{id}/stop | 停止扫描 | | POST | /scans/{id}/pause | 暂停扫描 | | POST | /scans/{id}/resume | 恢复扫描 | | DELETE | /scans/{id} | 删除扫描 |

人工智能代理

| 方法 | 端点 | 描述 | | — | — | — | | POST | /agent/run | 启动自主代理 | | GET | /agent/status/{id} | 获取代理状态和调查结果 | | GET | /agent/by-scan/{scan_id} | 扫描ID获取代理 | | POST | /agent/stop/{id} | 停止代理 | | POST | /agent/pause/{id} | 暂停代理 | | POST | /agent/resume/{id} | 简历代理 | | GET | /agent/findings/{id} | 获取详细信息 | | GET | /agent/logs/{id} | 获取代理日志 |

沙盒

| 方法 | 端点 | 描述 | | — | — | — | | GET | /sandbox | 列出容器和池状态 | | GET | /sandbox/{scan_id} | 健康检查容器 | | DELETE | /sandbox/{scan_id} | 销毁容器 | | POST | /sandbox/cleanup | 移除过期容器 | | POST | /sandbox/cleanup-orphans | 移除孤立容器 |

调度程序

| 方法 | 端点 | 描述 | | — | — | — | | GET | /scheduler | 列出已安排的作业 | | POST | /scheduler | 创建计划作业 | | DELETE | /scheduler/{id} | 删除职位 | | POST | /scheduler/{id}/pause | 暂停作业 | | POST | /scheduler/{id}/resume | 简历工作 |

漏洞实验室

| 方法 | 端点 | 描述 | | — | — | — | | GET | /vuln-lab/types | 按类别列出 100 种漏洞类型 | | POST | /vuln-lab/run | 运行按类型漏洞测试 | | GET | /vuln-lab/challenges | 列表挑战运行 | | GET | /vuln-lab/stats | 检测率统计数据 |

报告和仪表盘

| 方法 | 端点 | 描述 | | — | — | — | | POST | /reports | 生成报告 | | POST | /reports/ai-generate | 人工智能驱动的报告 | | GET | /reports/{id}/view | 查看 HTML 报告 | | GET | /dashboard/stats | 仪表盘统计信息 | | GET | /dashboard/activity-feed | 近期活动 |

WebSocket

ws://localhost:8000/ws/scan/{scan_id}

事件：scan_started，，，，progress_update``finding_discovered``scan_completed``scan_error

API 文档

交互式文档可在以下网址获取：

Swagger UI：http://localhost:8000/api/docs
ReDoc：http://localhost:8000/api/redoc

配置

环境变量

#&nbsp;LLM API Keys (at least one required)ANTHROPIC_API_KEY=your-key OPENAI_API_KEY=your-key GEMINI_API_KEY=your-key#&nbsp;Local LLM (optional)OLLAMA_BASE_URL=http://localhost:11434 LMSTUDIO_BASE_URL=http://localhost:1234 OPENROUTER_API_KEY=your-key#&nbsp;DatabaseDATABASE_URL=sqlite+aiosqlite:///./data/neurosploit.db#&nbsp;ServerHOST=0.0.0.0 PORT=8000 DEBUG=false

config/config.json

{ &nbsp;"llm": { &nbsp; &nbsp;"default_profile":&nbsp;"gemini_pro_default", &nbsp; &nbsp;"profiles": {&nbsp;...&nbsp;} &nbsp; }, &nbsp;"agent_roles": { &nbsp; &nbsp;"pentest_generalist": {&nbsp;"vuln_coverage":&nbsp;100&nbsp;}, &nbsp; &nbsp;"bug_bounty_hunter": {&nbsp;"vuln_coverage":&nbsp;100&nbsp;} &nbsp; }, &nbsp;"sandbox": { &nbsp; &nbsp;"mode":&nbsp;"per_scan", &nbsp; &nbsp;"kali": { &nbsp; &nbsp; &nbsp;"enabled":&nbsp;true, &nbsp; &nbsp; &nbsp;"image":&nbsp;"neurosploit-kali:latest", &nbsp; &nbsp; &nbsp;"max_concurrent":&nbsp;5, &nbsp; &nbsp; &nbsp;"container_ttl_minutes":&nbsp;60&nbsp; &nbsp; &nbsp;} &nbsp; }, &nbsp;"mcp_servers": { &nbsp; &nbsp;"neurosploit_tools": { &nbsp; &nbsp; &nbsp;"transport":&nbsp;"stdio", &nbsp; &nbsp; &nbsp;"command":&nbsp;"python3", &nbsp; &nbsp; &nbsp;"args": ["-m",&nbsp;"core.mcp_server"] &nbsp; &nbsp; } &nbsp; } }

发展

后端

pip install -r requirements.txt uvicorn backend.main:app --reload --host 0.0.0.0 --port 8000#&nbsp;API docs: http://localhost:8000/api/docs

前端

cd&nbsp;frontend npm install npm run dev &nbsp; &nbsp; &nbsp; &nbsp;#&nbsp;Dev server at http://localhost:5173npm run build &nbsp; &nbsp; &nbsp;#&nbsp;Production build

构建 Kali 沙盒

./scripts/build-kali.sh --test &nbsp; &nbsp;#&nbsp;Build + health check

MCP 服务器

python3 -m core.mcp_server &nbsp; &nbsp; &nbsp; &nbsp;#&nbsp;Starts stdio MCP server (12 tools)

项目地址：

https://github.com/CyberSecurityUP/NeuroSploit

一个专注网络数据安全的知识星球，星球提供以下资料：网络数据安全标准规范、制度体系、培训课件、攻防工具教程、解决方案、行业报告等人工智能安全标准规范、研究报告、培训材料等，每天都会更新

免责声明：

本文所载程序、技术方法仅面向合法合规的安全研究与教学场景，旨在提升网络安全防护能力，具有明确的技术研究属性。

任何单位或个人未经授权，将本文内容用于攻击、破坏等非法用途的，由此引发的全部法律责任、民事赔偿及连带责任，均由行为人独立承担，本站不承担任何连带责任。

本站内容均为技术交流与知识分享目的发布，若存在版权侵权或其他异议，请通过邮件联系处理，具体联系方式可点击页面上方的联系我。

本文转载自：网络安全民工《NeuroSploit 一人工智能自动化渗透测试平台》