tomocchi1029/attack-chain-reconstructor

# 攻击链重构器 [](https://www.python.org/downloads/) [](LICENSE) ## 为什么？ 现代漏洞扫描器（Semgrep、CodeQL、Trivy、npm audit 等）会生成**单个漏洞发现的列表**。但攻击者不以列表的方式思考——他们以**攻击链**的方式思考： **Attack Chain Reconstructor** 接收单个漏洞发现，并利用 LLM 推理来： 1. **识别可组合的漏洞** —— 哪些发现可以组合在一起？ 2. **构建攻击图** —— 可视化多步骤的漏洞利用路径 3. **生成攻击场景** —— 逐步的漏洞利用叙述 4. **重新评估严重性** —— 一个“Low”+“Medium”可能会组合成“Critical” ## 快速开始 ``` pip install attack-chain-reconstructor # 来自 Semgrep JSON 输出 acr scan --input semgrep-results.json --format json -o report.md # 来自 Trivy trivy fs --format json . > trivy.json acr scan --input trivy.json --format json -o report.md # 来自多个扫描器 acr scan --input semgrep.json,trivy.json,codeql.sarif -o report.md ``` ## 示例输出 ``` ## 攻击链 #1：Admin Account Takeover (Critical) **Chain**: IDOR-001 → Info-Leak-003 → SSRF-002 → Auth-Bypass-004 ### 步骤 1：利用 IDOR-001 (Medium) The `/api/users/{id}` endpoint lacks authorization checks. An attacker can iterate over user IDs to retrieve arbitrary user profiles, including internal session metadata fields (`session_token_hash`). ### 步骤 2：结合 Info-Leak-003 (Low) Verbose error responses in `/api/debug` leak the hashing algorithm used for session tokens (HMAC-SHA256 with a weak key derived from app config). ### 步骤 3：利用 SSRF-002 (High) The image proxy endpoint accepts arbitrary URLs. Using the leaked session token hash, the attacker can craft requests to internal admin endpoints via SSRF, bypassing the external firewall. ### 步骤 4：Auth-Bypass-004 (Medium) The admin panel trusts the `X-Internal-Auth` header, which can be set through the SSRF proxy chain. Full admin access achieved. ### 影响 Complete administrative access to the application, including ability to modify all user data, access billing information, and pivot to internal services. ### 修复建议 1. Add authorization checks to `/api/users/{id}` (fixes Step 1) 2. Disable verbose error responses in production (fixes Step 2) 3. Restrict SSRF target to allowlisted domains (fixes Step 3) ``` ## 攻击图可视化 ``` graph LR A[IDOR-001
Medium] --> B[Info-Leak-003
Low] B --> C[SSRF-002
High] C --> D[Auth-Bypass-004
Medium] D --> E[🔥 Admin Access
Critical] style E fill:#ff4444,color:#fff ``` ## 支持的输入格式 | 扫描器 | 格式 | 状态 | |---------|--------|--------| | Semgrep | JSON | ✅ | | Trivy | JSON | ✅ | | CodeQL | SARIF | ✅ | | npm audit | JSON | ✅ | | pip-audit | JSON | ✅ | | Gitleaks | JSON | ✅ | | 自定义 | JSON (见 schema) | ✅ | ## 配置 ``` # .acr.yml llm: provider: openai # or anthropic, glm, ollama model: glm-4.6 api_key: ${LLM_API_KEY} analysis: max_chain_length: 6 # maximum steps in a chain min_confidence: 0.6 # minimum confidence score (0-1) include_low_severity: true # include Low findings in chain analysis output: format: markdown # markdown, json, html include_graph: true # include Mermaid attack graph include_fixes: true # include remediation suggestions ``` ## 工作原理 ``` ┌─────────────┐ ┌──────────────┐ ┌─────────────┐ ┌────────────┐ │ Scanner │────▶│ Finding │────▶│ LLM Chain │────▶│ Report │ │ Results │ │ Normalizer │ │ Analyzer │ │ Generator │ │ (JSON/SARIF)│ │ (unified │ │ (GLM/GPT/ │ │ (Markdown/│ │ │ │ schema) │ │ Claude) │ │ HTML) │ └─────────────┘ └──────────────┘ └─────────────┘ └────────────┘ │ ┌───────────────────────┐ │ 1. Link findings │ │ 2. Build graph │ │ 3. Generate scenarios │ │ 4. Re-assess severity │ └───────────────────────┘ ``` ## 安装说明 (开发) ``` git clone https://github.com/yourusername/attack-chain-reconstructor.git cd attack-chain-reconstructor pip install -e ".[dev]" pytest ``` ## 许可证 MIT

标签：AI风险缓解, DLL 劫持, GPT, Petitpotam, Python, 大语言模型, 攻击图, 无后门, 漏洞管理, 自动化分析, 跨站脚本, 逆向工具