hex-tools/hexstrike-ai-pc

选用指南

AI 根据 README 整理 · 2026/10/28 05:16:30

使用方式

通过命令行使用，可编辑 engagement.json 配置文件。

查看 README 依据

[RPT] Report generated: findings 4 | severity: 1 high, 2 med, 1 low exec summary + remediation: 12 pages, client-ready ✓ [TIME] Engagement: 47 minutes | manual estimate: 4–6 hours | the "I'm in" ratio: ↑ ``` The .git exposure was found in minutes, the chain was explained with reasoning, and the report wrote itself. You still did the thinking — you just didn't do the typing. --- ## 📋 Example config — the framework, in file form What HexStrike saves to `engagement.json` (editable in the CLI): ```json {

部署要求

需要自带 LLM 模型的 API 密钥，本地优先设计，不依赖云端。

查看 README 依据

📝 **Report generation** — the killer feature: findings auto-compiled into a clean, client-ready report: executive summary, technical details, severity ratings, remediation advice. The most hated 20 hours of pentesting, compressed. 🎯 **CTF & bug bounty modes** — structured workflows for CTF boxes (HTB, THM, VulnHub) and bug bounty scopes: flag tracking, scope-guard rails, retainer-style session logs. 🔒 **Local-first design** — your engagement data stays on your box; bring-your-own model keys for the LLM layer. No engagement details shipped to someone else's cloud. ---

"target": { "host": "lab-10.10.14.x", "authorization": "written_on_file", "scope_guard": true }, "agents": { "recon": { "tools": ["nmap", "masscan"], "profile": "stealth" }, "enumeration": { "web": ["gobuster", "nikto"], "deep_mode": true }, "vuln_analysis": { "chain_correlation": true, "confidence_threshold": 80 }, "exploitation": { "guidance_mode": "reasoned_poc", "auto_exploit": false }, "report": { "format": "md+pdf", "severity": "cvss", "remediation": true } }, "llm": { "provider": "your_key", "local_first": true, "no_cloud_uploads": true },

已知限制

AI 发现的结果需要人工验证，不能直接用于报告；仅限授权测试。

查看 README 依据

- 📄 **Authorization is the entire game.** Written permission, defined scope, or dedicated lab environments (HTB, THM, VulnHub, CTF). Scanning systems you don't have permission to test is a crime in most jurisdictions — the tool doesn't change that, and scope_guard exists for a reason - 🧠 **AI findings ≠ verified findings.** LLMs hallucinate confidently: every agent conclusion needs human validation before it touches a real report. HexStrike accelerates your analysis — it doesn't replace it