YeepForge - Automated web application penetration testing fram

项目说明

<di="center">

YeepForge Agent

The AI Penetration Tester That Works While You Sleep

capture0x - powered by tmrswrr

![OWASP]()
![Python]()
![Ollama]()

> Give it a URL. Give it source code. Watch it hack.

<img src="assets/1.png" alt="YeepForge" width="800">

</div>

---

What Makes YeepForge Different?

Most security tools are menus of commands you run manually.

YeepForge Agent is different. It's an AI that *thinks like a penetration tester* - it reads the target, decides what to attack next, chains findings together, and delivers a professional C report. You provide a URL. The Agent does the rest.

---

<di="center">

<a id="agent"></a>

The Agent

</div>

How It Thinks

The Agent doesn't run tools randomly. After every result, it analyzes what it found and decides the smartest next move:

One command. Full assessment. Professional report.

---

Agent Intelligence - Context-Aware Decision Making

| What the Agent discovers | What it does next |
|--------------------------|-------------------|
| Login form on any page | → Jumps to test_authentication immediately |
| WordPress / Drupal / Joomla | → Prioritizes CMS scan |
| SQL error in any response | → Escalates test_sql_injection |
| .git or .env exposed | → Flags Critical, digs deeper |
| API endpoints in JS files | → Switches to GraphQL/JSON dir scan |
| WAF blocking requests | → Activates WAF bypass payloads |
| Source code provided | → Runs SAST before dynamic tests |
| SAST finds SQLi in code | → Confirms with dynamic test_sql_injection |
| PHP detected | → Prioritizes path tra + file inclusion |

---

Agent Modes

---

Agent Backends - Free Local or Cloud

| Backend | Cost | Speed | Quality | Setup |
|---------|------|-------|---------|-------|
| Ollama - mistral (default) | Free | Fast | Great | ollama pull mistral |
| Ollama - llama3.2:3b | Free | Fastest | Good | ollama pull llama3.2:3b |
| Ollama - qwen2.5-coder | Free | Medium | Great | ollama pull qwen2.5-coder |
| Claude API | Paid | Medium | Excellent | ANTHROPIC_API_KEY=sk-ant-... |

No API key required. Run entirely offline with Ollama.

MCP Server - dri from Claude Code / Cursor / Claude Desktop

YeepForge also ships as an MCP server: it
exposes all 28 scan tools (recon_target, crawl_target, test_sql_injection,
nuclei_scan, …) plus set_engagement over the Model Context Protocol, and a
host that already has an LLM drives the engagement with its own subscription -
no ANTHROPIC_API_KEY, no local model needed on YeepForge's side. The host is the
brain; YeepForge is the toolbox.

Set the target once with set_engagement and YeepForge injects it (plus
cookies/proxy/auth) into every later call. Full setup for Claude Code, Cursor and
Claude Desktop: docs/mcp.md.

---

Agent Skills - guided pentest workflows

YeepForge bundles Claude Agent Skills
under .claude/skills/, auto-discovered by Claude Code:

yeepforge-pentest - orchestrates a full engagement through the MCP tools:

authorization gate → recon → crawl → OWASP Top 10 testing → exploit → report.

yeepforge-sast - dri analysis (analyze_source_code).

security-fuzzing · security-payloads · security-patterns - SecLists

payload/wordlist/secret-pattern references for manual testing.

The three security-* skills are from
awesome-skills-security
(MIT, a SecLists redistribution); web-shell samples were excluded on purpose. See
.claude/skills/NOTICE.md for attribution.

---

<a id="code-analysis"></a>

SAST Code Analysis - 14 AI Skills

The Agent can analyze source code alongside dynamic testing. When you provide a codebase:

1. Agent reads architecture, frameworks, entry points
2. Traces every user input through the code
3. Finds SQLi/XSS/RCE/SSRF/JWT/IDOR in source
4. Then confirms findings with dynamic testing
5. Combined report: static + dynamic findings

---

Full Feature Set

Automated Engine

| Key | Module | What it does |
|-----|--------|-------------|
| [A] | Auto Full Scan | Crawl → test all OWASP → nikto → nuclei → C |
| [C] | Web Crawler | Links, forms, JS APIs, robots.txt, sitemap - plus optional headless rendering for SPAs |
| [W] | WAF Bypass | 14+ WAF fingerprints · 100+ bypass payloads per category |
| [O] | OOB Collaborator | interactsh-backed callbacks confirming blind SQLi, SSRF, XXE, Log4Shell, CMDi |
| [33]| AI Agent | Autonomous pentest · chat · payloads · SAST code analysis |

OWASP Top 10 (2021)

| Category | Module | Co |
|----------|--------|---------|
| A01 Broken Access Control | [2] | IDOR · path tra · forced browsing · JWT |
| A02 Cryptographic Failures | [3] | Sensiti · cleartext · weak hashes · JS secrets |
| A03 Injection | [4] | SQLi · XSS · SSTI · CMDi · XXE · NoSQL · LDAP |
| A04 Insecure Design | [5] | Business logic · race conditions · mass assignment |
| A05 Security Misconfiguration | [6] | Headers · CORS · debug endpoints · default creds |
| A06 Components | [7] | CVE scan · nikto · CMS scanner · Log4Shell |
| A07 Authentication Failures | [8] | Brute force · session · MFA bypass · credential stuffing |
| A08 Integrity Failures | [9] | Deserialization · SRI · CI/CD exposure |
| A09 Logging Failures | [10] | Log injection · stack trace exposure · lockout testing |
| A10 SSRF | [11] | Internal ports · cloud metadata · gopher · blind OOB |

Ad Exploitation

| Module | Techniques |
|--------|-----------|
| [17] HTTP Request Smuggling | CL.TE · TE.CL · TE obfuscation · H2.TE downgrade |
| [18] OAuth / SAML / OIDC | Redirect URI bypass · state CSRF · SAML XSW/XXE/replay |
| [19] WebSocket Security | CSWSH · message injection · origin bypass |
| [20] Cache Poisoning | Unkeyed headers · fat GET · cache deception · DoS |
| [21] API Security | GraphQL introspection · batch abuse · REST enumeration |

Web Application Attacks

| Module | Techniques |
|--------|-----------|
| [26] Open Redirect | 17 bypass · scheme abuse · ATO chain |
| [27] Clickjacking | Frame detection · PoC generator · sandbox bypass |
| [28] CRLF Injection | Header injection · response splitting · log injection |
| [29] Subdomain Takeover | 25 service fingerprints · CNAME hijacking · nuclei |
| [30] Account Takeover | Password reset · email change · 2FA bypass |
| [31] Ad File Upload | Polyglot · SVG XSS · ZIP slip · ImageMagick |
| [32] HTTP Parameter Pollution | WAF bypass · business logic · DOM HPP |
| [37] CSRF | Token detection · bypass · PoC generator · SameSite |
| [38] HTTP Verb Tampering | PUT/DELETE/TRACE · XST · method o |
| [39] Prototype Pollution | JS __proto__ · PHP type juggling · XPath · email injection |
| [40] LFI / RFI / PHP Wrappers | Local/remote inclusion · log poisoning · php:// filters |
| [41] Webshells | Generate · upload · interact · obfuscate · reverse shell |
| [42] Session Security | Cookie attributes · fixation · timeout · logout · cache |
| [43] XXE Injection | Classic · blind OOB · billion laughs · SSRF via XXE |
| [44] Race Conditions & IDOR | TOCTOU · limit-o · IDOR automation · business logic |
| [45] Passi (OSINT) | DNS · Google dorks · GitHub leaks · Shodan · certs - zero target traffic |

---

Quick Start

1. Install

2. Start Ollama (free AI backend)

3. Launch Agent

Agent + SAST (when you ha code)

---

Reports

Every assessment produces a professional report:

Formats: HTML (print-ready) · Markdown · JSON (CI/CD ready)

---

Installation & Requirements

Dependencies

AI Backend Setup

Configuration (.env)

---

Module Reference

---

Architecture

---

CLI Reference

---

Engagement Safety - scope, rate limiting, e

Every request made through the engine in utils/http.py is:

Scope-checked - out-of-scope hosts are blocked before the request leaves

your machine. With no explicit --scope, the target host and its subdomains
are the scope, so a scan cannot wander onto someone else's asset.

Rate-limited - a shared budget across the whole run (--rps, default 10),

with Retry-After honoured. This is what keeps you off a program's ban list.

Proxied - --proxy applies to every request, so Burp's history is the

complete engagement record.

Recorded - each finding can carry the raw request/response that proves it

plus a curl reproduction line, rendered into the HTML and Markdown reports.
Findings also carry a confidence level (Confirmed / Firm / Tentative).

See docs/engine.md for the full configuration reference and
for how to write a module against the client.

---

Ethical Use & Legal

YeepForge is built for:

Authorized penetration testing engagements

Bug bounty programs (in-scope targets only)

CTF competitions

Security research (your own systems)

Learning and education

Never test systems without explicit written authorization.

The author assumes no liability for misuse. By using YeepForge, you confirm you ha authorization for all targets.

YeepForge runs attacker-controlled data on your machine - see
SECURITY.md for the tool's own security posture, how to run it
safely, and how to report a in YeepForge itself.

---

Contributing

Pull requests are welcome. CONTRIBUTING.md covers the dev
setup, how to add a module, and the two non-negotiable rules: all HTTP goes
through utils/http.py, and no value is e into a shell string.

---

<di="center">

YeepForge v1.0 - Agent-First Web Security

*capture0x - powered by tmrswrr*

./run.sh → [33] YeepForge Agent → The AI does the rest.

</div>

项目信息

开发语言：Python

Star 数：8

技术要点

该项目属于安全软件方向的开源资源，代码结构清晰，适合安全研究人员学习参考。项目围绕Automated web application penetration testing framework: OWASP Top 10 + advanced attacks, an autonomous AI agent (Ollama/Claude/OpenAI), SAST source-code analysis, and an MCP server, with professional CVSS-scored reports.展开，提供了可运行的实现，对理解相关安全机制、工具原理与攻防技术具有较高的研究价值。

从技术实现角度看，此类项目的核心在于对底层原理的深入理解与工程化落地。开发者在研究过程中可以结合官方文档与社区讨论，梳理项目的初始化流程、核心模块职责与数据流转，逐步掌握其中的技术细节与设计思路。

使用与学习建议

1. 下载源码压缩包，解压到本地环境。

2. 根据项目类型准备对应的运行环境与依赖。

3. 按照 README 中的步骤完成配置与部署。

4. 在隔离的测试环境中运行验证，避免影响生产系统。

免责声明

本资源来源于开源社区，仅供安全研究、技术学习与合法授权范围内的攻防演练使用，请勿用于任何非法或侵害他人权益的用途。使用本项目产生的一切后果由使用者自行承担。