pale-knight/redteam-skill / skills/web-attack
web-attack
FlaggedHTTP/Web exploitation until a server shell or equivalent OS execution. Use after /web-recon has mapped the application. Covers injection, upload, LFI, SSRF/XXE, SSTI, deserialization, JWT/SAML, API logic, desync/cache/parser, and Web-controlled backend abuse. WAF stays here. Endpoint blocks after OS execution hand off to /edr-bypass then return. Direct non-HTTP service ports belong to /recon or /service-attack.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 5.3 KB |
| references/api-security.md | markdown | 20.3 KB |
| references/auth-bypass.md | markdown | 1.9 KB |
| references/cmdi.md | markdown | 5.0 KB |
| references/deserialization.md | markdown | 1.6 KB |
| references/http-desync.md | markdown | 5.2 KB |
| references/java-deserialization.md | markdown | 3.3 KB |
| references/jwt.md | markdown | 1.4 KB |
| references/lfi-rfi.md | markdown | 4.0 KB |
| references/orm-leak.md | markdown | 5.3 KB |
| references/parser-differentials.md | markdown | 4.4 KB |
| references/saml-security.md | markdown | 3.6 KB |
| references/server-side-client-abuse.md | markdown | 3.3 KB |
| references/sqli-mssql-chain.md | markdown | 6.3 KB |
| references/sqli-mysql-chain.md | markdown | 5.9 KB |
| references/sqli-oracle-chain.md | markdown | 4.7 KB |
| references/sqli-postgresql-chain.md | markdown | 4.6 KB |
| references/sqli.md | markdown | 8.0 KB |
| references/ssrf.md | markdown | 4.6 KB |
| references/ssti.md | markdown | 5.3 KB |
| references/upload.md | markdown | 3.6 KB |
| references/waf-bypass.md | markdown | 6.8 KB |
| references/web-cache.md | markdown | 4.1 KB |
| references/xss.md | markdown | 3.8 KB |
| references/xxe.md | markdown | 1.5 KB |
references/http-desync.md
View on GitHub# HTTP Desync / Request Smuggling 详细参考
> 适用:已经从 `/web-recon` 看到 CDN/WAF/reverse proxy → backend 的多组件 HTTP 边界。该类漏洞误报很常见,**必须证明不同组件对请求边界解释不同**。单纯 HTTP pipelining、客户端复用连接、两个正常响应连在一起都不是漏洞。
---
## 1. 前置条件
优先级高:
```
HTTP/2 frontend → HTTP/1.1 backend
CDN/WAF → Nginx/Apache/IIS/app server
不同Server header/error page交替出现
Connection keep-alive
Transfer-Encoding / Content-Length 行为不一致
```
低优先级:单一应用服务器、连接始终关闭、明确 end-to-end H2 且无转换。
---
## 2. 工具优先
### Burp HTTP Request Smuggler
优先使用官方/成熟 Burp 扩展做探测,再人工确认。
不要仅依据 scanner 的“possible”标记直接报漏洞。
### smuggler.py
```
python3 smuggler.py -u https://TARGET
```
用于经典 TE/CL 差异候选;现代 H2/0.CL/CRLF 类仍建议 Burp 工具和手工验证。
---
## 3. 经典请求边界
### CL.TE
前端按 Content-Length,后端按 Transfer-Encoding。
概念请求:
```http
POST / HTTP/1.1
Host: TARGET
Content-Length: 6
Transfer-Encoding: chunked
0
X
```
检测时用**无害 timeout/404 probe**,不要一开始污染其他用户队列。
### TE.CL
前端 TE,后端 CL。
### TE.TE
两边都识别 TE,但其中一端对混淆 `Transfer-Encoding` 语法处理不同:
```
Transfer-Encoding: chunked
Transfer-Encoding : chunked
Transfer-Encoding: xchunked
Transfer-Encoding: chunked, identity
```
具体有效语法强依赖产品版本。
---
## 4. Modern Desync Matrix
```
CL.TE
TE.CL
TE.TE
CL.0
0.CL
H2.CL
H2.TE
H2C downgrade
Client-side desync
Request tunnelling
Chunk extension / line-terminator differential
CRLF-powered desync
```
### CL.0
后端忽略 request body/Content-Length,body 残留成为下一请求前缀。
### 0.CL
前端认为无 body,后端按 Content-Length 等待/消费数据。
### H2.CL / H2.TE
HTTP/2 本身没有传统 TE chunked framing,但前端降级到 H1 时可能错误复制 `Content-Length`/TE 语义。
使用 Burp Repeater 的 HTTP/2 inspector/Request Smuggler,不要靠 curl 伪造非法 H2 header。
---
## 5. 如何避免“HTTP Pipelining = 漏洞”的误报
**错误判断:**
```
同一TCP连接发送两个正常请求
→ 收到两个正常响应
→ 误判RQP
```
真正需要:
```
同一攻击请求
前端认为边界在A
后端认为边界在B
↓
残留字节影响后续独立请求/响应
```
验证方式:
1. 用 Burp 的单连接控制,不让客户端自己做 pipelining。
2. 发送 harmless prefix,观察**另一个独立请求**是否被改变。
3. 重复多次确认,不把网络抖动/代理重试当结果。
---
## 6. 2025:HTTP/1.1 Desync Endgame
现代研究重点已经不只是 CL.TE/TE.CL,尤其关注:
```
0.CL / double-desync
chunk parser line terminator
upstream HTTP/1.1 downgrade
```
如果前端对客户端提供 H2,但 upstream 仍是 H1,优先级提高。
---
## 7. 2026:CRLF-Powered Desync
2026-08 新研究展示:某些服务器/代理允许 path/header 解析中的 CRLF 注入进一步“斩断”HTTP stream,从 header injection 升级为 desync,而且一部分可由浏览器 `fetch()` 触发。
### Recon 信号
先找**路径/参数可产生响应头或请求头注入**:
```
%0d%0a
%0a
encoded space + HTTP/1.1 + CRLF
```
不要把所有 CRLF injection 都直接判 desync;必须证明连接边界受影响。
### 验证策略
```
CRLF/header injection primitive
↓
让后端保持连接
↓
注入无害 TRACE/404/HEAD 前缀
↓
观察响应队列/下一请求错位
```
2026 研究表明 HEAD technique、100-continue 等可帮助观察盲 tunnelling/desync,但这些属于**高级验证**,优先在 HTB/本地靶场使用。
---
## 8. 2026:HTTP Terminator / Multipart Byteranges Differential
新研究发现部分实现会错误把 request 的:
```
Content-Type: multipart/byteranges; boundary=...
```
按 response-oriented multipart/byteranges 语义处理,造成新的 request boundary differential。
**状态:RESEARCH-GATED。** 这不是“见到 multipart/byteranges 就有漏洞”。
测试思路:
```
POST body使用multipart/byteranges结构
↓
在可疑 terminator 后放无害下一请求 marker
↓
比较前端/后端是否消费相同字节数
```
优先用最新 Burp/Turbo Intruder 方法;不要在 skill 里硬编码一个“通杀 payload”。
---
## 9. 从确认 Desync 到 Web Shell
可利用影响常见:
```
绕过前端ACL/WAF
内部路径访问
poison下一用户请求
response queue poisoning
cache poisoning
credential capture
```
`/web-attack` 的目标是 Shell,因此拿到 auth/admin 后继续找:
```
admin upload/plugin/theme editor
CI/build command
server-side template
internal debug/RCE endpoint
```
确认 desync 但不能继续到 Shell → 漏洞成立,但本模块 success condition 尚未达成。
---
## 10. 操作注意
- HTB/CTF 可做跨请求验证;生产授权环境优先单用户/无害 marker。
- 不要用会长期污染共享 cache/队列的 payload 作为第一验证。
- 记录原始 request/response、连接模式、H1/H2、重复次数。
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| receiving-code-reviewobra/superpowersUse when receiving code review feedback, before implementing suggestions, especially if feedback seems unclear or technically questionable - requires technical rigor and verification, not performative agreement or blind implementation | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| requesting-code-reviewobra/superpowersUse when completing tasks, implementing major features, or before merging to verify work meets requirements | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| using-git-worktreesobra/superpowersUse when starting feature work that needs isolation from current workspace or before executing implementation plans - ensures an isolated workspace exists via native tools or git worktree fallback | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| verification-before-completionobra/superpowersUse when about to claim work is complete, fixed, or passing, before committing or creating PRs - requires running verification commands and confirming output before making any success claims; evidence before assertions always | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |