OS Command Injection in OpenSSH Expansion Tokens

This repository is a small exploit/PoC for CVE-2023-51385 centered on malicious Git submodule URL handling. The structure is minimal: a README explaining setup and trigger conditions, a .gitmodules file containing the malicious submodule URL, and a bash payload script (shell.sh). The exploit abuses command substitution embedded in the submodule URL field, e.g. ssh://`bash shell.sh`foo.ichunqiu.com/bar, so that when a victim performs a recursive clone/submodule operation in a vulnerable environment, arbitrary local commands may execute. The README documents both a safe proof-of-execution payload (creating cve.txt) and the actual repository payload. It states that exploitation requires a vulnerable OpenSSH version earlier than 9.6p1 and an SSH client configuration with a ProxyCommand rule matching the crafted host pattern. Operationally, this is more than a pure detection script because the included shell.sh launches a real reverse shell using bash TCP redirection to 64.90.7.213:8889. Primary capability: arbitrary command execution on the client machine processing the malicious repository/submodule. Secondary capability in the included payload: interactive remote shell access for the attacker. This is best classified as a supply-chain style attack vector delivered through a malicious repository/submodule reference, with a network callback payload.

This repository provides a proof-of-concept (PoC) exploit for CVE-2023-51385, a command injection vulnerability in OpenSSH's ProxyCommand option. The exploit demonstrates how an attacker with access to a user's SSH configuration file (~/.ssh/config) or the ability to influence the ProxyCommand option can execute arbitrary shell commands. The provided payload creates a file (/tmp/pwned.txt) with a custom message, proving code execution. The exploit is documented in 'exploit.md' and is intended for educational and authorized penetration testing purposes. The repository contains no executable scripts, but provides clear step-by-step instructions for reproducing the vulnerability on affected OpenSSH versions (prior to 9.8, specifically <9.6). The main attack vector is local, requiring the attacker to modify the user's SSH configuration or command-line arguments.