RCE in n8n Git Node via core.hooksPath configuration

This repository is a proof-of-concept (PoC) exploit for CVE-2025-65964, demonstrating the use of a malicious Git pre-commit hook. The repository contains two files: a README.md describing the purpose, and evil_hooks/pre-commit, a shell script intended to be used as a Git pre-commit hook. When the pre-commit hook is triggered (e.g., during a git commit), it prints a message and creates the file /tmp/CVE-2025-65964-pwned, serving as evidence of code execution. The exploit requires the victim to install or use the provided pre-commit hook, making the attack vector local. No network endpoints or remote targets are involved. The code is simple and serves as a demonstration of the vulnerability rather than a weaponized exploit.

This repository demonstrates a proof-of-concept exploit for abusing Git's 'core.hooksPath' configuration in the context of automation platforms like n8n. The repository contains two Git hook scripts ('hooks/pre-commit' and 'hooks/post-checkout'), both of which execute a command that attempts to contact a unique DNS endpoint (yjvnpienggxhpzdumprw0xic2mka4vr9o.oast.fun) using curl, ping, or wget. The README and workflow description explain that if an attacker can set 'core.hooksPath' to a directory within a cloned repository, they can achieve arbitrary command execution on the system performing Git operations. The exploit is local (requires code execution on the victim's machine via Git hooks) and is intended for educational demonstration, with no harmful payloads. The unique DNS endpoint serves as a canary to prove code execution and potential exfiltration. The repository structure is simple, with documentation and two executable hook scripts.

This repository demonstrates a proof-of-concept exploit for a vulnerability in n8n (versions 0.123.1 through 1.119.1) that allows arbitrary code execution via a malicious git pre-commit hook. The exploit consists of a single shell script (evil_hooks/pre-commit) that, when executed, creates a file at /tmp/pwned_without_exec_node to indicate successful exploitation. The README.md provides detailed step-by-step instructions for setting up a vulnerable n8n instance in Docker, preparing a malicious git repository, and configuring an n8n workflow to clone and commit to the repository with a user-controlled core.hooksPath. The attack leverages the ability to redirect git's hooks path to a directory containing attacker-controlled scripts, which are then executed during git operations. The exploit is a local privilege escalation vector, as it requires the target to interact with a malicious repository and execute git operations with the manipulated configuration. The repository is structured with a README.md containing the exploit methodology and a single code file (evil_hooks/pre-commit) serving as the payload.

This repository demonstrates a proof-of-concept exploit for a vulnerability in n8n (versions 0.123.1 through 1.119.1) where a malicious git pre-commit hook can be executed on the server. The exploit consists of a single shell script (evil_hooks/pre-commit) that, when triggered, creates a file at /tmp/pwned_without_exec_node, proving arbitrary code execution. The README.md provides detailed step-by-step instructions to set up a vulnerable n8n instance, configure a workflow to clone a repository containing the malicious hook, and set the git core.hooksPath to the attacker-controlled directory. The exploit leverages the ability to manipulate git configuration and hooks within n8n's workflow automation, resulting in local code execution on the server. No network endpoints are directly targeted; the attack is local to the n8n server environment.

This repository is a proof-of-concept (POC) exploit for CVE-2025-65964. It contains a single Bash script ('evil/pre-commit') intended to be used as a malicious git pre-commit hook. When executed, the script creates a directory named 'success' in the /tmp directory, demonstrating the ability to execute arbitrary commands in the context of a git commit operation. The exploit is local in nature, requiring the victim to use the provided pre-commit hook. There are no network endpoints or remote attack vectors. The repository structure is minimal, with only a README and the exploit script.