Malware found in fake ComfyUI custom node on GitHub targets Claude skills

A security alert has emerged from the ComfyUI community: a malicious GitHub repository disguised as a custom node for Claude AI skills is circulating. The repository, named "comfyui-custom-node-skills" under user MusfiqurRahma, closely mirrors a legitimate, unaffiliated repo by jtydhr88. However, the malicious version is significantly larger and is not a fork, raising red flags. Upon download and inspection, a Reddit user discovered a nested ZIP archive within subfolders containing three files. Among them, "unit.exe" is instructed to execute a file called "packages.txt", which contains an obfuscated script that appears to be Lua-based. Furthermore, all links in the repository's README files were modified to point directly to this malicious ZIP, likely to infect unsuspecting users who download the custom node.

The attack vector leverages trust in open-source AI tools: ComfyUI users searching for Claude AI integration may download this fake plugin. The obfuscated script could perform any number of malicious actions—data exfiltration, credential theft, or cryptomining. The legitimate repository by jtydhr88 remains safe. The finder is creating a GitHub account to report both repos. Users should verify the authenticity of any custom node before execution, especially checking for forks, commit history, and code reviews. This incident highlights the growing risk of supply-chain attacks in the AI tool ecosystem.