Clean GitHub Repository Tricks AI Coding Agents Into Running Malicious Payloads

A security researcher demonstrated that a seemingly benign GitHub repository can execute a malicious payload when cloned and set up by agentic coding tools like Claude Code. The malicious instructions evade security scanners, AI agents, and human reviewers by hiding in repository setup instructions that AI agents follow during initialization. This represents a new supply chain attack vector specifically targeting AI-assisted development workflows.