A malicious instruction file placed in a user’s Downloads folder attempts to trick AI coding agents into installing a rogue npm package linked to the MALFEX malware campaign. Researchers say the file, named AGENTS.md, contains instructions that an agent follows when it reads local text files in the activated workspace.
Across reports, the described mechanism relies on how programming assistants ingest instructions from files in the working directory and on the fact that websites can save files to the Downloads folder without direct user involvement. In controlled testing, one agent executed the embedded command in most cases, while another succeeded less often due to approval prompts for actions like changing settings.
Outlets also note that the issue is presented as an abuse of intended agent behavior rather than a straightforward product bug. Security researchers and firms emphasize that similar risks can occur because coding agents frequently run commands found in local workspaces, including unauthorized instructions, even when security controls intervene during real-world attempts. Some coverage links the rogue package to additional malware steps, while other accounts focus more directly on how the attempt was detected and blocked on a customer machine.
