micro-app upstream commits installed malicious editor hooks

A repository bot's activity left micro-app contributors with executable editor settings in their source checkout. The September 21 change reached JD's established micro-frontend project, whose normal job was to let separately developed web applications coexist on one page. SafeDep identified the added files as part of the credential-stealing Mini Shai-Hulud worm's spread through compromised GitHub Actions.

The retained source snapshot shows that micro-app used actions-cool/issues-helper for routine issue maintenance. It also contains a new .claude directory and a VS Code task labeled Environment Setup. The task's label described a plausible convenience for someone preparing a development environment, but its command pointed to the same setup program as the new Claude Code session hook. Both paths led away from the framework's browser functionality and into the contributor's local tools.

Static inspection confirmed that the Claude hook invoked node .claude/setup.mjs on session startup, while the VS Code task requested execution when the folder opened. VS Code's automatic-task and workspace trust settings still governed that route. The setup script selected a Bun executable for the host platform and used it to run .claude/index.js. That file's digest matched the payload preserved from the other five repositories in this group; it was not a micro-app-specific implementation change.

SafeDep associated the first malicious commit with a nearby Close Inactive Issue workflow run, while treating the exact trigger as an inference. The source capture established the downstream delivery of the hooks, not how many contributors allowed them to execute. It also did not establish a poisoned npm release or malicious code in websites built with micro-app. The added execution path concerned the people working on the upstream checkout and the credentials available to their development sessions.