Observed incident

Malicious GitHub files target AI coding agents and trusted repository hooks

Reported

Published here

What happened

Bitdefender identified 22 malicious GitHub links carrying instructions and configurations designed to execute a concealed credential-stealing loader. One route persuades an assistant to run setup code; another uses session-start hooks after a developer trusts the repository.

Parties and reporting

Where this happened

Victim location not disclosed. Real organizations or systems were affected or targeted, but the sources do not say where. Explore the world map.

AI tools and systems: AI coding-agent instruction files, Claude Code session-start hooks, MCP configurations and editor tasks, JavaScript credential-stealing loader

Outcome

The analyzed payload can collect credentials, change local agent configurations and use stolen repository tokens to propagate. The report establishes malicious repository content and payload capabilities, but gives no confirmed developer-victim count, successful-execution total or completed takedown.

Operating lesson

Review repository instructions, hooks and tool definitions before trusting a project; isolate execution and restrict credentials available to coding agents.

Evidence and limits

Twenty-two counts malicious links, not infected users. Propagation and theft capabilities are not proof that each action occurred. Mentioned agent products are targets or configuration formats; the analysis does not establish a vulnerability in every named product. Vendor blocking claims are not an independent effectiveness evaluation.

Sources