Threat
Hackers Hide Malicious Rules in AI Coding Projects to Hijack Assistants
First reported · Discovered homeland411.com
Page published
Earliest dated coverage: 2 Nov 2026 · First observed: 5 Nov 2026 · Latest dated coverage: 2 Nov 2026
Coverage timeline
Single-source analysis — one report is available.
Why it matters
Repository instruction-file poisoning can turn seemingly passive project documentation into instructions for privileged coding agents, making rule-file review and restrictions on tool execution important safeguards for developer credentials and hosts.
Repository instruction-file poisoning hides prompt-injection directives in files such as `.cursorrules`, `CLAUDE.md`, and `AGENTS.md` that AI coding assistants ingest as project guidance. Homeland411 reports that payloads concealed with Unicode characters or HTML comments can redirect assistants toward unauthorized shell execution or secret theft, particularly when automatic execution or MCP tools are enabled.