Tego AI Discloses Second Claude Flaw in a Week: Hidden Link Silently Sends Files to Attackers

**Tego AI disclosed a vulnerability in Anthropic's Claude Code**: a repository-committed instruction file can include an @import directive pointing to a symbolic link which resolves outside the project, causing Claude Code to read external files and include their contents in the first outbound request without a visible approval or prompt. The issue can leak sensitive files (e.g., /etc/passwd) when a developer clones and trusts a repository; Tego confirmed the behavior against Claude Code v2.1.x, reported it via HackerOne in July 2026, and Anthropic closed the report as Informative.