Claude collaboration tools left the door wide open to remote code execution

Check Point researchers found and reported three serious vulnerabilities in Anthropic's Claude Code that convert repository-controlled configuration files into an attack surface: (1) malicious Hooks in .claude/settings.json allowing arbitrary shell commands and RCE; (2) an MCP consent bypass that auto-approves external tool servers and enables pre-dialog command execution; and (3) an overridable ANTHROPIC_BASE_URL that exposed developers' Anthropic API keys, allowing workspace file manipulation. The issues were demonstrated with proof-of-concept videos, responsibly disclosed, and Anthropic issued fixes and CVEs (including CVE-2025-59536 and CVE-2026-21852), underscoring a significant software supply-chain risk for development workflows.