review-loop

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 0.90). The scripts deliberately disable sandboxing and permission checks for peer LLM CLIs (Codex: --dangerously-bypass-approvals-and-sandbox + creating a trusted project entry; Claude: --dangerously-skip-permissions), and pass local repository/context into remote CLIs — behavior that intentionally weakens host safeguards and could enable remote code execution or data exfiltration when those CLIs communicate with external services.