Action guardrails for AI coding agents — native Claude Code enforcement, MCP integrations, and Git/CI checks.