I got tired of AI assistants confidently generating broken code. So I made one that checks its own work. LuckyD Code runs every edit through syntax → lint → AST → test verification before it tells you "done". Open source, terminal-native, with a web UI. Key differentiators: • Self-verification loop (no blind trust) • Code knowledge graph indexes your full project • Smart model routing saves $$$ on API costs • Autonomous fixer finds bugs and opens PRs • Docker sandbox for safe execution