Building a Harness Engineering Setup Tailored to Our Development Team

TL;DR AI
2 min readKey summary
At the Yojit CLCode Night 2 webinar, Vibers AI shared a team-specific harness engineering case built on Claude Code.
As the startup’s codebase and features grew, plan mode alone led to missed research, wrong implementations, and slower delivery.
They built “Maestro Dev,” splitting work into planning, implementation, validation, E2E, and PR stages.
A main agent acts as the orchestrator while sub-agents handle documentation and tasks; humans make the final spec call.
By formalizing tacit knowledge into skills and docs, the system raises team-wide consistency and stabilizes shipping speed and quality.
