The Vibe Code workflow
The Vibe Code workflow is Mastery HQ’s core methodology: you act as the director, and coding agents are your crew. Instead of typing code, you spawn named agents, hand them work in plain language (“tell Alex to fix the login bug”), keep every transcript visible on one canvas, and use built-in quality mechanisms — judged loops, the Model Council, the Sprint Board — to raise the bar and track progress. One person supervises many workstreams without losing the plot.
The loop, in practice
Section titled “The loop, in practice”- Cast your crew. Open agents for the work at hand — a Claude Code for the backend, a Codex for tests, a Gemini for docs. Names persist (“Alex” is Alex tomorrow too), and each agent’s profile accumulates what it has worked on, by category.
- Break work into cards. The Sprint Board holds the plan: Ready → In Progress → Complete. Assign a card straight to an agent: assign login bug to Alex — dispatch is automatic, and interrupted work resumes after a restart.
- Direct in plain language. The command bar routes to anyone by name; the meteor shows who got the job. Voice works everywhere.
- Watch, don’t hover. Glows show who’s running; who’s working gives a roll call; flip all agents shows every last prompt at once; the Audit Log records everything.
- Raise quality with loops. Don’t accept the first draft: loop: improve this landing copy until score 9 re-runs the work against a judge until it clears your bar. For decisions, run a Model Council — several models compete, a blind judge scores.
- Delegate the directing too. Hermes, the built-in orchestrator, can dispatch tasks and drive the workspace itself when you’d rather manage by intent.
Why it works
Section titled “Why it works”Visibility is the discipline: when every agent’s work is a window you can see, flip, and score, quality problems surface in minutes, not at merge time. The canvas is your studio floor; walk it.