Workflow

This is the MUXAI workflow: four stages that move from raw ideas to a working prototype inside a proper agent harness. Each stage has activities and tools. Learnings from Iteration feed back into Discovery — the loop never really ends.

1. Discovery

Start wide before you open the IDE. Clarify the problem, explore directions, and collect references while the cost of change is still low.

Activities: Research, ideation, brainstorming — user flows, competitive scans, mood boards, rough sketches, and “what if we tried…” conversations.

Tools: Sidekick, Gemini, Claude — use chat agents for exploration, synthesis, and early copy without committing to code.

Output: A clear brief — goal, audience, constraints, and what “done” looks like for this slice of work.

2. Context Artifacts

Turn discovery into durable context the agent can read on every run. If it only lives in chat, you will re-explain it tomorrow.

Activities: Write and organize .md files; capture reference designs (Figma frames, screenshots, links, inspiration).

Typical artifacts:

• Product or feature notes (PRODUCT.md, briefs)
• Design principles and review criteria (design.md, guidelines.md)
• Reference layouts, tokens, or component examples
• Links to Figma files or exported frames

Output: A small set of markdown and reference files that describe what to build and how it should feel — not just a one-off prompt.

3. Setup Harness

Prepare the repo and IDE so the agent has structure, rules, and skills — not a blank folder and good intentions.

Activities: Scaffold or open the project; add agent config, guidelines, skills, and design docs; wire in your context artifacts from stage 2.

Harness checklist:

• AGENTS.md — roles, boundaries, project conventions
• guidelines.md / design.md — design system expectations
• Skills — install from the Skills page (setup, design-review, shadcn, etc.)
• Context artifacts linked or copied into the repo
• File structure in IDE — Cursor, VS Code, or your editor with rules, skills, and docs in predictable paths

Output: A project the agent can enter cold and still produce on-brand work. Run the Setup skill once per new project if you are starting from scratch.

4. Iteration

Build in the harness. Ship slices, review against your system, and refine — prototype speed with production guardrails.

Activities: Prototyping — layout shell first, then sections; wire real components and tokens; preview locally; fix drift before calling a slice done.

Tools: VS Code, Claude Code, Cursor, Figma Make — pick the surface that matches the task (code in the repo, exploration in Figma, agent in the IDE).

Practices:

• Implement structure and navigation before polishing every detail
• Run design review when something feels off-brand
• Commit small, review diffs, and keep context files updated when decisions change

Output: A working prototype or shipped slice that matches your design system — not a one-off AI screen.

Iteration runbook (one slice)

Use this checklist for each feature branch. Details in Collaboration.

Step	Action
1	Create branch from latest main
2	Implement layout shell, then section UI
3	Run npm run dev — preview locally
4	Run design-review skill or manual check vs design.md
5	Quick accessibility pass — focus, contrast, labels
6	Review Git diff — no secrets, no scope creep
7	Open pull request with screenshots + Figma link
8	Address feedback; merge when approved

Skills by stage: Discovery → grill-me · Harness → setup · Iteration → design-review, make-interfaces-feel-better, hallmark

Feedback loop

What you learn in Iteration — what worked, what drifted, what was missing from context — goes back to Discovery for the next cycle. Update artifacts, skills, and harness docs so the next run starts sharper.

Common feedback actions:

• Add a glossary term or setup note you had to rediscover
• Tighten design.md after review findings
• Install or update a skill that closed a gap
• Refresh reference designs when the product direction shifts

---

Use the Setup and Skills sections of this wiki as living docs — extend them whenever the loop teaches you something new.