Wix Flow Alpha Program - Participant Guide

alpha-instructions.md

Wix Flow Alpha Program - Participant Guide

What is this?

You're participating in an alpha test of Wix Flow - a curated Claude Code setup designed to make AI-assisted Wix backend development effective out of the box. The setup includes managed instructions (CLAUDE.md), MCP servers for documentation, and skills for common workflows.

We want to understand how well this works on real tasks and where the gaps are.

Setup

Prerequisites

• Claude Code installed and working
• gh CLI authenticated (gh auth status should show you're logged in)
• Write access to wix-private/wix-flow-alpha-reflections (ask us if you don't have it)

Install the preset

Run from your service repository:

npx @anthropic-ai/nile-setup sync --preset claude-mcp

This installs:

• CLAUDE.md - instructions that guide Claude on Wix development patterns
• MCP servers - nile-docs for documentation search, nile for service operations
• Skills - proto scaffolding, service generation, platformization
• Commands - /wix-flow-reflect for session reflection

Verify it worked:

# Check MCP servers are configured
cat ~/.claude.json | grep nile

# Check skills are installed
ls ~/.claude/skills/

# Check commands are installed
ls ~/.claude/commands/

What we expect from you

Use Claude Code for real work

Use it on your actual tasks - the kind of work you'd normally do:

• Adding features to your service
• Fixing bugs
• Working with protos
• Writing tests
• Anything backend-related

Don't manufacture artificial tasks. We want signal from real work.

Run reflections after sessions

After a meaningful work session (not a quick one-liner), start a new Claude Code session and run:

/wix-flow-reflect

This will:

1. Analyze your most recent session
2. Show you a structured reflection (what worked, what didn't)
3. Ask for your confirmation
4. Upload the reflection + raw session log to our shared repo via PR

Do this at least 2-3 times during the alpha. More is better.

What gets uploaded

• A structured markdown reflection (task summary, issues found, suggestions)
• The raw conversation log (JSONL) so we can re-analyze if needed

The raw log contains everything from your session - prompts, code, tool outputs. It goes to a private Wix repo (wix-private/wix-flow-alpha-reflections).

Tips for good sessions

1. Be specific in prompts - "Add a GET endpoint for user preferences with caching" beats "add an endpoint"
2. Let Claude use the tools - approve MCP and skill invocations when prompted, that's what we're testing
3. Note when something feels wrong - if Claude gives bad advice, goes in circles, or misses something obvious, that's exactly what we want to capture
4. Don't fight the setup - if something in CLAUDE.md or a skill seems wrong, note it in the reflection rather than working around it

Timeline

• Duration: Up to 3 weeks
• Kick-off: See calendar invite
• Sync sessions: We'll schedule 1-2 check-ins to discuss how it's going
• Surveys: Baseline survey before you start, final survey at the end (Google Forms, linked in Slack)

Getting help

• Setup issues: Reach out in the alpha Slack channel
• Claude Code issues: Check /help or report at https://github.com/anthropics/claude-code/issues
• Feedback on the setup: Run /wix-flow-reflect or just tell us directly

Quick reference

Command	What it does
/wix-flow-reflect	Analyze last session and upload reflection
/wix-knowledge	Access Wix backend documentation
/platformizing-proto	Load proto validation context
/scaffolding-proto-package	Scaffold a new proto package
/generating-service	Generate service from protos