Working with Claude on Eden Stack

January 25, 2025 · Eden Stack Team

agenticclaudeaiworkflowskills

Working with Claude on Eden Stack

Eden Stack isn't just a template—it's designed for a new way of building software. Instead of writing every line yourself, you describe what you want and let AI implement it, following the patterns already established in the codebase.

This post shows you how to work with Claude effectively on Eden Stack projects.

The Core Idea

Traditional development:

You → Write code → Test → Debug → Repeat

Agentic development:

You → Describe intent → AI writes code → You review → AI fixes → Done

The key insight: AI is better at repetitive pattern-matching than you are. Once a pattern exists in your codebase, AI can replicate it faster and more consistently than you can.

Skills: Institutional Knowledge for AI

Eden Stack includes Claude skills—markdown files that teach Claude how things work in this specific codebase.

.claude/skills/
├── better-auth-elysia/     # Auth patterns
├── drizzle-neon/           # Database patterns
├── elysia-eden-treaty/     # API patterns
├── inngest-elysia/         # Background job patterns
├── stripe-elysia-inngest/  # Payment patterns
└── ...

When you mention a topic, Claude loads the relevant skill automatically. The skill contains:

  • Why things are done this way
  • How to implement the pattern
  • Examples from the actual codebase
  • Gotchas to avoid

Effective Prompting

Be Specific About What You Want

❌ "Add user profiles"
 
✅ "Add user profiles with:
- Database table for profile fields (bio, location, website)
- API endpoint for updating profile
- Profile settings page in the dashboard
- Mobile screen for editing profile"

The second prompt gives Claude clear scope and deliverables.

Reference Existing Patterns

"Add a feedback widget using the same pattern as the BookmarkButton component"

Claude will find the referenced pattern and replicate it.

Ask for Specific Technologies

"Add email notifications for new messages using React Email and Resend,
processed via Inngest background jobs"

This tells Claude exactly which stack to use.

Real Workflow Examples

Example 1: Adding a Feature

You say:

Add a "favorites" feature where users can star conversations.
Show favorites in a separate section of the sidebar.
Include a keyboard shortcut (Cmd+S) to toggle favorite.

Claude will:

  1. Create favorites table in schema
  2. Generate migration
  3. Create API endpoints in Elysia
  4. Create React hooks with TanStack Query
  5. Create UI components
  6. Add keyboard shortcut handler
  7. Run type-check to verify

Example 2: Fixing a Bug

You say:

When I log out on mobile, the session sometimes persists.
Check the auth flow and fix it.

Claude will:

  1. Examine the auth client configuration
  2. Check SecureStore clearing logic
  3. Trace the signOut flow
  4. Identify the issue
  5. Implement fix
  6. Verify with type-check

Example 3: Refactoring

You say:

The conversation list is slow with 100+ items.
Add virtualization and optimize the query.

Claude will:

  1. Install @tanstack/react-virtual
  2. Refactor ConversationList to use virtualization
  3. Add pagination to the API endpoint
  4. Update the React Query hook for infinite scroll
  5. Test performance

When to Intervene

AI isn't perfect. Here's when to step in:

Review Before Commit

Always review generated code before committing. Look for:

  • Logic errors: Does it actually do what you asked?
  • Security issues: SQL injection, auth bypasses, etc.
  • Performance: Unnecessary queries, missing indexes
  • Edge cases: Empty states, error handling

Provide Context When Stuck

If Claude keeps getting something wrong:

You're approaching this wrong. In this codebase, we handle X by Y.
Look at apps/api/src/routes/projects.ts for an example.

Direct references help Claude course-correct.

Make Design Decisions

Claude implements, but you decide:

  • What features to build
  • UX priorities
  • Architecture tradeoffs
  • Business logic

Don't let AI make product decisions.

Tips for Speed

1. Start with the Happy Path

"Add user onboarding. Just the basic flow—
we'll add edge cases later."

Get something working, then iterate.

"Add projects feature:
1. Database schema (projects, project_members)
2. API endpoints (CRUD + invite members)
3. Project list page
4. Project settings page
5. Member management UI"

One prompt, multiple outputs.

3. Use Follow-ups

After initial implementation:

"Now add email notifications when someone is invited to a project"

Build incrementally.

4. Let AI Debug

When tests fail:

"The project invite test is failing. Look at the error and fix it."

Claude can read error messages and fix issues faster than you can.

The 10x Developer Myth (Revisited)

The "10x developer" isn't someone who types 10x faster. They're someone who:

  1. Knows what to build
  2. Describes it clearly
  3. Reviews effectively
  4. Iterates quickly

With AI, you can be that developer—even if you couldn't before.

Anti-Patterns

Things that slow you down:

Over-Specifying

❌ "Create a function called handleSubmit that takes an event 
parameter of type React.FormEvent<HTMLFormElement> and calls 
preventDefault, then extracts the name field from FormData..."

Just say what you want, not how to do it.

Ignoring Errors

❌ [Claude generates code with type errors]
You: "That's fine, we'll fix it later"

Type errors compound. Fix them immediately.

Not Reading Output

❌ [Claude generates 200 lines of code]
You: "Commit it"

At minimum, skim the code. At best, understand it.

The Agentic Mindset

Think of Claude as a very fast, very literal junior developer:

  • Fast: Can write code in seconds
  • Literal: Does exactly what you ask (not what you mean)
  • Amnesiac: Forgets context between sessions
  • Pattern-matching: Excellent at replicating existing patterns

Work with these traits, not against them.

What's Next

As you get comfortable with agentic development:

  1. Create custom skills for your app's specific patterns
  2. Build prompt templates for common tasks
  3. Develop review checklists for AI-generated code
  4. Train your intuition for when AI is wrong

The future of development isn't AI vs. humans—it's AI + humans, each doing what they're best at.

Check out the AI Features guide for more on building AI-powered applications with Eden Stack.

Ready to build with Eden Stack?

One-time payment. Full source code. No lock-in.

View pricing