How to use Startbase
Debugging prompts
Structured prompt recipes for investigating errors, auditing your codebase, optimizing performance, and escaping persistent bugs.
Building with AI is fast, until something breaks. This guide gives you battle-tested prompts and workflows for quickly diagnosing and resolving issues inside Startbase.
Full codebase audit
When your project has grown large or you suspect structural problems, ask the AI to review the whole codebase without making changes first.
Perform a comprehensive audit of the entire codebase.
- Flag files or logic that are in the wrong place.
- Check separation of concerns (data vs UI vs state).
- Highlight overly complex or tightly coupled areas.
- Provide an ordered list of recommendations.
This is read-only - do not modify any code yet.The AI will respond with a structured report: misplaced logic, coupling hotspots, suggested refactors, ranked from critical to nice-to-have. You can then implement the fixes one by one.
Safe approach for delicate changes
When you are modifying something critical (authentication, payments, a core algorithm) prepend a cautionary guideline so the AI proceeds carefully.
The next change is in a critical area. Proceed with extreme care.
- Examine all related code and dependencies before editing.
- Do NOT touch unrelated components.
- If anything is unclear, explain your reasoning before continuing.
- Verify nothing else breaks after the change.
Task: Add Google OAuth alongside existing email/password auth.Performance optimization audit
If your app works correctly but feels slow, ask the AI to diagnose performance without writing code.
Analyze this project for performance bottlenecks:
- Unnecessary or duplicated network calls.
- Components re-rendering too often.
- Large unoptimized assets or bundles.
- Missing caching, memoization, or lazy loading.
Provide recommendations only - no code changes yet.Handling stubborn errors
Some bugs keep resurfacing in slightly different forms because the root cause was never addressed. Use these strategies to break the cycle:
- Ask the AI what it has already tried: "List every fix we have attempted for this error."
- Request a plain-English explanation: "Explain why this error occurs in simple terms."
- Explore a completely different approach: "Suggest an alternative architecture that avoids this problem entirely."
- If the code is too tangled, restore an earlier checkpoint and rebuild the feature from scratch with smaller steps.
Sample debugging workflows
Root cause vs. symptom
Always ask "why did this happen?" rather than just "how do I stop the error?" A quick patch might silence the message, but the underlying bug will resurface later.
Example: "You added a null check to fix the crash, but why was the value null in the first place? Can we address the source?"
Community debugging guidebook
These principles, shared by experienced community builders, can serve as guardrails you paste into your project knowledge file: