Debugging Prompts

When your app doesn't match your vision, here's how to diagnose and fix your prompts.

Understand Why Prompts Fail

Most prompt failures fall into one of four categories:

• Too vague — The AI made assumptions because there wasn't enough information.
• Too broad — Too many features were requested at once, causing some to be ignored or poorly implemented.
• Missing constraints — Rules, edge behaviors, and access conditions were not specified.
• Ambiguous terminology — Words like "dashboard", "profile", or "feed" mean different things in different apps. Without context, the AI picks a generic interpretation.

Diagnostic Checklist

Before resubmitting a prompt, run through this checklist:

1. Did you specify who the app is for?
2. Did you state the core problem the app solves?
3. Did you list features explicitly, not just reference a well-known app?
4. Did you define user roles and their permissions?
5. Did you include edge cases or constraints (e.g., validation rules, notification triggers)?
6. Was the prompt focused on an MVP, or did you request too much at once?

The "Reference App" Anti-Pattern

Problem: "Build something like Notion" or "Make a clone of Trello."

This fails because it gives the AI no information about your specific users, features, or constraints. The result is a generic interpretation that may not match what you actually need.

Fix: Replace references with explicit feature descriptions.

Instead of:

Build something like Trello.

Write:

Build a project management app for marketing teams. Users can create boards with columns (e.g., To Do, In Progress, Done). Cards can be assigned to team members, given due dates, and tagged with a priority level (Low, Medium, High). Admins can create new boards; members can only edit cards within boards they've been added to.

The "Everything at Once" Anti-Pattern

Problem: Packing an entire product roadmap into a single prompt.

When you ask for too much in one generation, the AI either ignores later features, generates them superficially, or produces an inconsistent structure.

Fix: Use phased prompting.

• Prompt 1 (MVP): Core screens, authentication, primary user flow.
• Prompt 2 (Extend): Secondary features, admin panel, notifications.
• Prompt 3 (Polish): Edge cases, additional roles, integrations.

Debugging a Specific Broken Feature

When one feature isn't working correctly, isolate it:

1. Identify the exact behavior that's wrong.
2. Describe the expected behavior and the actual behavior.
3. Submit a targeted correction prompt.

Template:

The [feature] is currently doing [actual behavior]. It should [expected behavior]. Specifically: [any additional context or constraint].

Example:

The search bar on the listings page is currently returning all results regardless of the search term. It should filter listings in real time by title as the user types. Only show listings where the title contains the search input.

When to Stop Prompting and Assign a Developer

Some issues are not prompt problems — they're engineering complexity problems. Consider assigning a task to an Imagine.bo developer when:

• You've sent 3+ correction prompts on the same issue without resolution.
• The feature requires a specific third-party API integration.
• The logic involves complex conditional business rules or multi-step workflows.
• The issue is related to performance, security, or data integrity.

Use the Hire a Human button in your dashboard. Frame the task clearly: what the feature should do, what it's currently doing, and any relevant constraints or examples.

Prompt Improvement Examples

For issues that aren't resolved through prompt refinement, use the Hire a Human feature from your project dashboard — or contact the Imagine.bo support team.

FAQ