V
Viban
Get Started

Task Refinement

Learn how to use AI-powered task refinement to transform simple descriptions into high-quality, actionable prompts.

What is Task Refinement?

Task refinement uses AI to enhance your task descriptions. A simple idea like "add user authentication" becomes a detailed specification with:

  • Clear objectives
  • Acceptance criteria
  • Scope boundaries
  • Implementation guidance

Using the Refine Button

  1. Create or open a task
  2. Click the Refine button
  3. Wait for AI processing (usually 5-10 seconds)
  4. Review the enhanced description
  5. Edit if needed, then save

Before and After

Before Refinement

Add dark mode to the app

After Refinement

## Objective
Implement a dark mode theme toggle that allows users to switch between light and dark color schemes.

## Requirements
- Add a theme toggle button in the header/navigation
- Implement dark color palette using CSS custom properties
- Persist user preference in localStorage
- Respect system preference (prefers-color-scheme) by default
- Ensure smooth transition between themes

## Acceptance Criteria
- [ ] Toggle button switches between light/dark modes
- [ ] Theme preference persists across sessions
- [ ] All components render correctly in both themes
- [ ] Transitions are smooth (no flash of wrong theme)
- [ ] System preference is respected on first visit

## Out of Scope
- Multiple theme options (only light/dark)
- Per-page theme settings
- Theme scheduling

How It Works

The refinement process:

  1. Analyzes your input: Understands intent and context
  2. Adds structure: Organizes into sections
  3. Expands details: Adds missing requirements
  4. Defines boundaries: Sets clear scope
  5. Creates checklist: Measurable acceptance criteria

Best Practices

Provide Context

More context leads to better refinement:

# Less context (okay)
Add search

# More context (better)
Add search to find products by name and description

Review and Edit

Always review refined descriptions:

  • Remove irrelevant suggestions
  • Add project-specific requirements
  • Adjust scope as needed

Iterative Refinement

You can refine multiple times:

  1. Start with a rough idea
  2. Refine once for structure
  3. Add your own details
  4. Refine again for polish

How It's Implemented

Task refinement uses Claude Code CLI with the Haiku model for fast, cost-effective processing:

  • Model: Haiku (optimized for speed)
  • Timeout: 60 seconds
  • Output format: Markdown with structured sections

The refinement prompt instructs the AI to:

  • Output only the refined task (no conversational text)
  • Add clear title and objective
  • Include acceptance criteria
  • Define appropriate scope

Troubleshooting

Refinement Takes Too Long

  • Check AI agent availability
  • Try refreshing and retrying
  • Check network connectivity

Output Quality Issues

  • Provide more input context
  • Try refining again
  • Manually edit the output