V
Vector
Sign inGet Started

Spec Generation#

Vector's spec generation transforms recommendations into structured documents that engineering teams (or AI coding assistants) can immediately act on.

What's in a Spec?#

Generated specs include:

Problem Statement#

  • Clear description of the issue
  • Quantified impact (metrics, revenue, users affected)
  • Current state vs. desired state

Evidence#

  • Specific data points supporting the recommendation
  • Comparison to benchmarks or industry standards
  • Historical trends

Proposed Solution#

  • Recommended approach
  • Key implementation details
  • Success criteria

Technical Considerations#

  • Affected components/pages
  • Dependencies
  • Potential risks
  • Estimated effort

Acceptance Criteria#

  • Definition of done
  • Metrics to track
  • A/B test suggestions

Export Formats#

Markdown (LLM-Ready)#

Optimized for AI coding assistants like Claude, Cursor, or Codex:

# Feature: Simplify Checkout Flow

## Problem
45% cart abandonment at payment step (industry avg: 28%)

## Solution
Reduce checkout from 4 steps to 2:
1. Pre-fill address from account
2. Combine review + payment screens

## Acceptance Criteria
- [ ] Single-page checkout
- [ ] Address auto-fill for logged-in users
- [ ] Payment form on same page as summary
- [ ] Mobile-optimized layout

## Metrics
- Target: Reduce abandonment to <35%
- Track: Checkout completion rate, time-to-purchase

PDF (Executive)#

Formatted for stakeholder review:

  • Professional layout
  • Charts and visualizations
  • Executive summary
  • ROI projections

Using Specs with AI Assistants#

Vector specs are designed to work directly with AI coding tools:

With Claude/Cursor#

  1. Generate spec from recommendation
  2. Copy the Markdown output
  3. Paste into Claude or Cursor with: "Implement this feature spec"
  4. Review generated code

With GitHub Copilot#

  1. Create a new issue with the spec content
  2. Reference the issue in your PR
  3. Copilot can use issue context for suggestions

Best Practices#

  1. Review before sharing: Specs are AI-generated—verify accuracy
  2. Add context: Include links to existing code or design files
  3. Set constraints: Mention tech stack, design system, testing requirements
  4. Iterate: Use spec as starting point, refine with engineering

Spec Quality#

Spec quality depends on:

  • Integration data quality: More data = better specs
  • Recommendation confidence: High-confidence recommendations produce better specs
  • Connected platforms: Multiple data sources provide richer context

Coming Soon#

  • GitHub Integration: Auto-create issues from specs
  • Linear Integration: Push specs directly to your backlog
  • Custom Templates: Define your own spec format
  • Team Sharing: Share specs with teammates