Prompt Engineering
Master the art of communicating with Shadowfax to unlock its full analytical potential.
Why Prompting Matters
Shadowfax is an AI-powered data analyst that transforms natural language into sophisticated analyses. The key to success? Effective prompting. Well-crafted prompts lead to accurate analyses, while vague prompts lead to confusion and suboptimal results.
The difference between:
"Show me sales data" → Agent has to guess what you mean
"@[sales_2024] Calculate monthly revenue with MoM growth rates for Q3" → Clear, actionable request
What You'll Learn
This section contains two comprehensive guides:
Learning to Prompt Better
A complete guide covering:
- How Shadowfax's AI agent works (Workbook, SQL views, reactive system)
- The three-layer context model (@mentions, Knowledge, prompts)
- Slash commands:
/plan,/visualize,/recommendations,/taxonomy - C.L.E.A.R. framework for effective prompts
- Advanced techniques and troubleshooting
- Working with Shadowfax's Vega-powered visualizations
Perfect for: Understanding the principles behind effective prompting and mastering advanced techniques.
Useful Prompts
A practical template library with 100+ copy-paste ready prompts for:
- Data exploration and quality checks
- Cleaning and preparation
- Metrics and KPIs
- Time-based analysis
- Segmentation and cohorts
- Joins and relationships
- Advanced visualizations
- Statistical analysis
Perfect for: Quick reference when you need a prompt for a specific task.
Core Concepts
Context is King
Shadowfax needs three types of context:
- Knowledge (persistent): Business rules, KPI definitions, data dictionaries
- @Mentions (per-prompt): Which specific data tables to analyze
- Prompts (per-request): What to do with that data
Slash Commands
/plan: Review the agent's approach before execution/visualize: Create sophisticated charts with Vega/recommendations: Get suggestions for next analyses/taxonomy: Discover data categories and create classification AI tables
The Workbook Advantage
- Every transformation is a SQL view you can inspect
- Fully reactive: changes propagate automatically
- Build iteratively and verify at each step
Quick Start
New to prompting? Start with Learning to Prompt Better to understand the fundamentals.
Need a specific prompt? Jump to Useful Prompts and find templates for your use case.
Working on complex analysis? Use the /plan command to collaborate with the agent on your approach.
Remember: Effective prompting = Better Context + Clear Instructions + Iteration
Happy prompting!