data-context-extractor_skill

Overview

This skill extracts company-specific data knowledge from analysts and generates a tailored data analysis skill for your warehouse. It discovers schemas, clarifies entity and metric definitions, documents common filters and gotchas, and packages reference files for repeated analyst use. Use it to onboard AI assistants to your data model and reduce query mistakes.

How this skill works

In Bootstrap mode the skill connects to your warehouse, lists schemas, identifies the most-used tables, and asks targeted questions about entities, identifiers, metrics, and hygiene rules. In Iteration mode it loads an existing skill, finds gaps, prompts focused questions, updates reference files for a domain, and repackages the skill. The output is a structured collection of reference files that capture entity definitions, metric formulas, table docs, and example queries.

When to use it

• Setting up AI-driven analysis for a new data warehouse (BigQuery, Snowflake, Postgres/Redshift, Databricks)
• Onboarding analysts or an AI assistant to company-specific terminology and metrics
• Documenting key tables, joins, and primary identifiers for reproducible queries
• Fixing recurring query errors caused by ambiguous entities, timezones, or NULLs
• Adding a new domain (marketing, finance, product) to an existing data knowledge base

Best practices

• Start with the 3–5 most-used tables per domain to get high-leverage coverage quickly
• Ask entity-disambiguation and primary-identifier questions early to avoid incorrect joins
• Capture exact metric formulas, source tables, and time-window conventions (e.g., trailing 7d vs calendar month)
• Document standard exclusion filters (test users, internal traffic, fraud) as mandatory query clauses
• Include 2–3 sample queries per table or metric to illustrate correct usage and common patterns

Example use cases

• Create a new company-data skill that maps 'user' vs 'account' and includes sample joins and IDs
• Add finance metrics and formulas (ARR, MRR, churn) into an existing skill to standardize reporting
• Document marketing tables and attribution rules so analysts stop double-counting conversions
• Fix timezone and NULL handling issues by adding clear caveats and filtering rules to table docs
• Produce a zipped package of reference files for handoff to BI, ML, or new hires