Next-Level SQL in VS Code: GitHub Copilot Custom Instructions, Plan Mode & Skills | Data Exposed

Microsoft Developer walks through using GitHub Copilot with the VS Code MSSQL extension to control Copilot context for T-SQL—covering custom instruction files, Plan Mode/Agent Mode for schema design, skills as slash commands, and inspecting Copilot’s debug payload while targeting SQL Server 2025 and Azure SQL Database.

Overview

This Data Exposed episode shows how to make GitHub Copilot produce team-aligned T-SQL in VS Code by controlling the context Copilot receives via the MSSQL extension.

The demo flow starts with a clean workspace (no context), then incrementally adds:

• A custom instruction file to teach your team’s T-SQL conventions
• Plan Mode to design a data model from a PRD (Product Requirements Document)
• Agent Mode + Schema Designer to generate/shape schema while connected to a real database
• A skills file that encodes your vector search architecture as a slash command so Copilot can generate targeted SQL without repeating lengthy prompts
• The GitHub Copilot debug panel to inspect what Copilot actually sends to the model (system prompt + injected context)

What the episode demonstrates

1) Baseline: Copilot with zero context

• Start from a clean VS Code workspace
• Observe what Copilot generates for SQL when it has no knowledge of your team’s conventions

2) Add a custom instruction file for team T-SQL conventions

• Introduce a single instruction file that teaches Copilot how your team wants T-SQL written
• Goal: make “awesome SQL” become “awesome, but also consistent with our standards”

3) Use Plan Mode to design from a PRD

• Use Plan Mode to translate a product requirements document into a proposed data model

4) Use Agent Mode + Schema Designer against a real database

• Use Agent Mode to bring the planned model to life using Schema Designer
• Connect Copilot to your actual database so it can operate with real schema context

5) Create a skills file + slash command for your architecture

• Create a skills file that teaches Copilot your vector search approach
• Use it as a slash command so Copilot can generate relevant SQL quickly
• The episode calls out generating SQL Server 2025 / Azure SQL Database T-SQL without needing to restate the architecture in every prompt

6) Inspect Copilot’s debug payload

• Open the GitHub Copilot debug panel
• Review the full request payload sent to the LLM, including:
  • The system prompt
  • Any injected context (instructions/skills/schema context)

Resources (from the video description)

• Install VS Code MSSQL extension: https://aka.ms/vscode-mssql
• Demos: https://aka.ms/vscode-mssql-demos
• Blogs: https://aka.ms/vscode-mssql-blogs
• Documentation: https://aka.ms/vscode-mssql-docs

Episode timestamps (from the video description)

• 0:00 Introduction
• 1:33 Demo
• 8:45 Demo
• 17:16 Demo
• 18:33 Getting started

Where to follow / watch more

• Data Exposed episodes: https://aka.ms/dataexposedyt
• Microsoft Azure SQL channel: https://aka.ms/msazuresqlyt
• Microsoft SQL Server channel: https://aka.ms/mssqlserveryt
• Microsoft Developer channel: https://aka.ms/microsoftdeveloperyt
• Twitter (Anna Hoffman): https://twitter.com/AnalyticAnna
• Twitter (AzureSQL): https://aka.ms/azuresqltw