VS Code + GitHub Copilot Workflows¶

Practical workflow patterns for real development tasks using Copilot’s three modes: completions, chat, and agent.

1. Understanding the Three Modes¶

Mode	Trigger	Best for
Completions	Just type — suggestions appear inline	Boilerplate, repetitive patterns, line-by-line edits
Chat	Open Copilot Chat panel (⌘⇧I)	Questions, explanations, planning, code review
Agent	@workspace in chat, or agent mode toggle	Multi-file features, refactoring, test generation, debugging

When to Use Each¶

Simple rename / one-liner    → Completions
"How does this function work?" → Chat
"Add auth middleware to all routes" → Agent

2. Feature Implementation Workflow¶

Scenario: Add a new “Parent-Child Retrieval” notebook to the RAG chapter.

Step 1: Plan with Chat¶

Open Copilot Chat and ask:

@workspace What RAG notebooks exist in 08-rag/? What numbering scheme do they use?
I want to add a parent-child retrieval notebook.

Copilot scans the workspace and returns the existing notebook list and naming convention.

Step 2: Build with Agent Mode¶

Switch to agent mode and describe the full task:

Create 08-rag/12_parent_child_retrieval.ipynb following the existing notebook
conventions in 08-rag/. The notebook should:
1. Explain parent-child retrieval with a diagram
2. Implement chunking with parent references
3. Show retrieval that expands from child chunks to parent docs
4. Compare against flat retrieval with precision and recall metrics
5. Include a summary comparison table

Use self-contained code with TF-IDF (no API keys).
Run all cells to verify they execute.

Agent mode will:

Read existing notebooks to learn the conventions
Create the notebook with proper structure
Execute cells and fix any errors
Report back with results

Step 3: Review with Chat¶

Select the generated code and ask:

/explain this retrieval implementation — are there edge cases I'm missing?

3. Debugging Workflow¶

Scenario: Tests are failing after a refactor.

Step 1: Get Context¶

@terminal What are the test failures?

Or paste the error directly into chat.

Step 2: Fix with Agent¶

The tests in tests/api/test_users.py are failing with a 422 error.
The endpoint was recently refactored. Fix the failing tests to match
the new endpoint signatures.

Agent reads the test file, the route file, the schemas, and the error — then fixes the mismatch.

Step 3: Verify¶

Agent mode runs the tests automatically. If they still fail, it iterates.

4. Code Review Workflow¶

Quick Explanation¶

Select code, then:

/explain

Targeted Review¶

@workspace Review src/services/payment.py for:
SQL injection risks
Missing error handling
Race conditions in the checkout flow

Review a Diff¶

Open the Source Control panel, select changed files, and ask:

Review these changes. Flag any breaking changes, missing tests,
or security issues.

5. Test Generation Workflow¶

Basic Test Generation¶

Select a function, then:

/tests

Comprehensive Test Suite¶

Use agent mode for thorough coverage:

Generate tests for src/api/routes/orders.py.
Include:
- Happy path for each endpoint
- 401 unauthorized
- 422 validation errors
- 404 not found
- Edge cases (empty list, max pagination)

Use the existing conftest.py fixtures and follow the test patterns
in tests/api/test_users.py.

6. MCP-Connected Workflows¶

MCP servers give Copilot access to external tools. Here are workflows that combine Copilot with MCP.

Database → Code Workflow¶

With a PostgreSQL MCP server connected:

@workspace Look at the database schema for the orders table.
Generate a SQLAlchemy model, Pydantic schemas, and CRUD routes
that match the actual table structure.

Copilot reads the live schema via MCP and generates code that matches exactly.

Browser Testing Workflow¶

With the Playwright MCP server connected:

Navigate to http://localhost:3000/checkout and fill in the form
with test data. Take a screenshot after each step.
Then verify the order appears in the database.

Documentation Workflow¶

With a fetch/web MCP server:

Fetch the API documentation from https://api.example.com/docs
and generate TypeScript types that match the response schemas.

7. Refactoring Workflow¶

Extract and Reorganize¶

@workspace The file src/services/orders.py is 800 lines long.
Split it into:
- src/services/orders/create.py
- src/services/orders/search.py
- src/services/orders/fulfillment.py

Update all imports across the codebase. Run tests after to verify
nothing broke.

Migrate a Pattern¶

@workspace Find all uses of os.getenv() in src/ and replace them
with pydantic-settings. The settings class is in src/config.py.

8. Documentation Workflow¶

Generate Docs for Existing Code¶

Select a module and use:

/doc

Generate a Full README¶

@workspace Generate a README.md for the 08-rag/ directory.
Include:
- Chapter overview
- Prerequisites
- List of all notebooks with one-line descriptions
- Suggested study order
- Link to the technique selection guide

9. Workflow Cheat Sheet¶

Task	Mode	Example prompt
Quick fix	Completions	Just type the fix — Copilot suggests the rest
Explain code	Chat	`/explain` on selection
Find a function	Chat	`@workspace Where is the retry logic?`
New feature	Agent	`Add rate limiting to all /api/ routes...`
Fix failing tests	Agent	`@terminal Fix these test failures`
Generate tests	Agent	`/tests` or detailed prompt
Refactor	Agent	`Split this 500-line file into modules...`
Code review	Chat	`Review this for security issues`
Documentation	Chat/Agent	`/doc` or `Generate README for...`
Database → code	Agent + MCP	`Read the users table schema and generate models`

10. Tips for Effective Prompts¶

Be Specific¶

Bad: Make this better

Good: Refactor this function to use async/await instead of callbacks. Keep the same function signature.

Provide Context¶

Bad: Fix the bug

Good: The /api/orders endpoint returns 500 when the cart is empty. The error is in the calculate_total function.

Reference Files¶

Bad: Update the tests

Good: Update tests/api/test_orders.py to cover the new discount_code parameter added in src/api/routes/orders.py

Iterate¶

Don’t expect perfection on the first try. Review the output and follow up:

Good start, but:
1. Use the existing `create_test_order` fixture instead of inline setup
2. Add a test for the edge case where discount > total

Previous: 03_copilot_instructions_guide.md — custom instructions for Copilot Next: 05_build_mcp_server.ipynb — hands-on: build your own MCP server