Skip to content

exercise-19-context-map.md

Latest commit

 

History

History
120 lines (76 loc) · 4.39 KB

exercise-19-context-map.md

File metadata and controls

120 lines (76 loc) · 4.39 KB

Exercise 19 — Create a Context Map Skill

Duration: 4 minutes
Copilot Feature: Skills (SKILL.md)
Goal: Install the Context Map skill from the awesome-copilot repository and use it to generate a codebase map that improves all subsequent Copilot interactions.

🟡 OPTIONAL EXERCISE

This exercise is not required to complete the mandatory track. The context map it produces gives Copilot richer codebase awareness, but Exercises 13–15 work without it. If you run this exercise, reference context-map.md in your prompts for Exercises 12 and 13.

Best after: Exercise 12  |  Return to mandatory track: Exercise 16 — Unit & Functional Tests

Background

A Skill is an on-demand workflow bundled in a SKILL.md file. Unlike instructions (always active) or agents (persona-driven), a skill is a specific multi-step process you invoke when you need it.

The Context Map skill generates a structured map of your entire codebase — all files, their purposes, relationships, and key patterns. When you attach this map to a Copilot prompt, Copilot gets instant, accurate context about your project without having to re-read every file.

Step 1 — Install the Context Map Skill

Create the skill directory and file:

  1. Create the folder .github/skills/context-map/
  2. Create the file .github/skills/context-map/SKILL.md

In Copilot Chat, send this prompt to fetch and save the skill:

Fetch the Context Map skill from this URL and save it to .github/skills/context-map/SKILL.md:
https://raw.githubusercontent.com/github/awesome-copilot/main/skills/context-map/SKILL.md

Then confirm the file was saved correctly by showing me its first 20 lines.

Alternative if network access is unavailable: Ask the instructor for the SKILL.md file, or copy it from: https://github.com/github/awesome-copilot/blob/main/skills/context-map/SKILL.md

Step 2 — Read the Skill

Open .github/skills/context-map/SKILL.md and read how the skill works. You'll see it instructs Copilot to systematically explore the codebase and generate a structured context document.

Step 3 — Run the Skill

In Copilot Chat (Agent mode, default agent), type / and look for context-map in the slash command list, or send:

Follow the instructions in .github/skills/context-map/SKILL.md to generate a context map for this project.

Save the output as .github/skills/context-map/context-map.md

The context map should cover:
- All directories and their purpose
- Key source files and what they do
- Main data models and their relationships
- API endpoints and their handlers
- Configuration patterns
- Test structure

Step 4 — Verify the Context Map

Open .github/skills/context-map/context-map.md. Check that it contains:

  • Directory tree with descriptions for each folder
  • Key files listed with their purpose
  • Main entities/models described
  • API routes listed with handler locations
  • Dependencies and their usage noted

Step 5 — Use the Context Map in a Prompt

Now test how the context map improves Copilot's responses. Send this prompt:

Using the context map in .github/skills/context-map/context-map.md, answer:
1. Which file handles task dependency blocking?
2. Where is the JWT authentication middleware?
3. What database tables are currently defined?
4. What's the response format for all API endpoints?

Compare the quality and accuracy of the answer vs. asking without the context map.

Why This Matters

In future prompts, prefix complex requests with:

Reference the context map at .github/skills/context-map/context-map.md for codebase context, then...

This single addition significantly reduces hallucinations (Copilot inventing file paths or function names) because it has accurate, up-to-date codebase knowledge.

Key Takeaway

Skills are reusable processes, not instructions or conversation. The Context Map skill is a good first skill for every project — it creates a shared mental model between you and Copilot about your codebase. You run it once (and re-run it as the codebase grows), and it pays dividends across every subsequent interaction.

Next optional: Exercise 20 — Database & SQL / PL/SQL

Return to Mandatory Track →: Exercise 16 — Unit & Functional Tests