ghc-copilot-foundations

name

ghc-copilot-foundations

description

GitHub Copilot track project #1. Learner installs the Copilot extensions, signs in, and writes a tiny calculator function with inline completions — learning Tab to accept, Esc to reject, Alt+] / Alt+[ to cycle, Ctrl+→ to accept word-by-word, and how to read ghost text. Auto-load when the learner is in `github-copilot/ghc-copilot-foundations` or asks to install Copilot, set up Copilot, learn the Copilot loop, or understand inline completions vs chat vs edit mode.

Project: ghc-copilot-foundations

Track: GitHub Copilot · Project: 1 of 9 · Time: ~60 minutes

The first time you press Tab on a Copilot suggestion that just works, you'll think it's magic. The second time, when it accepts a confidently wrong answer, you'll think it's broken. Both reactions skip the part where you learn to drive. This project teaches the loop — accept, reject, cycle, partial-accept — on a small calculator you build alongside.

Project goal

When this project is done, the learner can:

• Install GitHub Copilot and Copilot Chat in VS Code and sign in.
• Read the difference between inline completions (ghost text in the editor) and Copilot Chat (side panel conversation) and edit mode (multi-file changes).
• Use the four keystrokes every Copilot user must own: Tab (accept), Esc (reject), Alt+] / Alt+[ (cycle alternates), Ctrl+→ (accept word-by-word).
• Open the Copilot completions panel (Ctrl+Enter) to see multiple suggestions at once.
• Articulate when to use inline vs chat vs edit — in one sentence each, at interview level.

Scope guardrail

This is one tiny project, four keystrokes, three modes. We are not customizing Copilot (project #6), not writing prompts (project #7), not building agents (project #8). The point: the basic loop has to be muscle memory before anything else makes sense.

If the learner asks "how do I make Copilot follow my team's style?" — answer honestly: that's project #6. First master the keystrokes; you'll appreciate why customization matters once you've felt Copilot guess wrong.

Prerequisites

Prereq	Verify with
GitHub account with a Copilot license (Free, Individual, or Business)	github.com/settings/copilot shows "Copilot enabled"
VS Code 1.85+ installed	code --version
Python 3.10+ (used for the calculator example)	python --version

Phases

Phase 1 — Install, sign in, sanity check (~10 min)

Goal: Both Copilot extensions installed and signed in.

Steps:

1. VS Code → Extensions → search "GitHub Copilot" → Install. This installs both:
   • GitHub.copilot (inline completions engine)
   • GitHub.copilot-chat (the side-panel chat + slash commands + agents)
2. Bottom-right status bar → click the Copilot icon → Sign in to GitHub.
3. Browser opens → authorize the extension → return to VS Code.
4. Sanity check: create a file calc.py and type:

   def add(a, b):
   

   Pause for ~1 second. You should see ghost text appear suggesting return a + b. Press Tab to accept. If you see ghost text, you're done with setup.

Concepts to name out loud:

• This is ghost text — Copilot's suggestion overlaid on your editor in faded grey. Until you press Tab, nothing is committed.
• This is two extensions, one product — copilot does the inline completions, copilot-chat does the conversational side panel. You want both. They share auth but are independently updatable.
• This is per-language enablement — Copilot can be disabled for specific languages (e.g. you don't want completions while writing Markdown emails). Status bar icon → settings.

Common gotchas:

• No ghost text appears → check the status bar icon. If it shows a slash, Copilot is disabled for this file type. Click → enable.
• "Free trial limit reached" → you're on the Free tier and hit the monthly cap (a few thousand completions). Upgrade to Individual ($10/mo) or wait until the next billing cycle.
• Sign-in loop → browser is signed into a different GitHub account than the one with Copilot. Sign out, retry with the right account.

After-action prompt: "You installed two extensions. Why two? What does each do, and what would break if you only had one?"

Phase 2 — The four keystrokes (~15 min)

Goal: The learner has used Tab, Esc, Alt+] / Alt+[, and Ctrl+→ deliberately at least twice each.

Set up the playground — append to calc.py:

def subtract(a, b):

Drill 1 — Tab to accept: Pause. Ghost text appears. Press Tab. Code is committed.

Drill 2 — Esc to reject: Type a new line:

def multiply(a, b):

Ghost text appears. Press Esc. Ghost text disappears. Nothing committed.

Drill 3 — Alt+] to cycle to the next alternate: Type:

def divide(a, b):

Ghost text appears (likely return a / b). Press Alt+] — a different suggestion appears (likely with a zero-check). Press Alt+] again — another alternate. Press Alt+[ to go back. Press Tab when you like one.

Drill 4 — Ctrl+→ to accept word-by-word: Type:

def power(base,

Ghost text suggests something like exponent): return base ** exponent. Instead of accepting the whole thing, press Ctrl+→ — only the next word is accepted. Press again — the next word. This is how you take what's useful and leave the rest.

Concepts to name out loud:

• This is the four-keystroke loop — every working Copilot user does these without thinking. Practice them on purpose now, you'll save thousands of bad accepts later.
• This is why cycling matters — Copilot's first suggestion isn't always its best. The cycle key (Alt+]) shows you alternates that often differ meaningfully (e.g. with/without error handling, recursive vs iterative).
• This is word-by-word accept as the "I'll edit it anyway" superpower — when the suggestion is 80% right, take the 80% with Ctrl+→ and stop before the wrong part. Beats accepting then deleting.

Common gotchas:

• Alt+] doesn't work → some keyboards have the ] key in a different position; check Keyboard Shortcuts → search "Copilot: Next Suggestion" to remap.
• Tab accepts the wrong thing → you accepted ghost text when you meant to indent. Esc cancels ghost text; you can re-trigger it with Alt+\ if you want it back.

After-action prompt: "You used all four keystrokes. Which one was the most surprising — and when will you reach for it in real coding?"

Phase 3 — The completions panel (~10 min)

Goal: The learner has used Ctrl+Enter to open the completions panel and seen multiple full suggestions side-by-side.

Steps:

1. New function in calc.py:

   def factorial(n):
   

2. Pause for ghost text.
3. Press Ctrl+Enter — a new panel opens showing 5-10 full alternate completions.
4. Each completion is a different approach (iterative loop, recursive, with input validation, etc.).
5. Click "Accept Solution" on the one you like → it's inserted at the cursor.

Concepts to name out loud:

• This is the panel as the brainstorm view — when you don't know which approach is right, see multiple at once. Especially valuable for non-trivial functions where Copilot's first guess at intent might be wrong.
• This is a different cost — the panel calls the model more times to generate alternatives. Don't open it for every line. Reserve for genuine decision points.

After-action prompt: "You saw 5 alternate factorials. If two of them differ only by style (loop vs recursion), how do you decide which one to take?"

Phase 4 — Inline vs Chat vs Edit (~15 min)

Goal: The learner has opened each surface and used each at least once.

Inline (already comfortable): ghost text in the editor.

Chat — open the side panel (Ctrl+Alt+I on Windows, or click the Copilot icon in the activity bar):

• Type: Explain what the factorial function does and what edge cases it doesn't handle.
• Copilot replies in the side panel. You can ask follow-ups.
• Try a slash command: type /explain then highlight the factorial function in the editor → Copilot explains it.

Edit mode — multi-file changes (in the Chat panel, click the mode dropdown at the bottom → switch from Ask to Edit):

• Select calc.py in the working set.
• Prompt: Add type hints to every function in this file.
• Copilot proposes a diff across the file. Review the changes inline, accept or reject each.

Concepts to name out loud:

• This is inline as "complete what I'm typing" — fastest, no leaving the editor. Best for adding the next line, the next argument, the next test case.
• This is chat as "explain or design" — you stop typing code and talk. Best for "why is this broken," "what does this do," "design me an algorithm."
• This is edit as "change this code" — Copilot proposes a diff, you review. Best for refactors, multi-file changes, applying a pattern across a codebase.
• This is slash commands as shortcuts to common asks — /fix, /explain, /tests, /doc. Faster than typing the full prompt.

Common gotchas:

• Chat panel doesn't open → keyboard shortcut conflict. View → Open View → search "GitHub Copilot Chat".
• /explain works on nothing → you didn't have anything selected or open. Highlight code first.
• Edit mode applies changes you didn't expect → always review before accepting. The diff preview is non-destructive.

After-action prompt: "You used all three modes. Tell me in one sentence each: when inline beats chat, when chat beats edit, when edit beats inline."

Phase 5 — Build the calculator with intent (~10 min)

Goal: Use all four keystrokes and at least two of the three modes to finish the calculator.

Open the calculator file. You should have: add, subtract, multiply, divide, power, factorial.

Tasks:

1. Inline + Ctrl+→: Add a square_root(n) function. Accept the suggestion word-by-word, stop before any try/except (you'll add it deliberately).
2. Chat /fix: Highlight divide → in chat: /fix this function to handle division by zero. Accept the fix.
3. Edit mode: In Edit mode, prompt: Add a docstring to every function explaining what it does and what it returns. Keep them one line.
4. Verify: open a Python REPL, import calc, call each function. Confirm they work.

Concepts to name out loud:

• This is picking the right surface for each move — adding the next function is inline. Fixing one thing is /fix. Touching every function is Edit mode. Same product, three jobs.
• This is always running the code before declaring done — Copilot's confidence is unrelated to its correctness. Run it.

After-action prompt: "Your calculator works. What did Copilot get right first try, and what did it get wrong? What does the wrong-list tell you about where to be skeptical?"

When to break the method

• Learner is on Mac → keyboard shortcuts differ (Cmd not Ctrl for some). Show them the Keyboard Shortcuts view; search "Copilot" to see every binding for their OS.
• Learner already uses Copilot daily → skip phases 1-2, jump to phase 4 (the modes comparison is where most daily users have gaps).
• Time short → phases 1-2-4 are the must-do. Phase 3 (completions panel) and phase 5 (build the calculator) are reinforcement.

Definition of done

Observable, the learner can:

• Show ghost text appearing in calc.py and accept it with Tab.
• Cycle through at least 2 alternate suggestions with Alt+] / Alt+[.
• Accept a partial suggestion with Ctrl+→.
• Open the Copilot Chat panel and run /explain on a highlighted function.
• Run Edit mode on calc.py and accept a diff that touches multiple functions.
• Explain in one sentence each: ghost text, inline vs chat vs edit, slash command, completions panel.

Next project

→ ghc-prompting-for-completions — now that the loop is muscle memory, learn to steer completions on purpose: comments-as-prompts, signatures-as-prompts, naming-as-prompts. Five small functions, each prompted deliberately.