GitHub Copilot Novice tutorial

Novice · step 1 of 7

The 5-part prompt formula

Role, context, task, format, constraints.

Role: you're helping a small-business owner.
Context: [paste facts].
Task: [one clear job].
Format: [table / 5 bullets / 120 words].
Constraints: say "unknown" instead of guessing; cite sources.

A good result: the answer follows the format and length you set, and the audience and goal are reflected in its wording. Missing parts in the prompt show up as generic parts in the answer.

Why this step matters: The same model gives much better answers to a specific prompt. This is the cheapest improvement you can make.

Novice · step 2 of 7

Organize your work

Repository-wide custom instructions live in .github/copilot-instructions.md; path-specific instructions apply to matching files (GitHub docs).

Why this step matters: Grouping chats and files keeps context together, so you stop re-explaining and re-uploading.

Novice · step 3 of 7

Files and research

It works on the files in your open project and repository.

Ask it to explain code and list edge cases before changing anything.

Why this step matters: Grounding answers in your own files or live sources cuts made-up details and makes them checkable.

Novice · step 4 of 7

Limits you'll hit

Agent mode, code review, Copilot in the cloud, the Copilot CLI and the Copilot app use AI credits; 1 credit = $0.01 (GitHub).

• Start a new chat when a thread gets long; paste a short summary.
• Don't re-upload the same big files every time.

Why this step matters: Knowing when limits reset and what drains them stops you getting cut off mid-task.

Novice · step 5 of 7

Two-source rule

For anything you'll act on, confirm with two independent sources, one of them official.

1. Ask for the source of any fact you'll reuse.
2. Open the link and find the exact sentence (Ctrl+F).
3. If you can't find it, treat the claim as unverified.
4. For files, check two page or row references yourself.

Before you answer: separate what you checked from what you're assuming. Write "unknown" instead of guessing. Give a link for every number.

Why this step matters: Chatbots state wrong facts in the same confident tone as right ones. Checking is the only way to tell them apart.

Novice · step 6 of 7

Bad vs good prompts

Bad: help me

Good: Write pytest tests for parse_invoice() in invoices.py: the happy path, a missing total, a negative amount, and a DD/MM date. Don't change the function.

Why: a clear situation, task, format and limits.

Bad: is this right?

Good: Check this against the official page. Quote the sentence that supports it, or say it isn't there.

Why: asks for evidence instead of agreement.

Why this step matters: The same model gives much better answers to a specific prompt. This is the cheapest improvement you can make.

Novice · step 7 of 7

Three starter workflows

Why this step matters: Small steps with a checked output each fail visibly and early, instead of one big answer that is wrong somewhere.