Getting Help with Command Inputs from GitHub Copilot in the CLI
To reach a broader audience, this article has been translated from Japanese.
You can find the original version here.
Introduction
#GitHub Copilot functions as an extension for code editors like VS Code. By using GitHub Copilot in the CLI, you can also receive Copilot's assistance when entering commands in the terminal.
It has been introduced in GitHub's Blog with a video.
GitHub Copilot in the CLI is provided as an extension (gh-copilot) to the GitHub CLI (gh). For brevity, it will be referred to as gh-copilot henceforth.
With gh-copilot, you can ask questions in a chat-like interface on the terminal and receive answers. The questions are sent to GitHub's API, analyzed by Copilot's language model, and answered.
The documentation for gh-copilot can be found here:
gh-copilot is in public beta at the time of writing this article. To use it, you need a GitHub Copilot Individual subscription. I am using an Individual subscription. If the organization you belong to has a GitHub Copilot Business subscription and has activated gh-copilot, it can also be used.
(Added on 2024.03.22)
gh-copilot has become generally available (GA), and can now be used with all Individual, Business, and Enterprise Copilot plans.
Installation
#You need to install GitHub CLI beforehand. For instructions on installing GitHub CLI, refer to the Installation section in the GitHub repository README. On macOS, you can install it with brew install gh.
You must be logged into GitHub with GitHub CLI. To check if you are logged in, execute gh auth status. If logged in, it shows the status of your account, PAT, and the scopes of the PAT.
If you are not logged in, log in with gh auth login.
Then, install gh-copilot.
gh extension install github/gh-copilot
If you see the following output, the installation was successful.
✓ Installed extension github/gh-copilot
The first time you run gh copilot, you will be asked whether to allow GitHub to collect optional usage data to help us improve. Select Yes.
Main Commands of gh-copilot
#When you run gh-copilot, you will see a message and prompt like below.
gh-copilot can assist with the following three types of commands:
- Generic shell commands (shell)
- GitHub CLI commands (gh)
- Git commands (git)
The main commands of gh-copilot are the following two:
- suggest: Get suggestions for commands
- explain: Get explanations for commands
For suggest, you pass what you want to do in natural language.
For explain, you pass the command you want to execute, including arguments and options.
Below is a screenshot of using suggest (click to enlarge).
- You are asking "get kubernetes cluster information."
- You select
generic shell commandas the command type. - As a result, the command
kubectl get cluster-infowas suggested. - By using
Revise, you can modify the question or add additional information.- After adding "with yaml output" as a condition, you got
kubectl get cluster-info -o yaml.
- After adding "with yaml output" as a condition, you got
- You selected
Copy command to clipboardto copy the obtained command to the clipboard.
If selecting the command type every time is cumbersome, you can directly get the suggestion result by specifying the command type with -t.
For detailed usage, display the usage with gh copilot suggest --help or ask directly with gh copilot suggest.
Using explain on a Suggestion
#explain can also be used on suggestion results. After getting a suggestion result, select Explain command. In the following example, explain is applied to the suggestion result of the question "Terminate all processes whose names match a specific regular expression" to get explanations about the command and options.
Asking in Japanese
#Let's try asking in Japanese.
Normal responses are provided. Let's revise it since rmi is a bit old.
It looks good.
Let's ask something a bit more complicated.
The command generated from a vague instruction in Japanese seems usable.
GitHub Copilot's documentation states the following limitation:
GitHub Copilot in the CLI is trained primarily on natural language content written in English. As a result, when specifying prompts in languages other than English for GitHub Copilot in the CLI, performance may vary.
Currently, it seems that asking questions in English results in higher accuracy.
Conclusion
#This was an introduction to GitHub Copilot in the CLI. Currently, it has been trained on data up to 2021, so it does not support the latest versions of commands or commands released after 2022. Users are also responsible for checking for biases and inaccuracies inherent in LLM.
In the past, we used the man command or googled for CLI information, but it took quite some time for commands we don't frequently use.
For commands you have used before, reusing command history is the best way to utilize them. For efficient use of command history, please refer to the following article.
Nowadays, you can just ask ChatGPT and copy-paste, but having a dedicated UX in the terminal you're actually using is indeed convenient.
While it might feel like relying on Copilot could lead to a decline in technical skills, considering it as a new interface for using computers might be the right approach.
Having already used "natural language as a programming language," utilizing "natural language as commands" to quickly complete tasks seems like a good idea.
