GitHub Actions Toolkit
A toolkit for building GitHub Actions in Node.js
Usage •
API •
How to test your Action •
FAQ
Motivation
actions-toolkit is a wrapper around some fantastic open source libraries, and provides some helper methods for dealing with the GitHub Actions runtime. Actios all run in Docker containers, so this library aims to help you focus on your code and not the runtime. You can learn more about building Actions in Node.js to get started!
After building a GitHub Action in Node.js, it was clear to me that I was writing code that other actions will want to use. Reading files from the repository, making requests to the GitHub API, or running arbitrary executables on the project, etc.
So, I thought it'd be useful to build those out into a library to help you build actions in Node.js :tada:
Usage
Installation
$ npm install actions-toolkit
const { Toolkit } = require('actions-toolkit')
const tools = new Toolkit()
Bootstrap a new action
$ npx actions-toolkit my-cool-action
This will create a new folder my-cool-action
with the following files:
├── Dockerfile
├── index.js
├── index.test.js
└── package.json
API
Toolkit options
event (optional)
An optional list of events that this action works with. If omitted, the action will run for any event - if present, the action will exit with a failing status code for any event that is not allowed.
const tools = new Toolkit({
event: ['issues', 'pull_requests']
})
You can also pass a single string:
const tools = new Toolkit({
event: 'issues'
})
And/or strings that include an action (what actually happened to trigger this event) for even more specificity:
const tools = new Toolkit({
event: ['issues.opened']
})
secrets (optional)
You can choose to pass a list of secrets that must be included in the workflow that runs your Action. This ensures that your Action has the secrets it needs to function correctly:
const tools = new Toolkit({
secrets: ['SUPER_SECRET_KEY']
})
If any of the listed secrets are missing, the Action will fail and log a message.
Toolkit.run
Run an asynchronous function that receives an instance of Toolkit
as its argument. If the function throws an error (or returns a rejected promise), Toolkit.run
will log the error and exit the action with a failure status code.
The toolkit instance can be configured by passing Toolkit
options as the second argument to Toolkit.run
.
Toolkit.run(async tools => {
}, { event: 'push' })
tools.github
Returns an Octokit SDK client authenticated for this repository. See https://octokit.github.io/rest.js for the API.
const newIssue = await tools.github.issues.create({
...tools.context.repo,
title: 'New issue!',
body: 'Hello Universe!'
})
You can also make GraphQL requests:
const result = await tools.github.graphql(query, variables)
tools.log
This library comes with a slightly-customized instance of Signale, a great logging utility. Check out their docs for the full list of methods. You can use those methods in your action:
tools.log('Welcome to this example!')
tools.log.info('Gonna try this...')
try {
risky()
tools.log.success('We did it!')
} catch (error) {
tools.log.fatal(error)
}
In the GitHub Actions output, this is the result:
ℹ info Welcome to this example!
ℹ info Gonna try this...
✖ fatal Error: Something bad happened!
at Object.<anonymous> (/index.js:5:17)
at Module._compile (internal/modules/cjs/loader.js:734:30)
tools.command(command, (args, match) => Promise)
Respond to a slash-command posted in a GitHub issue, comment, pull request, pull request review or commit comment. Arguments to the slash command are parsed by minimist. You can use a slash command in a larger comment, but the command must be at the start of the line:
Hey, let's deploy this!
/deploy --app example --container node:alpine
tools.command('deploy', async (args: ParsedArgs, match: RegExpExecArray) => {
console.log(args)
})
The handler will run multiple times for each match:
/deploy 1
/deploy 2
/deploy 3
let i = 0
await tools.command('deploy', () => { i++ })
console.log(i)
tools.config(filename)
Get the configuration settings for this action in the project workspace. This method can be used in three different ways:
const cfg = tools.config('.myactionrc')
const cfg = tools.config('myaction.yml')
const cfg = tools.config('myaction')
If the filename looks like .myfilerc
it will look for that file. If it's a YAML file, it will parse that file as a JSON object. Otherwise, it will return the value of the property in the package.json
file of the project.
tools.getPackageJSON()
Get the package.json file in the project root and returns it as an object.
const pkg = tools.getPackageJSON()
tools.getFile(path, [encoding = 'utf8'])
Get the contents of a file in the repository.
const contents = tools.getFile('example.md')
tools.runInWorkspace(command, [args], [ExecaOptions])
Run a CLI command in the workspace. This uses execa under the hood so check there for the full options. For convenience, args
can be a string
or an array of string
s.
const result = await tools.runInWorkspace('npm', ['audit'])
tools.arguments
An object of the parsed arguments passed to your action. This uses minimist
under the hood.
When inputting arguments into your workflow file (like main.workflow
) in an action as shown in the Actions Docs, you can enter them as an array of strings or as a single string:
args = ["container:release", "--app", "web"]
# or
args = "container:release --app web"
In actions-toolkit
, tools.arguments
will be an object:
console.log(tools.arguments)
There is currently a known bug with strings with multiple word arguments being parsed incorrectly. Arguments that contain strings with multiple words need to currently pass the arguments in an array:
args = [ "To do", "100" ]
Or have a different action entrypoint file:
#!/bin/sh
sh -c "npm $*"
tools.token
The GitHub API token being used to authenticate requests.
tools.workspace
A path to a clone of the repository.
tools.exit
A collection of methods to end the action's process and tell GitHub what status to set (success, neutral or failure). Internally, these methods call process.exit
with the appropriate exit code. You can pass an optional message to each one to be logged before exiting. This can be used like an early return:
if (someCheck) tools.exit.neutral('No _action_ necessary!')
if (anError) tools.exit.failure('We failed!')
tools.exit.success('We did it team!')
tools.store
Actions can pass information to each other by writing to a file that is shared across the workflow. tools.store
is a modified instance of flat-cache
:
Store a value:
tools.store.set('foo', true)
Then, in a later action (or even the same action):
const foo = tools.store.get('foo')
console.log(foo)
Note: the file is only saved to disk when the process ends and your action completes. This is to prevent conflicts while writing to file. It will only write to a file if at least one key/value pair has been set. If you need to write to disk, you can do so with tools.store.save()
.
tools.context
tools.context.action
The name of the action
tools.context.actor
The actor that triggered the workflow (usually a user's login)
tools.context.event
The name of the event that triggered the workflow
tools.context.payload
A JSON object of the webhook payload object that triggered the workflow
tools.context.ref
The Git ref
at which the action was triggered
tools.context.sha
The Git sha
at which the action was triggered
tools.context.workflow
The name of the workflow that was triggered.
tools.context.issue
The owner
, repo
, and number
params for making API requests against an issue or pull request.
tools.context.repo
The owner
and repo
params for making API requests against a repository. This uses the GITHUB_REPOSITORY
environment variable under the hood.
How to test your GitHub Actions
Similar to building CLIs, GitHub Actions usually works by running a file with node <file>
; this means that writing a complete test suite can be tricky. Here's a pattern for writing tests using actions-toolkit, by mocking Toolkit.run
:
index.js
const { Toolkit } = require('actions-toolkit')
Toolkit.run(async tools => {
tools.log.success('Yay!')
})
index.test.js
const { Toolkit } = require('actions-toolkit')
describe('tests', () => {
let action
beforeAll(() => {
Toolkit.run = fn => { action = fn }
require('./index.js')
})
it('logs successfully', async () => {
const fakeTools = new Toolkit()
fakeTools.log.success = jest.fn()
await action(fakeTools)
expect(fakeTools.log.success).toHaveBeenCalled()
})
})
You can then mock things by tweaking environment variables and redefining tools.context.payload
. You can check out this repo's tests as an example.
FAQ
Can you get me into the GitHub Actions beta?
I'm sorry, but I cannot.
Aren't these just wrappers around existing functions?
Yep! I just didn't want to rewrite them for my next Action, so here we are.