Skip to content


In an Software Delivery Machine (SDM), a command is an action that can be triggered on demand, either from the command line in local mode, from chat in team mode, or in the Atomist web interface. This can be any action that you write in a TypeScript function.

What would you like your team to do more frequently or more consistently?

This page will show you how to:

  • create a command
  • test your command
  • respond to the person who invoked the command
  • wrap a shell script in a command
  • define parameters for your command

This page starts after you have created an SDM.

Create a command

First, we’ll write the function for our command. For this example, we will write the classic “hello, world” function. To be able to respond regardless of whether the command was invoked on the command line, in chat, or the web interface, we need to learn a little about what a command function looks like.

We call these command functions “listeners”, because they are always listening for someone to call them. All command listeners take a single argument, a CommandListenerInvocation, which provides some useful information to the command, like how to respond to the person who invoked the command.

Command listeners are asynchronous functions, returning a Promise. The Promise contains an object with at least a code property, whose value is a number. A code of zero means success. If the code is non-zero, the command execution is considered unsuccessful.

Armed with this information, we can write our command listener function.

import { HandlerResult, NoParameters } from "@atomist/automation-client";
import { CommandListenerInvocation } from "@atomist/sdm";
export async function helloWorldListener(ci: CommandListenerInvocation<NoParameters>): Promise<HandlerResult> {
    return ci.addressChannels("Hello, world");

You can see the CommandListenerInvocation has an addressChannels property, which sends a message to the appropriate places – in this case, to wherever the command was invoked. Like much of the API, this is an asynchronous function.

Register your command

The next thing to do is register your command in your SDM. First, we create a CommandHandlerRegistration.

import { CommandHandlerRegistration } from "@atomist/sdm";
const helloWorldCommand: CommandHandlerRegistration = {
    name: "HelloWorld",
    description: "Responds with a friendly greeting to everyone",
    intent: "hello",
    listener: async ci => {
        await ci.addressChannels("Hello, world");
        return { code: 0 };

We provide a unique name and description in the registration. The value of the intent property defines the command you enter to invoke the command listener function. Here we have defined our function inline. In the test, we would also have to update the invocation of the command to helloWorldCommand.listener(ci).

Once we have the registration, we can add the command to our SDM object.

import { Configuration } from "@atomist/automation-client";
import {
} from "@atomist/sdm";
import {
} from "@atomist/sdm-core";
function machine(configuration: SoftwareDeliveryMachineConfiguration): SoftwareDeliveryMachine {
    const sdm = createSoftwareDeliveryMachine({
        name: "My SDM",
    return sdm;
export const configuration: Configuration = {
    postProcessors: [configureSdm(machine)],

The configuration object should be exported from the index.ts of your SDM.

Run your command

If you are running your SDM in local mode, start the SDM as you normally would

atomist start --local

and then in another terminal run the hello command:

atomist hello

and the SDM should respond in the same terminal.

In team mode, once you have started your SDM you can send the intent to the Atomist bot

@atomist hello

and it will respond back to you in the channel.

In chat, you can send @atomist describe skill "hello" to see details of your command and its parameters.

Wrap a shell script

If you want to turn a shell script into a bot command, just call that shell script from a command listener. You can use the execPromise helper to capture the output and respond back with it.

import { execPromise } from "@atomist/automation-client";
import { CommandHandlerRegistration } from "@atomist/sdm";
const myScriptCommand: CommandHandlerRegistration = {
    name: "MyScript",
    description: "Run my-script and respond with its stdout",
    intent: "my script",
    listener: async ci => {
        const result = await execPromise("my-script", ["its", "args"]);
        return ci.addressChannels(result.stdout);

See also: Running external commands

Command parameters

Command parameters give you extra information that can be different each time the command runs.

By default, commands not require parameters. However, you can specify parameters to be gathered in chat by the Atomist bot or via the web interface or CLI. These parameters can be accessed in a typesafe manner.

To specify commands, use the optional parameters property when creating a command. Let’s make our hello command welcome us by name to wherever we are.

Defining parameters

Start by defining the parameters. A parameter definition object maps the name of each parameter to optional details. Pass it to the command registration so that it will know what parameters to gather.

const helloWorldParametersDefinition = {
      name: { description: "name", 
              required: true,
              pattern: /.*/, },
      location: {},

const helloWorldCommand: CommandHandlerRegistration<{ name: string, location: string }> = {
    name: "HelloWorld",
    description: "Responds with a friendly greeting to everyone",
    intent: "hello",
    parameters: helloWorldParametersDefinition,
    listener: async ci => {
        return ci.addressChannels(`Welcome to ${ci.parameters.location}, ${}`);

Atomist will now prompt anyone who invokes this command for name and location. In this case any value will be accepted, but we can use regular expressions to ensure that valid values are submitted.

The helloWorldParametersDefinition object has one property for each parameter. Each property’s name is the name of the parameter, and its value describes the parameter. These options are available in the parameter definition (all optional):

attribute type description default
description string short description same as name
pattern RegExp regular expression that the parameter’s value must match match any single line string (^.*$)
required boolean is the parameter required? false
displayName string name to display; may contain spaces same as name
validInput string describe what makes a valid parameter value blank
displayable boolean whether to show a parameter value after it’s been entered true
maxLength number maximum number of characters to accept no maximum
minLength number minmum number of characters needed 0
type ParameterType string, boolean, number, or Chooser string
order number when prompting, ask for smaller ‘order’ parameters first order doesn’t matter
group Group when prompting, put parameters in the same Group together none
control “input” or “textarea” input type for string parameters in dialogues “input”

We can combine parameter definitions using spreads. For example, this will also bring in some common parameters defined in another object constant:

parameters: {
    name: { description: "name", required: true, pattern: /.*/, },
    body: { description: "multi line body text", required: true, pattern: /[\s\S]*/, control: "textarea" }
    location: {},

The property definition mechanism applies to all commands, so you can apply it to code inspections and code transform commands.

Accessing parameters in the command

The CommandHandlerRegistration type is parameterized by the type of the properties object. In this case it’s the anonymous type { name: string, location: string }, but in more complex scenarios we’d use an interface. By default, no parameters are exposed.

We access the parameter values in the listener via the parameters property, as in

    listener: async ci => {
        return ci.addressChannels(`Welcome to ${ci.parameters.location}, ${}`);

Test your command

Testing your command listener requires that you mock the parts of CommandListenerInvocation you use and then call your function. Using the Mocha testing framework, it would look something like this.

import { NoParameters } from "@atomist/automation-client";
import { CommandListenerInvocation } from "@atomist/sdm";
import * as assert from "assert";
describe("helloWorldListener", () => {
    it("should respond successfully", async () => {
        let response: string;
        const ci: CommandListenerInvocation<NoParameters> = {
            addressChannels: async (m: string) => response = m,
        const result = await helloWorldListener(ci);
        assert(result.code === 0);
        assert(response === "Hello, world");

We assign the message sent to the response property so we can later confirm it was the message we expected.

What else would you like to do?

What is missing from this page? Please tell me! There’s a #docs channel in Atomist community Slack, or you can create an issue on this repository.