chapter02

Chapter02

Handle different kind of intents. Handle multiple acts with visual response

In this chapter, let's create skill which recognize multiple user intents and respond with multiple acts.

Let's see src/index.ts. In this chapter, GameControlManager is the ControlManager. It sets RootControl as root of control tree.

Handle different kind of intents

RootControl is as below. Different from chapter01, its canHandle and handle methods are a bit complex. But same rule there. handle will be called if canHandle returns true.

class RootControl extends ContainerControl {
    handleFunc: (
        input: ControlInput,
        resultBuilder: ControlResultBuilder,
    ) => Promise<void>;

    async canHandle(input: ControlInput) {
        if (InputUtil.isIntent(input, 'PlayLookThisWayIntent')) {
            this.handleFunc = this.handlePlayLookThisWayIntent;
            return true;
        } else if (InputUtil.isIntent(input, 'PlayRockPaperScissorsIntent')) {
            this.handleFunc = this.handlePlayRockPaperScissorsIntent;
            return true;
        } else if (InputUtil.isLaunchRequest(input)) {
            this.handleFunc = this.handleLaunchRequest;
            return true;
        }

        return false;
    }

    private async handleLaunchRequest(
        input: ControlInput,
        resultBuilder: ControlResultBuilder,
    ) {
        resultBuilder.addAct(
            new LiteralContentAct(this, {
                promptFragment: 'Hello',
            }),
        );
    }

    private async handlePlayLookThisWayIntent(
        input: ControlInput,
        resultBuilder: ControlResultBuilder,
    ) {
        resultBuilder.addAct(new PlayLookThisWayAct(this));
    }

    private async handlePlayRockPaperScissorsIntent(
        input: ControlInput,
        resultBuilder: ControlResultBuilder,
    ) {
        resultBuilder.addAct(new PlayRockPaperScissorsAct(this));
    }

    async handle(input: ControlInput, resultBuilder: ControlResultBuilder) {
        if (!InputUtil.isLaunchRequest(input)) {
            resultBuilder.addAct(
                new LiteralContentAct(this, {
                    promptFragment: "Okay, so let's start it...",
                }),
            );
        }

        await this.handleFunc(input, resultBuilder);
    }
}

As mentioned in chapter01, Control is not Intent Handler. Control sometimes includes some intent handlers. In this chapter, RootControl can respond to PlayLookThisWayIntent, PlayRockPaperScissorsIntent and LaunchRequest in different ways for each.

For example, if request is PlayLookThisWayIntent, RootControl sets handlePlayLookThisWayIntent method to handleFunc property. And then, it will be called in handle method. if request is PlayRockPaperScissorsIntent, handlePlayRockPaperScissorsIntent will be called.

In this way, you can respond to different kind of intents.

Handle multiple acts

As the name addPromptFragment implies, you can add prompts in pieces.

Let's see test/index.test.ts and you can see code as below.

    ...
    await tester.testTurn(
        "U: Let's play rock paper scissors",
        TestInput.of('PlayRockPaperScissorsIntent'),
        "A: Okay, so let's start it... Rock, paper, scissors, one two three",
    );
    ...

Alexa responds to Okay, so let's start it... Rock, paper, scissors, one two three to PlayRockPaperScissorsIntent intent. Let's see the flow to build this prompt.

As described as above, handlePlayRockPaperScissorsIntent will be called if request is PlayRockPaperScissorsIntent intent. In this method, PlayRockPaperScissorsAct act is added to resultBuidler.

    ...
    private async handlePlayRockPaperScissorsIntent(
        input: ControlInput,
        resultBuilder: ControlResultBuilder,
    ) {
        resultBuilder.addAct(new PlayRockPaperScissorsAct(this));
    }
    ...

PlayRockPaperScissorsAct is as below. It adds Rock, paper, scissors, one two three message as prompt fragment. and adds return value of simpleTextWithDelayDocument as APL Document.

APL (Amazon Presentaion Language) is the language to build visual experience for Alexa Skill Kit. In this APL document, just show the hand randomly after a bit delay.

class PlayRockPaperScissorsAct extends ContentAct {
    render(input: ControlInput, responseBuilder: ControlResponseBuilder) {
        responseBuilder.addPromptFragment(
            'Rock, paper, scissors, one two three',
        );

        const hands = ['Rock', 'Paper', 'Scissors'];
        responseBuilder.addAPLRenderDocumentDirective(
            'PlayRockPaperScissorsTemplate',
            simpleTextWithDelayDocument(
                hands[Math.ceil(Math.random() * hands.length)],
            ),
        );
    }
}

In this act, only Rock, paper, scissors, one two three is added as prompt. So when Okay, so let's start it... added?

You can see it in handle method.

    async handle(input: ControlInput, resultBuilder: ControlResultBuilder) {
        if (!InputUtil.isLaunchRequest(input)) {
            resultBuilder.addAct(
                new LiteralContentAct(this, {
                    promptFragment: "Okay, so let's start it...",
                }),
            );
        }

        await this.handleFunc(input, resultBuilder);
    }

In handle method, Okay, so let's start it... is added by using LiteralContent. LiteralContentAct is the kind of system act defined by framework. You can use it when you want to just add simple texts to response.

In other words, If request is PlayRockPaperScissorsIntent, LiteralContentAct is added in handle method at first. And then handlePlayRockPaperScissorsIntent in this.handleFunc is called and PlayRockPaperScissorsAct is added.

Finally, these acts are concatenated. By using it, you can implement building response logic to flexibly and DRY.

Let's deploy and run it.

See chapter01.

And I prepared the recognition model to interactionModel.json. You can build interaction model by using it.