How do I maintain state between a dispatch bot and skill bot? - botframework

Question

I am using the virtual assistant template provided via Microsoft's botframework-solutions Github repository, and I am having trouble maintaining the active dialog between our main entry point -- a bot developed that implements the dispatch model to determine which skill to send the user's utterance for further handling -- and the individual skill processes the user's input.
I have our dispatch model bot listening to POSTs made to localhost:3979/api/messages using the route method in mainDialog.ts to determine the relevant skill to pass the users' utterance to, which takes that utterance and determines which dialog within the skill should be used to handle the users' utterance. When we begin a dialog that implements a multi-step waterfall dialog, the dispatch model bot is not keeping track of the dialog that is active within the skill bot, and therefore the skill bot doesn't know how to route the users' utterance to the already active dialog. When we are POSTing directly to the skill bot and foregoing the dispatch model bot, the skill bot is able to keep track of the active dialog, which leads me to believe that we aren't properly managing state between the dispatch bot and the skill bot.
I've noticed that bot the virtual assistant and skill templates we're using from the botframework-solutions templates instantiate a new instance of AutoSaveStateMiddleWare within their defaultAdapter.ts, so it appears that reading and writing of conversation and user state should already be automatically managed.
mainDialog.ts within the dispatch bot, which routes utterances to the appropriate skill
protected async route(dc: DialogContext): Promise<void> {
// Get cognitive models for locale
const locale: string = i18next.language.substring(0, 2);
const cognitiveModels: ICognitiveModelSet | undefined = this.services.cognitiveModelSets.get(locale);
if (cognitiveModels === undefined) {
throw new Error('There is no value in cognitiveModels');
}
// Check dispatch result
const dispatchResult: RecognizerResult = await cognitiveModels.dispatchService.recognize(dc.context);
const intent: string = LuisRecognizer.topIntent(dispatchResult);
if (this.settings.skills === undefined) {
throw new Error('There is no skills in settings value');
}
// Identify if the dispatch intent matches any Action within a Skill if so, we pass to the appropriate SkillDialog to hand-off
const identifiedSkill: ISkillManifest | undefined = SkillRouter.isSkill(this.settings.skills, intent);
if (identifiedSkill !== undefined) {
// We have identified a skill so initialize the skill connection with the target skill
await dc.beginDialog(identifiedSkill.id);
// Pass the activity we have
const result: DialogTurnResult = await dc.continueDialog();
if (result.status === DialogTurnStatus.complete) {
await this.complete(dc);
}
} else if (intent === 'l_NOVA_general') {
// If dispatch result is general luis model
const luisService: LuisRecognizerTelemetryClient | undefined = cognitiveModels.luisServices.get(this.luisServiceGeneral);
if (luisService === undefined) {
throw new Error('The specified LUIS Model could not be found in your Bot Services configuration.');
} else {
const result: RecognizerResult = await luisService.recognize(dc.context);
if (result !== undefined) {
const generalIntent: string = LuisRecognizer.topIntent(result);
// switch on general intents
switch (generalIntent) {
case 'Escalate': {
// start escalate dialog
await dc.beginDialog(EscalateDialog.name);
break;
}
case 'None':
default: {
// No intent was identified, send confused message
await this.responder.replyWith(dc.context, MainResponses.responseIds.confused);
}
}
}
}
} else {
// If dispatch intent does not map to configured models, send 'confused' response.
await this.responder.replyWith(dc.context, MainResponses.responseIds.confused);
}
}
The following code from the skill bot's dialogBot.ts which listens to all turn events
public async turn(turnContext: TurnContext, next: () => Promise<void>): Promise<any> {
// Client notifying this bot took to long to respond (timed out)
if (turnContext.activity.code === EndOfConversationCodes.BotTimedOut) {
this.telemetryClient.trackTrace({
message: `Timeout in ${ turnContext.activity.channelId } channel: Bot took too long to respond`,
severityLevel: Severity.Information
});
return;
}
const dc: DialogContext = await this.dialogs.createContext(turnContext);
if (dc.activeDialog !== undefined) {
const result: DialogTurnResult = await dc.continueDialog();
} else {
await dc.beginDialog(this.rootDialogId);
}
await next();
}
The mainDialog.ts within the skill bot that routes to the claims_claimsStatus waterfall dialog
protected async route(dc: DialogContext): Promise<void> {
// get current activity locale
const locale: string = i18next.language.substring(0, 2);
const localeConfig: Partial<ICognitiveModelSet> | undefined = this.services.cognitiveModelSets.get(locale);
// Populate state from SkillContext slots as required
await this.populateStateFromSkillContext(dc.context);
if (localeConfig === undefined) {
throw new Error('There is no cognitiveModels for the locale');
}
// Get skill LUIS model from configuration
if (localeConfig.luisServices !== undefined) {
const luisService: LuisRecognizerTelemetryClient | undefined = localeConfig.luisServices.get(this.solutionName);
if (luisService === undefined) {
throw new Error('The specified LUIS Model could not be found in your Bot Services configuration.');
} else {
let turnResult: DialogTurnResult = Dialog.EndOfTurn;
const result: RecognizerResult = await luisService.recognize(dc.context);
const intent: string = LuisRecognizer.topIntent(result);
switch (intent) {
case 'claims_claimStatus': {
turnResult = await dc.beginDialog(StatusDialog.name);
break;
}
.
.
.
Expected results: When using the dispatch bot to route to the skill bot, after beginning a waterfall dialog within the claims_claimStatus dialog, the dialogContext.activeDialog should be statusDialog for further steps of the waterfall dialog
Actual results: When using the dispatch bot to route to the skill bot, after beginning a waterfall dialog within the claims_claimStatus dialog, the dialogContext.activeDialog should is undefined for further steps of the waterfall dialog