wolf-packs/wolf-core

🐺 Declarative development for state driven dynamic prompt flow

wolf-core's Issues

Move the stages from a directory structure into a file structure

Description:
right now all the stages are in its own directory with a index.ts. we could just have files under stages
Tasks:

run onComplete for every abilities completed on current turn

Is your feature request related to a problem? Please describe.
if a user fills in multiple slots on a single turn, (either through utterance or through slotData getter function), the bot should be able to run the onComplete functions on all abilities that are completed.

Describe the solution you'd like
the result of runOnComplete should return an array of messages for the middleware to add.
and the addMessage now should be addMessages to take that array of messages.

Cancellation Support

Description: Ability for Wolf to identify and execute cancellation flows mid activeAbility

Epic: Hot-loading Abilities Data

Description: On any given turn, the developer should be able to hot-swap the abilities.ts data and Wolf should be able to run uninterrupted with a seamless transition to the user.

Construction of the new abilities.ts (dynamic or static) is independent of this feature, but this epic should define best patterns for passing in the new abilities.ts data.

Tasks:

Define interface ActionResult

Description:

interface ActionResult {
  actionResult: PendingWolfState
  func: () => Promise // returns ability.onComplete promise

Allow Multiple Abilities

Is your feature request related to a problem? Please describe.
NA

Describe the solution you'd like
A user can say "added an alarm called wakeup at 6am. And removed alarm called workout". Wolf should ideally do two abilities in the single turn.

Describe alternatives you've considered
NA

Additional context
NA

Update Outtake signature

Description: Remove context.reply() bind as second argument in outtake. preview1.2 changes this to context.sendActivity().

Outtake now returns a messageArray so the user can implement replies from the bot. This allows wolf to avoid using async await

`onComplete` can return, an activity or a list of strings or Activities

Description:
Currently onComplete only allows the user to return (or resolve) either a string or null. We should allow the dev to return objects.

Epic: Slot Data API

Description: Define and implement a set of APIs that allow developers to set/change/update slot data. Wolf should handle this state updates in S1 before proceeding to S2.

Tasks:

• #75 Implement post /api/slotdata [EXAMPLE BOT]
• #77 Save lookup into storage [EXAMPLE BOT]
• #79 Implement slotData processing in S1
• #84 Run all onCompletes
• #80 Cleanup apiStorage
• #76 Implement post /api/new-conversation

Epic slated for v2 (breaking changes with middleware interface)

addMessageToQueue should check paramaters

Description: helpers/message should check if argument string is undefined or null.

Define acknowledge structure in ability.ts

Description: Move ability acknowledge to ability.ts.
Separate <ability>/index.ts can be removed. Users should handle storing value to convoState on their own inside onComplete function.

Test Feature

Tasks:

Dependent ability conversations

Description: Conversation flow should have multiple abilities and slots to represent a conversation with many pieces of information required from the user. The abilities should be able to have dependencies on others abilities to represent a structured conversation flows.

Example:
Abilities:

1. Onboard user (name)
2. Departure Information (date, airport)
3. Return Information (date, airport)

Abilities 2 and 3 have a dependency on ability 1

User: I would like to book a flight from ORD to EWR.
Bot: I can certainly help you with that, but first.. what is your name?
User: Foobar
Bot: Thank you Foobar!
Bot: What date would you like to fly from ORD to EWR
User: 8/1/2018
Bot: When would you like to return?
User: 8/2/2018
Bot: Okay your round trip flight from ORD to EWR is booked! You will depart on 8/1/2018 and return on 8/2/2018.

Change OuttakeResult

Description: Refactor outtake to return

{ messageArr: string[], richMessageArr: MessageQueueItem[] }

Implement retry multiple slot matches

Description: Currently..MatchNotValidData currently holds information on one slot that has been matched but fails validation.

Desired behavior: MatchNotValidData can hold multiple (array) slots that have been matched but fails validation which will allow Wolf to recognize that a match has been picked up.

Discussion: What order should we retry and attempt to fill these matches? One by one? all at once?

Save lookup onto storage

Description: Create new MemoryStorage() to save slot data api data. Lookup data should be made available for S1. (read-only)

Feature: #74

Refactor Wolf state into new namespace

Description: Move Wolf state into a new bot namespace to separate from conversation and user state.

https://github.com/Microsoft/botbuilder-js/blob/master/doc/botbuilder/classes/botbuilder.conversationstate.md#constructors-1

Rename ability acknowledge

Description: ability.acknowledge: () => string -> ability.onComplete: () => Promise

Must resolve to string or null

Implement post /api/slotdata/

Description: Implement server.post('/api/slotdata', (req: any, res: any)

• Accept payload and test values.
• Values that pass validators should be stored in a global lookup table for S1 to retrieve (one turn life).
• Regardless of pass/fail return descriptive response back to caller.

Feature: Epic #74

Implement addMessageToQueue

Description: Implement function for users to add string to pendingWolfState.messageQueue

Refactor Nlp Interface

Description: Refactor Nlp interface and usage to promote self documenting naming conventions.
I.e. entity.entity is ambiguous.

Correction Support

Description: Ability for users to correct prefilled slots

• Most recent (Oh, I meant JFK)
• Past filled slots (I want to change my departure airport)

passing `convoState` in `nextAbility`

Is your feature request related to a problem? Please describe.
user should have access to convoState when deciding which ability to start next.
However in order to have an accurate convoState, nextAbility has to be executed after the slot's onComplete is executed.

onComplete can be async, and the result of onComplete is done in the wolfMiddleware.

Describe the solution you'd like
We should need to make another stage (say, kickoff) to be run after execute and before outtake. this proposed stage, is to find the nextAbility and kickoff the ability with the next query (or finish the ability if no slot is present)

Since the nextAbility could finish (when it has no slot), and requires another execute... and so on and so forth. this means the execute and kickoff is in a loop until there is no more nextAbility to finish.

add a `forceComplete` or `exitAbility` function to the Slot `query`

Description:
Example: removeAlarm's alarmName slot shouldn't ask a user for an alarm name if the alarms array is empty

`shouldRetry` property for slots

Is your feature request related to a problem? Please describe.
Dev should be able to define a function that based on turn count or existing wolf state, whether wolf needs to run the retry function if a response is invalid.

This is helpful if Dev wants the only retry 3 times, and go to another slot.

Describe the solution you'd like

Describe alternatives you've considered
We could the logic in the retryQuery function but it feels dirty.

Additional context

Setup testing framework

Description: Setup testing framework and scripts

Outtake messages are sent out of order

Describe the bug
During turns when the bot is outputting multiple messages, the messages our being outputted

To Reproduce
Steps to reproduce the behavior:

• Start the alarm bot
• "add alarm called Hao"
• "kevin"

(the messages are sometime out of order)

Desktop (please complete the following information):

• OS: Windows

Port to Redux

Description: Port Wolf to use Redux.

Create Travis CI Pipeline

Description: Setup Travis that will fail based on tests.

Rename slot acknowledge

Description: slot.acknowledge: () => string -> slot.onFill: () => string

Epic: Test and CI

Description: Set up framework for testing with basic tests and CI

Tasks:

• #13
• #14
• #15

Implement slot.query

Description: slot.query:string -> slot.query: (pass the convostate getter function, etc..) => string

slot.query should be a function with access to convo, allowing dynamic prompting queries.
Return: string

Implement post /api/new-conversation

Description: Implement server.post('/api/new-conversation', (req, res) => {})

• Should call botbuilder directline and start new conversation

Feature: #74

remove reduxDevTool dependency on wolf

Is your feature request related to a problem? Please describe.
the user shouldn't have to build and install the redux dev tool everytime they install botbuilder-wolf

Describe the solution you'd like
Delegate the starting of devServer and the redux devtool redux middleware to the user.
provide the user a simple function to call (like wolfStoreBuilder), and allows the user to decorate the function by passing in the devtool middleware, and then pass the decorated function into wolfMiddleware

Describe alternatives you've considered
A clear and concise description of any alternative solutions or features you've considered.

Additional context
Add any other context or screenshots about the feature request here.

NLP results that has no intent but detected entities breaks the bot

Describe the bug
Example: alarmBot
if a user just says "called kevin", NLP will result {'intent': null, entities: {name: 'alarmName', value: 'kevin'}
this causes fillSlot to identify an "invalid" slot within the defaultActivity, which causes the validateSlot to try to run the retry off of the slot, but there is no such slot in the given ability.

To Reproduce
Steps to reproduce the behavior:

1. User says "called Kevin"
2. Bot breaks

Expected behavior
Bot just runs the default ability

Update repo for v4.0.0-Preview1.2

Description: Updates code to work with newest preview release: v4.0.0-Preview1.2

Tasks:

• examples/
• src/

Update Documentation

Description: Update repository documentation

Tasks:

• Wiki: bootstrap
• Wiki: roadmap
• Wiki: contributions + conduct
• Wiki: engineering guidelines
• Main README.md

Redux Observable

Description: Also allows for actions to kick off other actions and allow asynchronous processes.

Create Wolf Middleware

Description: Refactor src/ code into middleware and store OutakeResult in convoState for user to access.

Implement api slotData processing in S1

Description: If available, S1 should process data from slotData api.

• identify correct ability and slot
• fill slot data
• update slot status
• run onFill

Feature: #74

Slot Entity Types

Description: Allow slots to list entity types or "tags", which will allow Wolf to more intelligently match entities with potential slot matches and improve fill accuracy.

Current implementation: slot names are tightly coupled with entity types.

Refactor state and store

Description: Refactor all state naming to be consistent and accurately reflect state objects throughout the stages.

convertsationState -> convertsationStore
state -> conversationState

• look into where to initialize pendingWolfState = wolfState. pendingWolfState should not have any references outside of the Wolf stages. Currently used in nlp logic in bot.ts..

Define abilities for flightBot

Description: Define multiple abilities (A, B, C) for flight bot.
Abilities B and C should have a dependency on A

Refactor slot structure

Description: Refactor slot interface in src/types/ and example bot alarmBot
Property names should be self-descriptive

Example: entity -> name

Confirmation Slots Order

Description: Currently, confirmation slots must be ordered after the "calling slot" to ensure the confirmation slot is not picked up by the evaluate stage as the next pending slot to be completed.

Desired results: confirmation slots should only be prompted for then called by the "calling slot".

Add ImmutableJS?

Is your feature request related to a problem? Please describe.
NA

Describe the solution you'd like
We are mutating the pendingWolfState sometimes. To ensure we are not mutating states, we should use ImmutableJS for our pendingWolfState

Describe alternatives you've considered
Alternative is keep what we have, but try really hard to make sure we are not mutating the state.

Additional context
NA

Basic Tests

Description: Create basic unit tests for current code base

cleanup apiStorage

Description: Clean/purge apiStorage = new MemoryStorage()

Feature: #74

Epic: Support Asynchronous Operations

Description:
We will need to allow users to run async functions in the user-defined submit function, and possibility other stages.

Tasks:

• #29
• #31
• #35
• #36
• #37
• #38
• #39
• #40
• #41

Allow Ability to have other properties

Is your feature request related to a problem? Please describe.
Currently there is not a way to "inject" other dependencies onto the onComplete function. Any developer could want to use the context of the onComplete Function.

Describe the solution you'd like
Add a open string key with any on the Wolf Ability

Describe alternatives you've considered
In the functional scope: this works until you have to write unit tests.

Additional context
This is for better unit tests for the developers who wants to test the onComplete function