Skip to main content

Tasks

What is a Task?​

A Task is a defined piece of work assigned to agents, characterized by:

  • Clear Instructions: Details exactly what needs to be accomplished.
  • Defined Outcome: Specifies the expected result upon completion.
  • Assigned Responsibility: Allocated to a specific agent equipped to handle the task.

Tasks are essential in organizing efforts and driving projects towards successful outcomes.

Creating a Task​

To create a task, you start by initializing an instance of the Task class with the necessary properties. Here's how you can do it:

import { Task } from 'kaibanjs';

const searchTask = new Task({
    description: 'Search for detailed information about the sports query: {sportsQuery}.',
    expectedOutput: `Detailed information about the sports event, 
    including key players, key moments, 
    final score, and other useful information.`,
    agent: searchAgent  // Ensure searchAgent is defined and imported if needed
});

This example demonstrates how to define a task with a clear description, expected outcomes, and an associated agent responsible for its execution.

Task Attributes​

title (optional)​

The title of the task, which can be used as a concise summary or label.

  • Type: String
  • Example: Update Client Data
  • Default: '' (empty string)

description​

Describes what the task entails and the work to be performed.

  • Type: String
  • Example: Search for detailed information about the sports query.

expectedOutput​

Specifies the anticipated result or product from completing the task.

  • Type: String
  • Example: Detailed report including key players, key moments, and final score.

isDeliverable​

Specifies if the task outcome should be considered a final deliverable. Typically, KaibanJS treats the result of the last task as the deliverable, but this can be set to true for tasks whose results are critical at other stages.

  • Type: Boolean
  • Example: true
  • Default: false

agent​

The agent assigned to execute the task.

  • Type: Agent
  • Example: Refer to a specific Agent object instance, such as searchAgent.

status​

Indicates the current operational state of the task. This property is read-only and provides insights into the task's lifecycle.

  • Type: Enum (Read-only)
  • Example: [TODO, DOING, BLOCKED, REVISE, DONE, AWAITING_VALIDATION, VALIDATED]
  • Enum Defined At: Tasks Status Definitions

externalValidationRequired​

Indicates whether the task requires external validation before being considered complete.

  • Type: Boolean
  • Default: false

feedbackHistory​

An array that stores the history of feedback provided for the task.

  • Type: Array (Read-only)
  • Default: []

Each Feedback object in the feedbackHistory array has the following structure:

  • content: String - The feedback message
  • status: Enum - The status of the feedback (PENDING or PROCESSED)
  • timestamp: Number - Unix timestamp of when the feedback was provided

id​

A unique identifier for the task, autogenerated by the system. This property is read-only.

  • Type: String (Read-only)
  • Example: "579db4dd-deea-4e09-904d-a436a38e65cf"

Human-in-the-Loop (HITL) Features​

KaibanJS supports Human-in-the-Loop functionality for tasks, allowing for manual intervention and validation when necessary. This feature enhances the accuracy and reliability of task outcomes by incorporating human oversight into the workflow.

Key HITL features for tasks include:

  • External validation requirements
  • Feedback provision and history
  • Task revision based on human input

These features enable more complex workflows where human expertise or judgment is necessary, ensuring higher quality results and more reliable task completion.

For a detailed explanation of HITL features and how to implement them in your KaibanJS projects, please refer to our Human-in-the-Loop (HITL) documentation.

Conclusion​

Tasks drive the actions of agents in KaibanJS. By clearly defining tasks and their expected outcomes, you help AI agents work efficiently, whether alone or in teams. Understanding how tasks are carried out ensures that agents are well-prepared and that tasks are completed correctly.

We Love Feedback!

Is there something unclear or quirky in the docs? Maybe you have a suggestion or spotted an issue? Help us refine and enhance our documentation by submitting an issue on GitHub. We’re all ears!