Structure task output

[anthonydevs17]

Add Structured Output Feature

This PR adds Structured Output support to KaibanJS, allowing users to define and enforce specific output structures for their agent tasks using Zod schemas.

Changes

• Add outputSchema property to Task class
• Implement schema validation in ReactChampionAgent
• Add OUTPUT_SCHEMA_VALIDATION_ERROR state
• Integrate with workflowLogs for error tracking
• Add schema validation feedback loop in agent processing

Implementation Details

• Use Zod for schema validation
• Add schema validation to agent's output processing
• Integrate with existing state management system
• Add validation error handling in agent loop
• Track validation events in workflowLogs

Testing

• Add tests for schema validation
• Add tests for error handling
• Add tests for agent recovery from validation errors

[darielnoel]

Solves #171

[darielnoel]

It is better to call this variable this way. -> INVALID_OUTPUT_SCHEMA_FEEDBACK.

[darielnoel]

Love this prompt 🤗

[darielnoel]

We cannot overwrite the expectedOutput here... both variables serves different purposes:

expectedOutput and outputSchema serve different purposes:

• expectedOutput: A human-readable description used in prompts to tell the AI agent what output format you want (used for instruction)
• outputSchema: A Zod schema that validates the actual output structure (used for validation)

Example:

const task = new Task({
  description: "Extract article metadata",
  expectedOutput: "Get the article's title and list of tags", // For the AI to understand
  outputSchema: z.object({                                    // For validation
    title: z.string(),
    tags: z.array(z.string())
  })
});

[darielnoel]

Let's use OUTPUT_SCHEMA_VALIDATION_ERROR as the name for this agent status.

[darielnoel]

Update error message to: "The output does not match the expected schema structure"

[darielnoel]

Here we should implement a different handler in the store. Cause they are two separated use cases.