Class WorkflowGenerator

Namespace

JD.AI.Workflows

Assembly

JD.AI.Workflows.dll

Generates structured workflow definitions from natural language descriptions. Uses a prompt-based approach: the description is analyzed and decomposed into steps with tool mappings, parameters, and control flow.

public sealed class WorkflowGenerator

Inheritance

object

WorkflowGenerator

Inherited Members

object.Equals(object)

object.Equals(object, object)

object.GetHashCode()

object.GetType()

object.ReferenceEquals(object, object)

object.ToString()

Methods

Compose(string, IReadOnlyList<AgentWorkflowDefinition>)

Composes multiple workflows into a single composite workflow. Steps from each child workflow become nested steps in the composite.

public AgentWorkflowDefinition Compose(string name, IReadOnlyList<AgentWorkflowDefinition> workflows)

Parameters

name string

workflows IReadOnlyList<AgentWorkflowDefinition>

Returns

AgentWorkflowDefinition

DeriveToolStepName(string)

public static string DeriveToolStepName(string tool)

Parameters

tool string

Returns

string

DryRun(AgentWorkflowDefinition, IReadOnlySet<string>?)

Generates a dry-run preview of a workflow without executing. Shows what each step would do, required tools, and validation.

public WorkflowDryRunResult DryRun(AgentWorkflowDefinition workflow, IReadOnlySet<string>? availableTools = null)

Parameters

workflow AgentWorkflowDefinition

availableTools IReadOnlySet<string>

Returns

WorkflowDryRunResult

ExtractFromHistory(IReadOnlyList<ChatMessageContent>, int, string?)

Extracts a workflow definition from conversation history by analyzing the tool calls and user/assistant interactions from the last N turns.

public AgentWorkflowDefinition ExtractFromHistory(IReadOnlyList<ChatMessageContent> messages, int turnCount = 10, string? name = null)

Parameters

messages IReadOnlyList<ChatMessageContent>

turnCount int

name string

Returns

AgentWorkflowDefinition

FormatDryRun(WorkflowDryRunResult)

Formats a dry-run result for display.

public static string FormatDryRun(WorkflowDryRunResult result)

Parameters

result WorkflowDryRunResult

Returns

string

Generate(string, string?)

Generates a workflow definition from a natural language description. This is a deterministic fallback that parses common patterns; for AI-powered generation, the agent loop itself generates the workflow and calls SaveAsync on the catalog.

public AgentWorkflowDefinition Generate(string description, string? name = null)

Parameters

description string

name string

Returns

AgentWorkflowDefinition

GenerateAsync(string, Kernel, string?, IReadOnlySet<string>?, CancellationToken)

Generates a workflow from a natural language description using LLM. Falls back to heuristic generation if LLM fails.

public Task<WorkflowRefinementResult> GenerateAsync(string description, Kernel kernel, string? name = null, IReadOnlySet<string>? availableTools = null, CancellationToken ct = default)

Parameters

description string

kernel Kernel

name string

availableTools IReadOnlySet<string>

ct CancellationToken

Returns

Task<WorkflowRefinementResult>

RefineAsync(AgentWorkflowDefinition, string, Kernel, CancellationToken)

Refines an existing workflow using AI-powered analysis. Applies the user's natural language feedback to add, remove, reorder, or modify steps. Returns a new version of the workflow with the refinements applied.

public Task<WorkflowRefinementResult> RefineAsync(AgentWorkflowDefinition workflow, string feedback, Kernel kernel, CancellationToken ct = default)

Parameters

workflow AgentWorkflowDefinition

feedback string

kernel Kernel

ct CancellationToken

Returns

Task<WorkflowRefinementResult>