workflow-conditional

README

Conditional Workflow Example

This example demonstrates how to use conditional workflow agents to implement branching logic in workflows.

Overview

The conditional workflow agent executes different branches based on state conditions, similar to if/else logic in programming. This is useful for:

• Routing data to specialized processors based on type
• Implementing priority-based handling systems
• Creating decision trees for complex business logic
• Handling multiple conditions with custom priority ordering

Features Demonstrated

1. Basic Conditional Logic: Simple if/else branching based on state values
2. Priority-Based Evaluation: Conditions evaluated in priority order (highest first)
3. Multiple Matching: Allow multiple branches to execute for comprehensive processing
4. Default Branches: Fallback processing when no conditions match
5. Complex Conditions: Custom condition functions with access to full state

Usage

# Run with mock agents
go run main.go

# Run with actual LLM providers (requires API keys)
export ANTHROPIC_API_KEY=your-key
export OPENAI_API_KEY=your-key
go run main.go

Examples

1. Basic Conditional Processing

Routes different data types to specialized processors:

• Text data → Text processor (Claude)
• Image data → Image processor (GPT-4 Vision)
• Structured data → Data processor (GPT-4)
• Unknown data → Generic processor (fallback)

2. Priority-Based Issue Triage

Handles issues based on severity with priority ordering:

• Severity 9-10 → Critical handler (Priority 100)
• Severity 7-8 → High priority handler (Priority 75)
• Severity 4-6 → Medium priority handler (Priority 50)
• Severity 1-3 → Low priority handler (Priority 25)
• Severity 0 → No handler (demonstrates no match scenario)

3. Multiple Match Validation

Runs multiple validation checks simultaneously:

• Syntax validation
• Security scanning
• Performance analysis
• Compatibility checking

All applicable checks run based on the required_checks configuration.

Code Structure

Basic Conditional Workflow

// Create conditional workflow
workflow := workflow.NewConditionalAgent("processor").
    AddAgent("text", func(state *domain.State) bool {
        dataType, _ := state.Get("data_type")
        return dataType == "text"
    }, textProcessor).
    AddAgent("image", func(state *domain.State) bool {
        dataType, _ := state.Get("data_type")
        return dataType == "image"
    }, imageProcessor).
    SetDefaultAgent(genericProcessor)

// Run with state
result, err := workflow.Run(ctx, initialState)

Priority-Based Conditions

workflow := workflow.NewConditionalAgent("triage").
    AddBranchWithPriority("critical", criticalCondition, criticalStep, 100).
    AddBranchWithPriority("high", highCondition, highStep, 75).
    AddBranchWithPriority("medium", mediumCondition, mediumStep, 50)

Multiple Matches

workflow := workflow.NewConditionalAgent("validation").
    WithAllowMultipleMatches(true).
    WithEvaluateAllConditions(true).
    AddAgent("syntax", syntaxCondition, syntaxValidator).
    AddAgent("security", securityCondition, securityValidator)

Configuration Options

• AddAgent(name, condition, agent): Add a branch with agent
• AddBranch(name, condition, step): Add a branch with custom step
• AddBranchWithPriority(name, condition, step, priority): Add prioritized branch
• SetDefaultAgent(agent): Set fallback agent
• SetDefaultBranch(step): Set fallback step
• WithAllowMultipleMatches(bool): Allow multiple branches to execute
• WithEvaluateAllConditions(bool): Evaluate all conditions even after finding matches
• WithHook(hook): Add monitoring hooks

Condition Functions

Condition functions receive the current state and return a boolean:

func textCondition(state *domain.State) bool {
    if dataType, exists := state.Get("data_type"); exists {
        return dataType == "text"
    }
    return false
}

func severityCondition(minSeverity int) func(*domain.State) bool {
    return func(state *domain.State) bool {
        if severity, exists := state.Get("severity"); exists {
            return severity.(int) >= minSeverity
        }
        return false
    }
}

Execution Flow

1. Condition Evaluation: Conditions evaluated in priority order (highest first)
2. Branch Execution: Matching branches execute based on configuration
3. State Flow: Each branch receives and can modify the workflow state
4. Result Merging: Final state contains results from executed branches
5. Default Handling: Default branch executes if no conditions match

Error Handling

• Branch Errors: Individual branch failures can be handled or cause workflow failure
• Validation: All branches and conditions validated before execution
• Graceful Degradation: Default branches provide fallback behavior

Metadata

The workflow adds execution metadata including:

• executed_branches: List of branches that executed
• total_branches: Total number of defined branches
• has_default: Whether a default branch is configured

Next Steps

See the loop workflow example for iterative processing, or combine conditional and parallel workflows for complex decision trees.