ADK-TSSequential Agents
Execute agents one after another in a specific order
Sequential agents execute a list of sub-agents one after another in a predetermined order. Each agent runs after the previous one completes, creating a predictable pipeline workflow.
How Sequential Agents Work
- Ordered Execution: Agents run in the exact order you specify
- Sequential Processing: Each agent waits for the previous one to complete
- Context Sharing: Results flow through the conversation context
- State Passing: Use session state to pass structured data between agents
When to Use Sequential Agents
Sequential agents are perfect for:
- Multi-step pipelines - Research → Analysis → Summary
- Data processing workflows - Extract → Transform → Load
- Quality assurance - Generate → Review → Approve
- Progressive refinement - Draft → Edit → Finalize
Order Matters
Use sequential agents when the order of execution is important and each step depends on the previous ones.
Configuration
Basic Setup
- agents: List of agents to execute in order
- name: Unique identifier for the workflow
- description: Summary for other agents to understand the workflow
Advanced Options
- max_iterations: Limit the number of execution cycles
- global_instruction: Instructions that apply to all sub-agents
- disallow_transfer: Control agent delegation behavior
Common Patterns
Research Pipeline
Execute research, then analysis, then summarization in sequence.
Content Creation Workflow
Generate content, review for accuracy, then format for publication.
Data Processing Pipeline
Extract data, transform it, then load it into the target system.
Quality Assurance Workflow
Create initial content, perform quality checks, then make final adjustments.
Best Practices
Design Considerations
- Keep the number of sequential steps reasonable (3-7 agents typically)
- Ensure each agent has a clear, specific purpose
- Use descriptive names and instructions for better coordination
- Consider error handling between steps
Agent Design
- Make each agent's purpose clear and specific
- Design agents to work well with the outputs of previous agents
- Use consistent naming conventions across the workflow
Data Flow
- Plan how data will flow between agents
- Use session state for structured data passing
- Consider what context each agent needs
Error Handling
- Design agents to handle unexpected inputs gracefully
- Consider retry logic for critical steps
- Plan for partial failure scenarios