ADK-TSLoop Agents
Repeat agent execution until conditions are met
Loop agents repeatedly execute a single sub-agent until a stopping condition is met. This enables iterative improvement, retry logic, and progressive refinement workflows.
How Loop Agents Work
- Iterative Execution: Run the same agent multiple times
- Condition Checking: Evaluate stopping conditions after each iteration
- Context Accumulation: Build upon previous results in each iteration
- Controlled Termination: Stop when conditions are met or limits are reached
When to Use Loop Agents
Loop agents are perfect for:
- Iterative improvement - Refining content until quality criteria are met
- Retry logic - Attempting tasks until successful completion
- Progressive research - Deepening understanding through multiple iterations
- Quality assurance - Repeated validation until standards are achieved
Convergence Required
Use loop agents when you expect the process to converge toward a solution or when you have clear stopping criteria.
Configuration
Basic Setup
- agent: The agent to execute repeatedly
- name: Unique identifier for the loop workflow
- description: Summary for other agents to understand the workflow
Loop Control
- max_iterations: Maximum number of loop cycles (required)
- stop_condition: Function or condition to determine when to stop
- iteration_limit: Alternative way to set maximum iterations
Advanced Options
- accumulate_context: Whether to build conversation history across iterations
- timeout: Maximum time for the entire loop execution
- on_iteration: Callback function to execute after each iteration
Common Patterns
Content Refinement
Repeatedly improve written content until it meets quality standards.
Research Deepening
Iteratively explore topics to build comprehensive understanding.
Problem Solving
Keep attempting different approaches until a solution is found.
Validation Loops
Repeatedly check and fix issues until validation passes.
Stopping Conditions
Built-in Conditions
- Maximum iterations: Stop after a specified number of cycles
- Timeout: Stop after a maximum time period
- Success detection: Stop when the agent indicates completion
Custom Conditions
- Quality thresholds: Stop when output quality meets criteria
- State validation: Stop when session state reaches target values
- External triggers: Stop based on external system conditions
Best Practices
Loop Safety
- Always set reasonable maximum iteration limits to prevent infinite loops
- Design clear stopping conditions to ensure termination
- Monitor resource usage during extended loop execution
- Consider timeout mechanisms for time-sensitive operations
Iteration Design
- Ensure each iteration makes meaningful progress
- Design the agent to build upon previous results
- Avoid repetitive behavior that doesn't add value
- Plan for diminishing returns in later iterations
Stopping Strategy
- Define clear, measurable stopping criteria
- Use multiple stopping conditions for robustness
- Test stopping conditions thoroughly
- Consider edge cases and failure modes
Performance Considerations
- Balance iteration thoroughness with execution time
- Monitor resource consumption during loops
- Consider breaking long loops into smaller chunks
- Plan for graceful degradation on timeout
Monitoring and Control
Progress Tracking
- Log iteration progress and results
- Monitor convergence toward stopping conditions
- Track resource usage across iterations
- Measure improvement between iterations
Error Handling
- Handle agent failures within iterations
- Implement retry logic for transient errors
- Consider partial results when loops terminate early
- Plan for recovery from iteration failures