Skip to main content
Parallelization
- Launch multiple subagents concurrently for independent tasks
- Use general-purpose agents to isolate context and reduce token usage
- Each subagent processes work independently and returns synthesized results
Model Selection
- Use appropriate models based on task complexity
- DeepAgent inherits all model selection capabilities from base Agent
- Can override model for specific executions
Monitoring
- Track todo status with
get_todos()
- Monitor completion progress in debug mode
- Access files at any time with
get_files()
Quality Assurance
Todo Completion Enforcement
- Automatic verification that all todos are completed
- Up to 10 continuation iterations if todos remain incomplete
- Debug warnings if max iterations reached with incomplete todos
File Validation
- All file operations return confirmation or error messages
- Read before edit enforcement in prompts
- Exact string matching for edit operations
Subagent Quality
- Subagents can be specialized for review and validation
- Results can be verified by main agent before proceeding
- Isolated context prevents subagent errors from affecting main agent
