reasoning-controls
Use when adjusting reasoning depth, budgets, or metrics visibility - provides guidance for selecting and applying reasoning controls safely.
When & Why to Use This Skill
The Reasoning Controls skill provides a robust framework for managing the cognitive intensity and operational costs of AI agents. It enables developers to precisely calibrate reasoning depth, establish strict latency and token budgets, and monitor execution metrics, ensuring a perfect balance between high-quality output and resource efficiency.
Use Cases
- Cost Management: Reducing reasoning depth for high-volume, simple tasks to minimize API expenditure and improve throughput.
- Complex Problem Solving: Scaling up reasoning budgets for intricate tasks like architectural design, deep debugging, or strategic planning.
- Latency Optimization: Setting strict time-based constraints for real-time applications to ensure the agent remains responsive to user interactions.
- Performance Auditing: Utilizing metrics reporting to track reasoning patterns and optimize the agent's decision-making logic over time.
| name | reasoning-controls |
|---|---|
| description | Use when adjusting reasoning depth, budgets, or metrics visibility - provides guidance for selecting and applying reasoning controls safely. |
Reasoning Controls
Overview
Control reasoning depth and cost trade-offs using consistent settings and metrics.
When to Use
- Adjusting reasoning depth or thinking mode
- Setting budget limits for cost or latency
- Reporting reasoning metrics
Avoid when:
- The task doesn’t require explicit reasoning controls
Quick Reference
| Task | Load reference |
|---|---|
| Adjust reasoning | skills/reasoning-controls/references/adjust.md |
| Budget controls | skills/reasoning-controls/references/budget.md |
| Metrics reporting | skills/reasoning-controls/references/metrics.md |
Workflow
- Determine the control goal (depth, budget, metrics).
- Load the matching reference.
- Apply the control with the appropriate parameters.
- Report settings and effects.
Output
- Updated reasoning settings
- Metrics or confirmation output
Common Mistakes
- Over-allocating budget for simple tasks
- Changing depth without explaining trade-offs