reasoning-strategy-selection

What problem does it solve?

Helps engineers choose and configure the right agent reasoning strategy, iteration budget, and tool gates so agents meet task requirements for cost, capability, and output quality. It reduces guesswork when deciding between single-pass, multi-step, branching, or adaptive approaches and ensures outputs meet format and verification needs.

Core Features & Use Cases

• Strategy Guidance: Maps common task types to recommended strategies (reactive, plan-execute-reflect, adaptive, tree-of-thought, reflexion).
• Configuration Patterns: Prescribes iteration limits, strategy switching behavior, fallback strategies, and cost tracking to avoid runaway usage.
• Tool & Output Gates: Enforces required tool calls and activates an output quality pipeline to validate and repair formats like JSON, markdown, CSV, and code.
• Observability & Pitfalls: Shows how to subscribe to strategy switch events and warns about common mistakes like incorrect strategy names or missing cost tracking.

Quick Start

Configure an agent with .withReasoning defaultStrategy set to "adaptive", maxIterations 12, enable verification, and enable cost tracking.