reasoning

Reasoning Techniques Pattern

Overview

The Reasoning Techniques Pattern equips agents with structured approaches to thinking through problems before acting. Rather than jumping directly from input to output, reasoning-enhanced agents explicitly decompose problems, explore possibilities, verify their work, and reflect on their reasoning process — dramatically improving accuracy on complex tasks.

Core Principle: Slow down to speed up accuracy — structured thinking catches errors that fast pattern-matching misses.

When This Skill Applies

Activate this pattern when:

• Multi-step problems require logical deduction or causal reasoning
• Mathematical, coding, or analytical tasks require precise computation
• The agent must justify its conclusions (auditable decisions)
• Complex questions benefit from exploring multiple solution paths
• Self-consistency across multiple reasoning attempts is needed
• The agent must detect and correct its own errors before responding

Rule of thumb: If a human expert would "show their work" — the agent should too. Use reasoning techniques whenever accuracy matters more than speed.

Key Reasoning Techniques

1. Chain-of-Thought (CoT)

Break reasoning into explicit intermediate steps:

Problem → Step 1 → Step 2 → ... → Step N → Answer

2. ReAct (Reason + Act)

Interleave reasoning with tool use:

Thought → Action (tool call) → Observation → Thought → Action → ... → Answer

3. Tree of Thought (ToT)

Explore multiple reasoning branches and select the best:

Problem → Branch A → Branch B → Branch C
                    ↓ (evaluate each)
              Select Best Path → Answer

4. Self-Consistency

Generate multiple independent solutions, pick the majority answer:

Problem → Solution 1 → Answer A
        → Solution 2 → Answer A (consensus)
        → Solution 3 → Answer B

5. Self-Critique (Reflexion)

Agent critiques its own output and corrects it:

Problem → Initial Answer → Critique → Revised Answer → Critique → Final Answer