home / skills / zpankz / mcp-skillset / thought-based-reasoning

thought-based-reasoning skill

safe

This skill leverages chain-of-thought techniques to guide step-by-step reasoning, improving complex problem-solving accuracy and explainability for multi-step

npx playbooks add skill zpankz/mcp-skillset --skill thought-based-reasoning

Review the files below or copy the command above to add this skill to your agents.

Files (3)

SKILL.md

2.4 KB

---
name: thought-based-reasoning
description: Use when tackling complex reasoning tasks requiring step-by-step logic, multi-step arithmetic, commonsense reasoning, symbolic manipulation, or problems where simple prompting fails - provides comprehensive guide to Chain-of-Thought and related prompting techniques
---

# Thought-Based Reasoning Techniques for LLMs

## Overview

Chain-of-Thought (CoT) prompting and its variants encourage LLMs to generate intermediate reasoning steps before arriving at a final answer, significantly improving performance on complex reasoning tasks. These techniques transform how models approach problems by making implicit reasoning explicit.

## Quick Reference

| Technique | When to Use | Complexity | Accuracy Gain |
|-----------|-------------|------------|---------------|
| Zero-shot CoT | Quick reasoning, no examples available | Low | +20-60% |
| Few-shot CoT | Have good examples, consistent format needed | Medium | +30-70% |
| Self-Consistency | High-stakes decisions, need confidence | Medium | +10-20% over CoT |
| Tree of Thoughts | Complex problems requiring exploration | High | +50-70% on hard tasks |
| Least-to-Most | Multi-step problems with subproblems | Medium | +30-80% |
| ReAct | Tasks requiring external information | Medium | +15-35% |
| PAL | Mathematical/computational problems | Medium | +10-15% |
| Reflexion | Iterative improvement, learning from errors | High | +10-20% |

## When to Use Thought-Based Reasoning

**Use CoT techniques when:**
- Multi-step arithmetic or math word problems
- Commonsense reasoning requiring logical deduction
- Symbolic reasoning tasks
- Complex problems where simple prompting fails

**Start with:**
- **Zero-shot CoT** for quick prototyping ("Let's think step by step")
- **Few-shot CoT** when you have good examples
- **Self-Consistency** for high-stakes decisions

## Progressive Loading

**L2 Content** (loaded when core techniques needed):
- See: [references/core-techniques.md](./references/core-techniques.md)
  - Chain-of-Thought (CoT) Prompting
  - Zero-shot Chain-of-Thought
  - Self-Consistency Decoding
  - Tree of Thoughts (ToT)
  - Least-to-Most Prompting
  - ReAct (Reasoning + Acting)
  - PAL (Program-Aided Language Models)
  - Reflexion

**L3 Content** (loaded when decision guidance and best practices needed):
- See: [references/guidance.md](./references/guidance.md)
  - Decision Matrix: Which Technique to Use
  - Best Practices
  - Common Mistakes
  - References

Overview

This skill provides a practical guide to thought-based reasoning techniques for large language models, centered on Chain-of-Thought and related prompting strategies. I present when to use each method, performance tradeoffs, and concrete tactics to elicit step-by-step reasoning for complex tasks. The goal is to improve accuracy on multi-step arithmetic, commonsense deduction, symbolic manipulation, and problems that fail with simple prompts.

How this skill works

I walk you through core techniques—zero-shot and few-shot Chain-of-Thought, Self-Consistency, Tree of Thoughts, Least-to-Most, ReAct, PAL, and Reflexion—and explain how each encourages the model to generate intermediate steps before a final answer. The skill explains how to craft prompts, supply examples, sample alternative reasoning traces, and iterate on solutions. It also covers deciding which technique fits a task, how to combine methods, and how to decode or aggregate multiple reasoning paths for higher confidence.

When to use it

Multi-step arithmetic or math word problems where intermediate steps are required
Commonsense and logical deduction tasks that need transparent reasoning
Symbolic reasoning, puzzles, or problems with discrete steps
Problems that produce incorrect answers with direct prompting
High-stakes decisions where sampling multiple reasoning chains improves reliability

Best practices

Start simple: try zero-shot CoT with 'Let's think step by step' for quick gains
Use few-shot CoT when you have clear, representative examples in the desired format
Apply Self-Consistency by sampling multiple chains and taking the most common answer
Break complex tasks into subproblems with Least-to-Most to reduce search space
Use Tree of Thoughts for problems that benefit from explicit search and backtracking
Iterate: collect failure cases, refine prompts or examples, and consider Reflexion-style reruns

Example use cases

Solving multi-step algebra or word problems that require showing work
Designing diagnostic prompts for model evaluation and error analysis
Building assistants that perform stepwise planning or multi-turn reasoning
Generating program-like derivations for symbolic math or code synthesis
Creating high-confidence outputs for critical decisions by aggregating chains

FAQ

When should I prefer few-shot CoT over zero-shot CoT?

Use few-shot when you can provide high-quality, consistent examples that mirror your target problem; it usually yields larger accuracy gains but needs careful example design.

How many sampled chains are enough for Self-Consistency?

There’s no fixed number; 5–20 diverse samples often help, with diminishing returns beyond that. Monitor answer convergence and cost.