gh-giuseppe-trisciuoglio-developer-kit/skills/ai/prompt-engineering/references/cot-patterns.md

Chain-of-Thought Reasoning Patterns

This reference provides comprehensive frameworks for implementing effective chain-of-thought (CoT) reasoning that improves model performance on complex, multi-step problems.

Core Principles

Step-by-Step Reasoning Elicitation

Problem Decomposition Strategy

• Break complex problems into manageable sub-problems
• Identify dependencies and relationships between components
• Establish logical flow and sequence of reasoning steps
• Define clear decision points and validation criteria

Verification and Validation Integration

• Include self-checking mechanisms at critical junctures
• Implement consistency checks across reasoning steps
• Add confidence scoring for uncertain conclusions
• Provide fallback strategies for ambiguous situations

Zero-Shot Chain-of-Thought Patterns

Basic CoT Initiation

Let's think step by step to solve this problem:

1. First, I need to understand what the question is asking for
2. Then, I'll identify the key information and constraints
3. Next, I'll consider different approaches to solve it
4. I'll work through the solution methodically
5. Finally, I'll verify my answer makes sense

Problem: {problem_statement}

Step 1: Understanding the question
{analysis}

Step 2: Key information and constraints
{information_analysis}

Step 3: Solution approach
{approach_analysis}

Step 4: Working through the solution
{detailed_solution}

Step 5: Verification
{verification}

Final Answer: {conclusion}

Enhanced CoT with Confidence

Let me think through this systematically, breaking down the problem and checking my reasoning at each step.

**Problem**: {problem_description}

**Step 1: Problem Analysis**
- What am I being asked to solve?
- What information is provided?
- What are the constraints?
- My confidence in understanding: {score}/10

**Step 2: Strategy Selection**
- Possible approaches:
  1. {approach_1}
  2. {approach_2}
  3. {approach_3}
- Selected approach: {chosen_approach}
- Rationale: {reasoning_for_choice}

**Step 3: Execution**
- {detailed_step_by_step_solution}

**Step 4: Verification**
- Does the answer make sense?
- Have I addressed all parts of the question?
- Confidence in final answer: {score}/10

**Final Answer**: {solution_with_confidence_score}

Few-Shot Chain-of-Thought Patterns

Mathematical Reasoning Template

Solve the following math problem step by step.

Example 1:
Problem: A store sells apples for $2 each and oranges for $3 each. If John buys 4 apples and 2 oranges, and spends exactly $14, how much does each fruit cost?

Step 1: Set up the equation
Let a = cost of apples, o = cost of oranges
4a + 2o = 14

Step 2: Simplify the equation
Divide both sides by 2: 2a + o = 7

Step 3: Test integer solutions
If a = 2, then 2(2) + o = 7 → o = 3
If a = 3, then 2(3) + o = 7 → o = 1

Step 4: Verify the original problem
For a=2, o=3: 4(2) + 2(3) = 8 + 6 = 14 ✓
For a=3, o=1: 4(3) + 2(1) = 12 + 2 = 14 ✓

Both solutions work, so apples could cost $2 and oranges $3, OR apples $3 and oranges $1.

Example 2:
Problem: Sarah is twice as old as Tom was 3 years ago. In 5 years, Sarah will be 1.5 times Tom's current age. How old are they now?

Step 1: Define variables
Let S = Sarah's current age, T = Tom's current age

Step 2: Set up equations
"Sarah is twice as old as Tom was 3 years ago": S = 2(T - 3)
"In 5 years, Sarah will be 1.5 times Tom's current age": S + 5 = 1.5T

Step 3: Solve the system
From first equation: S = 2T - 6
Substitute into second: (2T - 6) + 5 = 1.5T
2T - 1 = 1.5T
0.5T = 1
T = 2

Step 4: Find Sarah's age
S = 2(2) - 6 = 4 - 6 = -2

Step 5: Check for validity
Negative age is impossible, so I must have made an error.
Let me re-examine the problem...

Actually, let me try again:
S = 2(T - 3) = 2T - 6
S + 5 = 1.5T
Substitute: (2T - 6) + 5 = 1.5T
2T - 1 = 1.5T
0.5T = 1
T = 2 years old
S = 2(2 - 3) = 2(-1) = -2

This gives a negative age. The problem might have no solution with integer ages.

Now solve:
Problem: {math_problem}

Step 1: Define variables
{variable_definitions}

Step 2: Set up equations
{equation_setup}

Step 3: Solve the system
{solution_process}

Step 4: Verify the solution
{verification}

Final Answer: {answer}

Logical Reasoning Template

Analyze the logical argument and determine if it's valid.

Example 1:
Premise 1: All birds can fly
Premise 2: Penguins are birds
Conclusion: Therefore, penguins can fly

Step 1: Analyze the structure
This is a syllogism with form:
All A are B
C is A
Therefore, C is B

Step 2: Evaluate premise validity
Premise 1: "All birds can fly" - This is false (penguins, ostriches cannot fly)
Premise 2: "Penguins are birds" - This is true

Step 3: Check logical validity
The logical structure is valid, but since Premise 1 is false, the conclusion may not be true

Step 4: Real-world verification
In reality, penguins cannot fly despite being birds

Conclusion: The argument is logically valid but soundness fails due to false premise

Example 2:
Premise 1: If it rains, then the ground gets wet
Premise 2: It is raining
Conclusion: Therefore, the ground gets wet

Step 1: Analyze the structure
This is modus ponens:
If P, then Q
P
Therefore, Q

Step 2: Evaluate premise validity
Premise 1: "If it rains, then the ground gets wet" - Generally true
Premise 2: "It is raining" - Given as true

Step 3: Check logical validity
Modus ponens is a valid argument form

Step 4: Verify the conclusion
Given the premises, the conclusion follows logically

Conclusion: The argument is both logically valid and sound

Now analyze:
Argument: {logical_argument}

Step 1: Analyze the argument structure
{structure_analysis}

Step 2: Evaluate premise validity
{premise_evaluation}

Step 3: Check logical validity
{validity_check}

Step 4: Verify the conclusion
{conclusion_verification}

Final Assessment: {argument_validity_assessment}

Self-Consistency Techniques

Multiple Reasoning Paths

I'll solve this problem using three different approaches and see which result is most reliable.

**Problem**: {complex_problem}

**Approach 1: Direct Calculation**
{first_approach_reasoning}
Result 1: {result_1}

**Approach 2: Logical Deduction**
{second_approach_reasoning}
Result 2: {result_2}

**Approach 3: Pattern Recognition**
{third_approach_reasoning}
Result 3: {result_3}

**Consistency Analysis:**
- Approach 1 and 2 agree: {yes/no}
- Approach 1 and 3 agree: {yes/no}
- Approach 2 and 3 agree: {yes/no}

**Final Decision:**
{majority_result} appears in {count} out of 3 approaches.
Confidence: {high/medium/low}

Most Likely Answer: {final_answer_with_confidence}

Verification Loop Pattern

Let me solve this step by step and verify each step.

**Problem**: {problem_description}

**Step 1: Initial Analysis**
{initial_analysis}

Verification: Does this make sense? {verification_1}

**Step 2: Solution Development**
{solution_development}

Verification: Does this logically follow from step 1? {verification_2}

**Step 3: Result Calculation**
{result_calculation}

Verification: Does this answer the original question? {verification_3}

**Step 4: Cross-Check**
Let me try a different approach to confirm:
{alternative_approach}

Results comparison: {comparison_analysis}

**Final Answer:**
{conclusion_with_verification_status}

Specialized CoT Patterns

Code Debugging CoT

Debug the following code by analyzing it step by step.

**Code:**
{code_snippet}

**Step 1: Understand the Code's Purpose**
{purpose_analysis}

**Step 2: Identify Expected Behavior**
{expected_behavior}

**Step 3: Trace the Execution**
{execution_trace}

**Step 4: Find the Error**
{error_identification}

**Step 5: Propose Fix**
{fix_proposal}

**Step 6: Verify the Fix**
{fix_verification}

**Fixed Code:**
{corrected_code}

Data Analysis CoT

Analyze this data systematically to draw meaningful conclusions.

**Data:**
{dataset}

**Step 1: Understand the Data Structure**
{data_structure_analysis}

**Step 2: Identify Patterns and Trends**
{pattern_identification}

**Step 3: Calculate Key Metrics**
{metrics_calculation}

**Step 4: Compare with Benchmarks**
{benchmark_comparison}

**Step 5: Formulate Insights**
{insight_generation}

**Step 6: Validate Conclusions**
{conclusion_validation}

**Key Findings:**
{summary_of_insights}

Creative Problem Solving CoT

Generate creative solutions to this challenging problem.

**Problem:**
{creative_problem}

**Step 1: Reframe the Problem**
{problem_reframing}

**Step 2: Brainstorm Multiple Angles**
- Technical approach: {technical_ideas}
- Business approach: {business_ideas}
- User experience approach: {ux_ideas}
- Unconventional approach: {unconventional_ideas}

**Step 3: Evaluate Each Approach**
{approach_evaluation}

**Step 4: Synthesize Best Elements**
{synthesis_process}

**Step 5: Develop Final Solution**
{solution_development}

**Step 6: Test for Feasibility**
{feasibility_testing}

**Recommended Solution:**
{final_creative_solution}

Implementation Guidelines

When to Use Chain-of-Thought

• Multi-step problems: Tasks requiring sequential reasoning
• Complex calculations: Mathematical or logical derivations
• Problem decomposition: Tasks that benefit from breaking down
• Verification needs: When accuracy is critical
• Educational contexts: When showing reasoning is valuable

CoT Effectiveness Factors

• Problem complexity: Higher benefit for complex problems
• Task type: Mathematical, logical, and analytical tasks benefit most
• Model capability: Newer models handle CoT more effectively
• Context window: Ensure sufficient space for reasoning steps
• Output requirements: Detailed explanations benefit from CoT

Common Pitfalls to Avoid

• Over-explaining simple steps: Keep proportional detail
• Circular reasoning: Ensure logical progression
• Missing verification: Always include validation steps
• Inconsistent confidence: Use realistic confidence scoring
• Premature conclusions: Don't jump to answers without full reasoning

Integration with Other Techniques

CoT + Few-Shot Learning

• Include reasoning traces in examples
• Show step-by-step problem-solving demonstrations
• Teach verification and self-checking patterns

CoT + Template Systems

• Embed CoT patterns within structured templates
• Use conditional CoT based on problem complexity
• Implement adaptive reasoning depth

CoT + Prompt Optimization

• Test different CoT formulations
• Optimize reasoning step granularity
• Balance detail with efficiency

This framework provides comprehensive patterns for implementing effective chain-of-thought reasoning across diverse problem types and applications.