Prompt Library ⚙️ Technical AI Chain-of-Thought Debugging
GPT-4o ⚙️ Technical Advanced

AI Chain-of-Thought Debugging

Diagnose and fix flawed AI reasoning by inspecting chain-of-thought outputs, locating where logic breaks down, and redesigning the prompt.

👁 0 views ⎘ 0 copies ♥ 0 likes

The Prompt

# AI Chain-of-Thought Debugging

You are a prompt debugging specialist. Debug a reasoning failure in [MODEL, e.g. GPT-6 Astra, Claude Fable 5.1] on the following task.

## The Failing Task

Task description: [DESCRIBE THE TASK]
Observed failure: [WHAT WENT WRONG — e.g. the model reached the wrong conclusion, made an arithmetic error, contradicted itself]

## Chain-of-Thought Inspection

Read the model's reasoning output step by step. Identify:
1. The exact step where the logic first goes wrong
2. The root cause: wrong assumption, missing information, shortcut, or hallucination
3. Whether the error is systematic (happens on similar inputs) or one-off

## Diagnosis

Classify the failure type:
- Faulty premise: [DESCRIBE]
- Reasoning gap: [DESCRIBE]
- Conflicting instructions in the prompt: [DESCRIBE]
- Insufficient context: [DESCRIBE]

## Redesigned Prompt

Write an improved prompt that prevents this class of error. Explain every change you made and why it addresses the root cause.

## Regression Test

Write [NUMBER] test cases that verify the fix does not introduce new failures.

📝 Fill in the blanks

Replace these placeholders with your own content:

[MODEL, e.g. GPT-6 Astra, Claude Fable 5.1]
[DESCRIBE THE TASK]
[WHAT WENT WRONG — e.g. the model reached the wrong conclusion, made an arithmetic error, contradicted itself]
[DESCRIBE]
[NUMBER]

How to use this prompt

1
Copy the prompt

Click "Copy Prompt" above to copy the full prompt text to your clipboard.

2
Replace the placeholders

Swap out anything in [BRACKETS] with your specific details.

3
Paste into GPT-4o

Open your preferred AI assistant and paste the prompt to get started.