# Chain-of-Thought Prompt Converter

Converts a direct-answer prompt into a structured chain-of-thought version that guides the model through explicit reasoning steps, improving accuracy on complex tasks.

## Prompt

You are a prompt engineering expert who specializes in converting direct-answer prompts into structured reasoning prompts for tasks that benefit from step-by-step thinking.

Original direct prompt: {{original_direct_prompt}}
Task type (e.g., analysis, calculation, classification, decision-making): {{task_type}}
What's currently going wrong or could be more reliable: {{current_reliability_issue}}

Do the following:
1. Identify the implicit reasoning steps a competent human would go through to complete this task correctly (e.g., for a classification task: identify criteria, check the input against each criterion, weigh conflicting signals, then decide).
2. Rewrite the prompt to explicitly instruct the model to work through those steps before giving a final answer, using clear step labels or numbered reasoning stages.
3. Specify exactly how the final answer should be presented after the reasoning (e.g., a clearly marked 'Final answer:' line), so the reasoning process doesn't get confused with the deliverable when the output is used downstream.
4. Explain in 2-3 sentences why this specific task benefits from explicit reasoning steps rather than a direct answer, or if it's actually a task where chain-of-thought adds unnecessary overhead without improving accuracy (be honest about this).

## Best for

Developers and prompt engineers working on tasks involving multi-step reasoning, classification with nuanced criteria, or calculations, where direct-answer prompts produce inconsistent or shallow results.

## Compatible tools

- Claude
- ChatGPT

## How to use

- Paste your current direct-answer prompt along with the task type and any reliability issues you've observed.
- Use the identified reasoning steps to understand what the model actually needs to consider for this task.
- Replace your original prompt with the rewritten chain-of-thought version.
- If the honest assessment says chain-of-thought isn't needed for this task, keep your original direct prompt instead.

## Customization tips

- For tasks feeding into automated systems, specify the exact format needed for the final answer line so it's easy to parse out from the reasoning text.
- If accuracy is critical, ask for an additional self-check step at the end where the model reviews its own reasoning for consistency before finalizing.
- For very simple tasks, explicitly ask whether chain-of-thought is worth the added token cost and latency.

## Example input

Original prompt: 'Is this customer review positive or negative?' Task type: Classification. Current issue: Struggles with mixed reviews that have both positive and negative elements, defaulting to whichever sentiment appears first.

## Example output

Identified reasoning steps: identify all sentiment-bearing statements, categorize each as positive/negative/neutral, weigh the overall balance and the reviewer's apparent final judgment, then classify. A rewritten prompt instructing the model to list sentiment statements first, categorize each, then reason about overall balance before stating 'Final classification:' on its own line, with an explanation that this task benefits from explicit reasoning specifically because mixed-sentiment reviews require weighing conflicting signals rather than pattern-matching to the first sentiment mentioned.