ChatGPT/Project/How To/Refine the Project Prompt using ChatGPT

• How To Refine a ChatGPT Project #Prompt Using o3

  • My Notes

    • 2025-05-16 Fri
      • I've had great luck using this template to iteratively improve the ChatGPT/Project/Instructions to align with my expectations. These models are excellent prompt engineers, but, perhaps ironically, they need to be prompted a certain way in order to access their highest capabilities in this regard. For example, I used it to dramatically increase the alignment of ChatGPT/Project/Instruction/Diataxis Logseq quickly, using ChatGPT/Dictation

  • Goal

    • Produce a concise, well-structured prompt that meets project-specific length or formatting targets (e.g., 7 000 – 7 500 chars) by iterating with a reasoning model such as o3, using ChatGPT/Code/Interpreter to measure length.

  • Preconditions

    • Access to ChatGPT with the o3 reasoning model.
    • Code Interpreter (python_user_visible) enabled.
    • A draft prompt that needs trimming or re-organization.
    • Knowledge of your project’s formatting rules (length limits, required sections, special markup, etc.).

  • Procedure

    • 1. Map Your Existing Prompt

      • Skim the draft and label logical parts (intro, bad example, critique, current rules, request).
      • Note all hard requirements (length cap, mandatory headings, prohibited markup, etc.).

    • 2. Build a Generic Prompt Template Skeleton

      • Ask ChatGPT to extract section headings and convert them into a reusable template.
      • Use the structure below; fill [placeholders] with topic-specific text:
        markdown

        # Prompt Template: Refine Project Rules
        
        ## 1 Context  
        > Currently in your project instructions: **[summary of situation]**
        
        ## 2 Anti-Pattern Example  
        <EXAMPLE_OF_HOW_NOT_TO_DO_IT>  
        **[paste flawed sample]**  
        </EXAMPLE_OF_HOW_NOT_TO_DO_IT>
        
        ## 3 Problem Analysis  
        The above example has problems:  
        1. **[issue A]**  
        2. **[issue B]**  
        3. **[issue C]**
        
        ## 4 Current Project Rules  
        <PROJECT_RULES>  
        **[paste current rules]**  
        </PROJECT_RULES>
        
        ## 5 Refinement Request  
        Please rewrite the rules so that **[desired outcome]**.  
        **Constraints:**  
        - Keep length within 7000-7500 characters (it MUST be <8000 characters).  
        - Follow all formatting guidelines in Section 4.  

    • 3. Instruct the o3 Model to Rewrite

      • Send the filled-in template to ChatGPT (o3). You may need to select o3 from the model drop-down at the top.
      • Explicitly request: “Use Code Interpreter to count characters and iterate until the result fits the required range.”

    • 4. Measure with Code Interpreter

      • When ChatGPT returns a draft, have it run:
        python

        rules = """...rewritten text..."""
        print(len(rules))

      • If the count is outside your target range, ask for automatic trimming or expansion and re-run the count.

    • 5. Validate Output

      • Confirm the rewrite matches all project formatting rules (headings, markup, fences, etc.).
      • Ensure no prohibited elements remain (e.g., unsupported tags or placeholders).

    • 6. Finalize

      • Save the approved prompt to your documentation repo or knowledge base.
      • Archive the template for future refinements across other projects.

  • Troubleshooting

    • Output exceeds length requirement → Ask o3 to “compress explanatory prose, keep tables.”
    • Key formatting rule ignored → Remind model: “Ensure compliance with project’s formatting checklist.”
    • Prohibited markup persists → Reinforce rule: “Strip all unsupported tags or symbols.”
    • Code Interpreter unavailable → Manually copy text into a local script (e.g., wc -m) to count characters.

  • References

    • OpenAI Platform documentation.
    • Prompt Engineering best practices (OpenAI Cookbook).