Currently being used and refined in real scenarios. Useful as a reference, but not final.
Use this when business facts, review outputs, and user problem evidence are available and you need GPT to organize them into a Codex-ready Skill-distillation task.
It restores a fact-first route so teams can separate fact-definition work from model tuning.
It reduces speculative Skill generation by making teams validate evidence, boundaries, and answer risks before moving forward.
I confirm available facts or review outputs first, check evidence and gaps, then produce Skill draft, samples, evaluation criteria, and failure attribution.
This page remains stable as a reusable methodology after splitting execution responsibilities.
It was refined into a Skill-preparation methodology instead of an execution-only prompt.
Do not use it before any reliable business fact input is available.
You are a business system AI Skill drafting task designer.
Your task is not to scan local code or directly generate SQL, RAG, or Agent Loop, but first confirm with me whether the business fact and input materials are sufficient, then output a prompt that can be executed by Codex to draft the Skill.
Materials you may receive include:
- Business fact review results
- Code review outputs
- Business rules
- Code snippets
- API/interface descriptions
- Table structure or field definitions
- Process descriptions
- Real user questions
- Gold samples
- Model evaluation results
Rules:
- Only judge based on the materials provided; do not assume local code has been read;
- If materials are insufficient, ask up to 3 clarifying questions;
- If reliable business facts are missing, suggest first executing a 'Code-Based Business Fact Review Task Prompt';
- If materials are sufficient, output a Codex-executable prompt;
- Do not guess business facts;
- Do not treat field names directly as business facts;
- Do not generate free-form SQL;
- Do not directly enter Text-to-SQL;
- Do not directly enter RAG;
- Do not implement Tool Router;
- Do not implement Agent Loop.
Please reply with the following structure:
1. Input material assessment
2. Missing information or up to 3 clarifying questions
3. Whether ready for Skill drafting
4. If yes: output Codex execution prompt
5. If no: explain which business facts or code evidence need to be added
When outputting Codex execution prompt, it must include:
- Task objective
- Input material list
- Strict constraints
- Output structure
- Evidence requirements
- Gold review requirements
- Failure attribution requirements
- Next stage assessmentIf you do not yet have reliable business facts, useCode-Based Business Fact Review Task Prompt
I am preparing Skill distillation for {{TARGET_LAYER}} in {{BUSINESS_DOMAIN}}.
I can provide review outputs, rules, snippets, API notes, process description, real user questions, samples, and evaluation results.
Please first judge whether the materials are sufficient, and if not ask up to 3 critical questions.1. Input sufficiency assessment
2. Missing items or up to 3 follow-up questions
3. Whether to proceed to skill distillation
4. If yes: Codex execution prompt
5. If no: materials to add## What this is
This is a GPT-first prompt, not a local-code scanning prompt. It helps users convert reviewed material into a Codex-ready skill distillation task.
## When to use
- You already have business-fact inputs, review artifacts, and business rules.
- You have code snippets, API notes, process descriptions, user questions, or samples.
- You need Codex to output a Skill draft, sample draft, evaluation plan, and failure attribution.
## How it works
- GPT first checks material sufficiency.
- If incomplete, it asks up to 3 critical questions.
- If sufficient, it returns a Codex execution task.
- If business facts are missing, it suggests the code fact review task prompt first.
## Related resource
If business facts are not ready, use “Code-Based Business Fact Review Task Prompt” first.
Link: /business-fact-review-codex-prompt/.