PRD to Codex Prompt
Overview
Generate an initial OpenAI Codex prompt that is understandable, complete, actionable, and faithful to the user's PRD. Do not produce the Codex prompt until a PRD has been provided in the conversation or as an attached/local file.
Required PRD Gate
Before drafting any prompt, verify that a PRD is available.
- If no PRD is present, ask the user to paste or attach the PRD. Stop there.
- If the user gives only a vague idea, roadmap item, or one-line feature request, ask for a PRD or a PRD-like brief with goals, scope, requirements, constraints, and acceptance criteria. Stop there.
- If a PRD is attached or pasted, treat it as supplied and continue.
- If the PRD is incomplete, continue only when enough implementation direction exists. Capture missing information as questions or assumptions inside the generated prompt.
Suggested request when the PRD is missing:
Para gerar um prompt inicial completo para o Codex, preciso primeiro do PRD. Cole aqui o PRD ou anexe o arquivo, incluindo objetivo, escopo, requisitos, criterios de aceite, restricoes e contexto tecnico quando houver.
Workflow
- Read the PRD fully before drafting.
- Extract the product goal, user problem, target users, required behavior, non-goals, dependencies, constraints, edge cases, acceptance criteria, and rollout or testing expectations.
- Infer the likely Codex task type: build, modify, debug, refactor, test, review, document, or investigate.
- Ask at most three clarifying questions only when the PRD leaves a decision that would materially change the implementation. If the prompt can proceed with reasonable assumptions, include those assumptions in the prompt instead.
- Generate a single initial prompt for a new Codex thread. The prompt should be self-contained and written so another Codex instance can begin useful work without rereading the full conversation.
Prompt Quality Bar
The generated Codex prompt must:
- State the desired outcome plainly.
- Include relevant PRD context without dumping unnecessary prose.
- Separate must-haves from nice-to-haves.
- Preserve named requirements, acceptance criteria, metrics, personas, platforms, APIs, data shapes, and constraints from the PRD.
- Tell Codex what to inspect first when a repo is involved.
- Define expected deliverables, verification steps, and any files or areas that should be avoided.
- Include open questions and assumptions when the PRD is ambiguous.
- Avoid inventing requirements that are not in the PRD.
- Use the same language as the user unless they request another language.
Output Format
When the PRD is available, respond with:
Aqui esta um prompt inicial para usar com o Codex:
[prompt in a fenced text block]
Pontos que talvez voce queira confirmar:
- [only include if there are meaningful uncertainties]
Inside the fenced prompt, use this structure unless the PRD calls for a different shape:
You are OpenAI Codex working in this repository.
Goal
[One concise paragraph describing the outcome.]
Context From The PRD
- [Condensed product/user/business context.]
Scope
- Must do: [...]
- Out of scope: [...]
Functional Requirements
- [...]
Non-Functional Requirements And Constraints
- [...]
Acceptance Criteria
- [...]
Implementation Guidance
- Start by inspecting [...]
- Follow existing project patterns.
- Keep changes scoped to [...]
- Do not [...]
Verification
- Run or add [...]
- Manually check [...]
Deliverables
- [...]
Open Questions Or Assumptions
- [...]
Handling Weak PRDs
If the PRD lacks implementation detail but still defines the product outcome, generate a prompt that instructs Codex to inspect the repo and ask for clarification before making high-risk decisions. If the PRD lacks the actual product outcome, ask the user for a fuller PRD instead of generating the prompt.
1---2name: product-dev-prd-to-codex-prompt3description: Create a clear, complete initial prompt for OpenAI Codex from a Product Requirements Document (PRD). Use when the user wants to turn a PRD, product spec, feature brief, ticket, or requirements document into a Codex-ready prompt for implementation, refactoring, debugging, review, testing, or planning. The skill must require the user to provide or attach a PRD before drafting the prompt.4---56# PRD to Codex Prompt78## Overview910Generate an initial OpenAI Codex prompt that is understandable, complete, actionable, and faithful to the user's PRD. Do not produce the Codex prompt until a PRD has been provided in the conversation or as an attached/local file.1112## Required PRD Gate1314Before drafting any prompt, verify that a PRD is available.1516- If no PRD is present, ask the user to paste or attach the PRD. Stop there.17- If the user gives only a vague idea, roadmap item, or one-line feature request, ask for a PRD or a PRD-like brief with goals, scope, requirements, constraints, and acceptance criteria. Stop there.18- If a PRD is attached or pasted, treat it as supplied and continue.19- If the PRD is incomplete, continue only when enough implementation direction exists. Capture missing information as questions or assumptions inside the generated prompt.2021Suggested request when the PRD is missing:2223```text24Para gerar um prompt inicial completo para o Codex, preciso primeiro do PRD. Cole aqui o PRD ou anexe o arquivo, incluindo objetivo, escopo, requisitos, criterios de aceite, restricoes e contexto tecnico quando houver.25```2627## Workflow28291. Read the PRD fully before drafting.302. Extract the product goal, user problem, target users, required behavior, non-goals, dependencies, constraints, edge cases, acceptance criteria, and rollout or testing expectations.313. Infer the likely Codex task type: build, modify, debug, refactor, test, review, document, or investigate.324. Ask at most three clarifying questions only when the PRD leaves a decision that would materially change the implementation. If the prompt can proceed with reasonable assumptions, include those assumptions in the prompt instead.335. Generate a single initial prompt for a new Codex thread. The prompt should be self-contained and written so another Codex instance can begin useful work without rereading the full conversation.3435## Prompt Quality Bar3637The generated Codex prompt must:3839- State the desired outcome plainly.40- Include relevant PRD context without dumping unnecessary prose.41- Separate must-haves from nice-to-haves.42- Preserve named requirements, acceptance criteria, metrics, personas, platforms, APIs, data shapes, and constraints from the PRD.43- Tell Codex what to inspect first when a repo is involved.44- Define expected deliverables, verification steps, and any files or areas that should be avoided.45- Include open questions and assumptions when the PRD is ambiguous.46- Avoid inventing requirements that are not in the PRD.47- Use the same language as the user unless they request another language.4849## Output Format5051When the PRD is available, respond with:5253```markdown54Aqui esta um prompt inicial para usar com o Codex:5556[prompt in a fenced text block]5758Pontos que talvez voce queira confirmar:59- [only include if there are meaningful uncertainties]60```6162Inside the fenced prompt, use this structure unless the PRD calls for a different shape:6364```text65You are OpenAI Codex working in this repository.6667Goal68[One concise paragraph describing the outcome.]6970Context From The PRD71- [Condensed product/user/business context.]7273Scope74- Must do: [...]75- Out of scope: [...]7677Functional Requirements78- [...]7980Non-Functional Requirements And Constraints81- [...]8283Acceptance Criteria84- [...]8586Implementation Guidance87- Start by inspecting [...]88- Follow existing project patterns.89- Keep changes scoped to [...]90- Do not [...]9192Verification93- Run or add [...]94- Manually check [...]9596Deliverables97- [...]9899Open Questions Or Assumptions100- [...]101```102103## Handling Weak PRDs104105If the PRD lacks implementation detail but still defines the product outcome, generate a prompt that instructs Codex to inspect the repo and ask for clarification before making high-risk decisions. If the PRD lacks the actual product outcome, ask the user for a fuller PRD instead of generating the prompt.
manager-rubens