Engineer’s task brief from GPT-6 Pro
Before a task goes to Codex: GPT-6 Pro finds what you haven’t thought through and writes a brief an engineer could take.
1.5K saves on X, as of Posted
I want to do this: [DESCRIBE WHAT YOU WANT TO DO IN ONE SENTENCE].
Assume an engineer who can solve problems independently will take over, but they haven’t seen our conversation. Using the materials I’ve provided, first help me clarify what problem I’m trying to solve, whether any requirements conflict, and what missing information could affect the approach. Ask me about decisions I need to make. Don’t turn assumptions into confirmed requirements.
Once you have enough information, write a task brief I can send directly to that engineer. Include the necessary background, deliverables, and any project location, starting branch, and change boundaries already established in the materials. Define specific checks and what counts as passing. If the materials don’t establish the files, paths, or test commands, mark those as needing confirmation when the engineer takes over. Don’t invent them.
Make the brief understandable without reading our conversation. Leave implementation details to the engineer’s judgment. Don’t prescribe every step or expand the scope. At the end, output only the task brief.Next Give GPT-6 Pro your ideas, screenshots, relevant code and constraints, then paste it.
Before you start coding, have GPT-6 Pro point out what you haven’t thought through. Don’t ask Pro to “optimize my prompt”. Ask it to “write a task brief I can hand directly to an engineer”.
Asking a model to write a prompt for AI may lead it to underestimate the agent doing the work and spell out every step. Framing it as an assignment for an engineer tells it to write for someone who can make independent decisions but lacks the project context.
Read the brief and make sure it reflects what you meant. Then give it to Codex, along with the materials it needs, and let it get to work.
Share this tip
