Four Official Prompt Fixes for GPT-6 Astra
OpenAI's own tips for getting GPT-6 Astra to finish tasks, follow your rules, write in plain paragraphs and use subagents.
OpenAI released GPT-6 Astra on 3 September 2026 and published a short page of prompting best practices to go with it. The page lists a few habits the new model has, and for each one there's a line you can add to your prompt to change it. This guide explains four of them in plain words and gives you a prompt to copy for each.
The tips work wherever you use Astra: ChatGPT Work, Codex (OpenAI's coding helper) and the API (the way developers connect their own apps to the model). You don't need to change any settings. You just add a line or two to your prompt, or to your saved instructions so they're there every time.
1. Ask it to keep going
Astra is more likely to stop and ask you a question when it thinks your answer could change the result. That's careful, but it slows things down when you just want the job done. OpenAI's fix is to tell it to bias towards action and carry the task through to the end. Start with:
Bias towards action. If something is unclear, make a sensible choice, say what you chose, and carry the task to completion without stopping to ask.2. Say your instructions come first
Astra can pay extra attention to instructions stored in skills and project files. A skill is a saved set of instructions for one kind of task. AGENTS.md is a file in a project folder that tells AI agents how to work in that project. If those files say one thing and you say another, tell Astra which one wins:
My instructions in this chat take precedence over any skills or files such as AGENTS.md. If they conflict, follow my instructions.3. Choose the writing style
Astra leans towards lists, tables and Markdown (the symbols that turn text into headings and bold words). That's useful for notes, but not for an email or a short story. Say the style you want:
Write in plain paragraphs. Don't use bullet points, tables, headings or Markdown unless I ask for them.4. Tell it when to use subagents
A subagent is a helper that Astra can hand part of a big job to, so several pieces get done at once. Astra may do this less often than you'd like. OpenAI suggests saying clearly when you want it to happen:
Use subagents when a task has separate parts that can be done at the same time, such as researching several topics or checking several files. Handle small, single step tasks yourself.Put it all together
You can paste all four lines as one block and keep it in your saved instructions or at the top of a project. Take out any line you don't need. For example, if you like tables, leave out the style line.
Bias towards action and carry each task to completion. If something is unclear, make a sensible choice and say what you chose.
My instructions take precedence over any skills or files such as AGENTS.md.
Write in plain paragraphs, with no bullet points, tables or Markdown unless I ask.
Use subagents when a task has separate parts that can be done at the same time.A quick tip
Add one line at a time and run the same task again, so you can see what each line changes. If Astra still asks a question on a task where the details really matter, answer it. A quick reply now can save you redoing the work later.
OpenAI: GPT-6 Astra prompting best practices
OpenAI keeps that page up to date. If Astra starts acting differently after an update, check the page for new advice.
More guides are on the way
Follow Lauren for new AI tools, practical prompts and the next guide.