OpenAI dropped the GPT-6 Astra migration guide. Instead of manually working through your codebase against their new best practices, they have an official skill that lets Codex read the specs and apply the recommended migration changes.
The skill is called openai-docs. You run it like this:
$ openai-docs migrate this project to GPT-6 AstraIt inspects your project and applies relevant model and prompt changes, flagging broader API or architecture changes when needed. Tested it on a mid-sized repo. Worked out of the box.
Understanding GPT-6 Astra's behaviour
GPT-6 Astra is designed to be a more effective collaborator. Its behaviour may differ from previous models in several important ways.
Initiative and follow-through
The model may ask clarifying questions when additional information could materially affect the result. If you prefer it to make reasonable assumptions and continue, state that explicitly in your instructions.
Instruction following
GPT-6 Astra follows general instructions more closely, including instructions found in files such as AGENTS.md. Review accessible skills, configuration files and project documentation to ensure they do not contain unintended directives.
Personality and writing style
The model may produce detailed, highly formatted responses and repeat certain phrases. Specify your preferred tone, structure, response length and formatting requirements.
Subagent delegation
GPT-6 Astra may delegate tasks less often than expected. Define clear rules for when it should use subagents and how much parallel work it should perform.
Testing and verification
For coding tasks, the model may perform broader testing than a small task requires. Specify the desired testing scope when speed or minimal verification is important.
