OpenAI has released documentation for GPT-6 Astra, focusing on how to reduce unwanted behaviors. The model is more likely to ask clarifying questions than GPT-5.6 Sol, which can lead to it stopping prematurely. To counter this, OpenAI suggests prompts that encourage the model to infer user intent from context and show a bias toward action.
The model is more sensitive to context and longer instructions. Unclear or contradictory instructions in skill files, such as AGENTS.md, can cause the model to stall or deviate. OpenAI recommends auditing all accessible skill files and giving explicit priority to user instructions.
To help developers trace errors, OpenAI suggests a debugging prompt that requires the model to name the specific skill file and quote the exact instruction that caused a pause or change in direction.
Regarding writing style, GPT-6 Astra tends to use lists and tables by default. For prose, developers should explicitly request concise paragraphs in plain language and active voice. OpenAI also provided a blocklist of "slop words" to avoid, such as "delve into," "leverage," and "it's worth noting," as well as made-up hyphenated compounds.
The model can delegate work to sub-agents in parallel, but does so less frequently than expected. Developers should specify exactly when and how much it should delegate. Additionally, for coding tasks, the model may run excessive tests for small changes; OpenAI recommends limiting test reruns to cases where new failures occur.
Developers can use Codex with the OpenAI Docs skill to automatically migrate projects to GPT-6 Astra using the command: $openai-docs migrate this project to GPT-6 Astra.
