OpenAI has updated its model documentation with specific prompting guidance and engineering recommendations for GPT-6 Astra. The documentation notes that while GPT-6 Astra collaborates more effectively by asking clarifying questions, it can prematurely pause workflows where users expect continuous initiative.
To counter this, OpenAI advises developers to explicitly instruct the model to infer user intent, demonstrate a bias toward action, and execute read-only or non-destructive tasks autonomously before seeking approval. The guidance also includes debugging strategies, advising developers to prompt the model to cite the exact skill or instruction files (such as SKILL.md) that cause unexpected stalls.
Additionally, the documentation outlines instructions for controlling Astra’s default formatting and writing style. Developers are provided with concrete rules to restrict repetitive formatting patterns, avoid over-nested lists, and suppress unwanted disclaimers or passive phrasing.
Why it matters
Helps developers optimize system prompts and skill configurations for OpenAI’s latest GPT-6 Astra model.
Provides concrete techniques to prevent autonomous agents from stalling during complex task execution.
Offers actionable formatting controls to improve output quality and eliminate generic AI prose.
Source: the-decoder.com
