OpenAI shares prompting tips for GPT-6 Astra, including a blocklist of slop words

OpenAI 发布 GPT-6 Astra 提示词指南，含 slop 词屏蔽清单

OpenAI's docs show GPT-6 Astra asks clarifying questions more often than GPT-5.6 Sol, which makes it a better collaborator but also causes it to stop when users expect action. To push it toward initiative, prompts should tell it to infer intent and show a bias toward action. The model is sensitive to contradictory instructions in skill files like AGENTS.md, so OpenAI recommends auditing them and giving user instructions explicit priority. A debugging prompt can force the model to name the exact file and line that caused a pause. For writing style, Astra overuses lists, tables, and repeated phrases. OpenAI published a blocklist of slop words—including “delve into,” “leverage,” and “it’s worth noting”—and warns against made-up compound terms. The model also under-delegates to sub-agents; developers need to spell out when and how much to hand off.