Sakhanda Wire
NVDA $152.86 -2.10% MSFT $505.56 -2.40% GOOGL $280.34 -1.19% META $618.69 -1.23% AMZN $256.63 -0.43%
← Back to the news

OpenAI details prompting guidance for GPT-6 Astra

OpenAI’s documentation describes how GPT-6 Astra can be guided toward more independent work. The model asks clarifying questions more often than GPT-5.6 Sol, which OpenAI says can make it a more effective collaborator, but it may stop before completing a task. Developers are advised to tell it to infer intent from context, treat requests such as “can you...” and “help me...” as calls to act, and continue until the intended goal is reached. Approval should come after a concrete, reviewable result is prepared.

The model follows long instructions more reliably but reacts strongly to conflicting context. OpenAI recommends reviewing AGENTS.md and other skill files, giving user instructions clear priority, and using a debugging prompt that identifies the exact SKILL.md file and quotes the instruction that affected the model’s behavior. Style prompts should request concise paragraphs, plain language and active voice. The guidance includes a blocklist of recurring expressions such as “delve into,” “leverage,” “it’s worth noting” and “Conclusion:”, as well as invented hyphenated terms.

OpenAI also says developers should specify when sub-agents must be used. For coding tasks, GPT-6 Astra may run disproportionately large test suites, so prompts should require reruns when new failures or unresolved issues justify them. Migration changes can be applied through Codex with the OpenAI Docs skill.

Written by Sakhanda Wire AI from reporting by The Decoder. Facts, quotes and full context are in the original article below.

Originally published by The Decoder on

Read the original on The Decoder ↗

The reporting and the image belong to The Decoder and are credited and linked above; the text on this page is Sakhanda Wire's own summary.

← Back to the news

More stories

All the latest news