OpenAI shares prompting tips for GPT-6 Astra including a blocklist of slop words
decoder ·
OpenAI published prompting guidance for GPT-6 Astra, noting it asks more clarifying questions than GPT-5.6 Sol and can stop prematurely. The company recommends prompting the model to infer user intent and show a bias toward action, only pausing for approval when a concrete, reviewable result is ready. GPT-6 Astra is more sensitive to context, so unclear instructions in skill files like AGENTS.md can cause work to stop; OpenAI suggests auditing those files and using a debugging prompt that forces the model to name the exact skill and quote the instruction. The documentation also includes a list of "slop words" to avoid, such as "delve," "leverage," and "it's worth noting," and advises using plain language, active voice, and concise paragraphs instead of lists and tables. On coding tasks, GPT-6 Astra tests too thoroughly for small changes and delegates too rarely to sub-agents; developers should specify delegation parameters and instruct the model to rerun tests only when justified. Full prompts and migration tips are available in OpenAI's model documentation, including a Codex command using the OpenAI Docs skill.
decoder