OpenAI AI Updates: September 13, 2026

1. OpenAI Told Developers to Strip Their Prompt Scaffolding Back for GPT-6 Astra

OpenAI published guidance on its developer blog, “Rethinking Skills and Prompts for GPT-6 Astra” by Eric Provencher, arguing that the prompting habits built up around earlier models now actively degrade Astra’s output. Skill descriptions should be short and task-specific, because long ones get truncated and the model then misses the selection criteria it needed. Explicit step-by-step sequences should be cut, since the model handles nuance and ambiguity on its own. In AGENTS.md, blanket rules that force a documentation read on every change should become conditional pointers: architecture docs only for service boundary work, database docs for schema changes, deployment notes when shipping. Safe operations should be explicitly permitted to stop repeated confirmation requests, and prompts should define what “done” means up front rather than setting checkpoints that halt the model early. The guidance also flags legacy restrictions carried over from earlier models as a cause of Astra stopping when it should continue. Source