OpenAI tells developers to strip down prompts and approval rules for GPT-6 Astra

OpenAI is advising developers to trim accumulated instructions when moving their coding workflows to GPT-6 Astra. In a blog post summarized by The Decoder, OpenAI's Eric Provencher writes that overly long skill descriptions, blanket reading requirements, and rigid approval rules can get in the model's way. Instructions that have piled up over time can consume context or cause GPT-6 Astra to stop work too early. He recommends tying instructions more tightly to specific tasks, defining more clearly when a job is done, and reviewing skills, AGENTS.md files, and task prompts whenever switching models. The underlying argument is that more capable models need less hand-holding, which matches earlier OpenAI guidance on model transitions. Skills are prompts stored as Markdown files that can bundle resources and scripts, and Provencher says they work best for specific workflows or applications. Their names and descriptions are loaded into the model's context so Codex can pick the right one. Too many skills force Codex to truncate those descriptions, stripping out information it needs to choose correctly, and conflicting or overly broad descriptions can pull in instructions the model does not need. The guidance affects teams that built layered guardrails for earlier models and now find them counterproductive. It arrives as Astra continues rolling out to ChatGPT subscribers and developer tooling.