OpenAI tells devs to trim Skills, prompts for GPT-6 Astra

An OpenAI developer blog post tells teams building coding agents on the new GPT-6 Astra to trim the scaffolding they carried over from earlier models. "With each release, it's been worth revisiting those assumptions, but with GPT-6 Astra, it's more important than ever," it says, arguing that "skills, AGENTS.md, and your task prompts are all shaping how the model gets work done."

The specific complaint is bloat. "Reading a skill takes up context, bringing you closer to compaction and introducing guidance that may not apply," the post warns, and a description like "Create and validate Postgres schema migrations. Use when working with databases, queries, models, or persistence" is the kind that fires whenever the model touches a table. Astra also runs tests unprompted: "Previous models needed encouragement to run tests and check their work. GPT-6 Astra does that on its own," making the old 'make sure to run the tests' line into noise.

There is a trade-off. "GPT-6 Astra is thorough, but it can be more tentative about how far to take a task," the post says, and developers used to GPT-5.6 Sol continuing for long stretches may see Astra pause for review sooner. Its suggested fix is to define completion up front, then ask Astra itself to audit the existing scaffolding.