OpenAI publishes official prompting guide for GPT-6.2 Astra & Sol in Codex

daniel_mac8 · x · 2026-11-20

OpenAI released an official guide for prompting GPT-6.2 Astra and GPT-6.2 Sol in Codex. Core advice: stop judging reasoning effort by answer length and stop stuffing routine coding rules into AGENTS.md. Distilled rules: state your review bar upfront (e.g. does an icon read cleanly at 24px); pick reasoning effort by investigation depth (Medium for quick surveys, High to trace one cause across systems, Extra High when multiple causes interact); use subagents for cleanly branching tasks — in Ultra mode Codex spawns them autonomously; iterate and verify. Full docs on ChatGPT Learn; append .md to page URLs for Markdown versions.

Original post →

More from coding & agent

coding & agent channel →