Rethinking skills and prompts for GPT-6 Astra | OpenAI Developers
Sep 11, 2026

Summary
This post details how to rethink skills and prompts for GPT-6 Astra, focusing on improving agentic workflows. It introduces best practices for writing concise and context-aware skill descriptions, emphasizing progressive disclosure and avoiding overly specific guidance. It also provides updated guidance for `AGENTS.md` to reflect GPT-6 Astra's improved judgment, suggesting adjustments to decision boundaries and persistence instructions to better align with the model's capabilities and desired task completion.
More from the Agentic Execution Environment trail
- Bringing my LED display to life with GPT-Live-1 and Codex | OpenAI Developers
- Meet Rosalind Workbench: Empowering every scientist to be their own research team | OpenAI Developers
- Automating repetitive work at OpenAI with Codex
- Codex as a platform: build on the open agent harness | OpenAI Developers
- Custom Code Review rules for Codex | OpenAI Developers