OpenAI's guide to retuning AGENTS.md and skills for GPT-6 Astra
- Source
- x.com
- Date
Get more out of GPT-6 Astra by revisiting your skills, AGENTS.md, and task prompts. Make skill triggers specific, load guidance when it's relevant, and define what done looks like. https://t.co/UGF0AC8Z5Y
Context
OpenAIDevs pointed developers to a guide on retuning skills, AGENTS.md files, and task prompts for GPT-6 Astra, OpenAI's newest model. The guide frames AGENTS.md as a table of contents that separates permanent rules from contextual guidance, rather than a single long instruction document, and recommends making triggers specific enough that Astra doesn't load the wrong skill for a given request.
Its central argument is that scaffolding built to compensate for a weaker predecessor model can now work against a more capable one: heavy hand-holding, redundant , and bloated skill descriptions that once helped a predecessor model like GPT-5.6 can instead crowd Astra's and blunt its performance. pvncher published a separate, practical audit of the same problem around the same time, covering how progressive disclosure of skill content and explicit stopping conditions reduce that same context bloat.
- agent skill — A reusable instruction file that teaches an agent how to do one job well — the procedure, the tools, and what counts as done.
- guardrails — The checks around a model that block bad inputs and outputs — filters, validators, and permission rules the model itself can't override.
- context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
Checking sign-in…
Loading comments…