Auditing Skills and AGENTS.md for GPT-6 Astra
- Source
- x.com
- Date
Context
Instructions accumulated for older coding agents may need revision when the model changes. In a guide to auditing skills and repository instructions, pvncher recommends keeping descriptions short and loading detailed guidance only when a task needs it. The guide also argues for explicit completion criteria: if a request includes running the implementation and fixing failures, say so. Applied to OpenAI's GPT-6 Astra, the advice is to remove obsolete procedural demands, describe decision boundaries carefully because the model takes them seriously, and give a concrete definition of finished work.
- 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.
- context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
“Models have gotten much better at understanding nuance and ambiguity, so overly specific guidance can now hinder results where it previously helped.”
“GPT-6 Astra is thorough, but it can be more tentative about how far to take a task.”
“A new model is a good opportunity to clean your house.”
“The local tests use disposable fixtures and have no production access. Run them, fix failures caused by the requested change, and rerun affected tests without asking for approval at each step.”
Checking sign-in…
Loading comments…