One Prompt Fix Stops GPT-6 Coding Agents From Pausing on Casual Chatter

altryne · x · 2026-09-25

altryne shared a prompt snippet for agents.md that stops coding agents from pausing all work when humans send casual messages.

The core idea: instruct the model that human messages are not always steering — only messages that start, change, cancel, or continue work count as directives. Casual chat, acknowledgements, or questions should be answered normally from current context (or with a single reaction), without reading, writing, restoring/clearing checkpoints, updating Up Next, or scheduling waits. Directly copy-pasteable for teams running long tasks on GPT-6-class models.

Original post →

More from coding & agent

coding & agent channel →