← sources

Be honest… how many permission prompts does your coding agent get before you stop reading them?

source: reddit (reddit:ClaudeAI)published 2026-09-28 15:15ingested 2026-09-29 02:37

retained text

For me it's about an hour. After that I'm clicking approve on autopilot, which means the prompts aren't protecting much anymore.

So what are people actually doing instead? Separate user account? Running it in a VM or a devcontainer? Keeping SSH keys and browser sessions out of its reach somehow? Or just YOLO mode and a good backup?

I work on agent isolation at Jozu, so I have opinions, but I'm really asking. I want to know what setups people trust enough to stop babysitting the agent.

document fetched 2026-09-29 00:23 · canonical https://reddit.com/r/ClaudeAI/comments/1wshclh/be_honest_how_many_permission_prompts_does_your