Hey, thanks for the report. This is a known issue. When the Plan mode prompt times out, the agent keeps working instead of waiting.
There are already a few similar reports:
- While planning, agent questions timing out resulted in agent writing code without permission
- ‘Plan’ mode is not respected by the Agent
The team is aware. As a workaround, you can add a rule in .cursor/rules/ that reinforces Plan mode behavior:
---
description: Plan mode safety
alwaysApply: true
---
CRITICAL: In PLAN mode, NEVER edit files or run commands. Only describe the plan. Wait for explicit user approval before any implementation.
It’s not perfect, but it helps reduce how often it happens. I’ll pass this along to the team.