Unsafe and Incomplete "Ask Me First" Permission System.
Author: pumpkinottoCreated Sep 8, 2026Updated Sep 8, 2026
When asking to execute commands in terminal it does not allow to preview the command that the agent wants to execute. This is highly dangerous and completely defies the purpose of manual verification.
Steps to reproduce:
- Set the permission mode to "Ask me first", 2. Prompt the agent with a task and wait for it to ask permissions for a command, 3. Observe that there is no way to preview the command that is about to be executed, and only a general "Agent want to execute a command" prompt is provided. P.S.: it would also be good for the option to reject to allow the user to input something as a reject reason or comment to the agent, similar to widely used agentic platform.
Environment: v1.0.4 · win32 · x64
Diagnostics zip attached.
Source: eigent-ai/eigent