Feature Idea: Pre-Action Review

[a-bonus]

Integrate an optional, automated AI review step into Desktop Commander's execution workflow before it takes action.

After generating a plan (e.g., with Claude), DC can send it to a different AI, like Gemini Pro 2.5 exp (free), for critical analysis of potential issues, risks, or improvements.

DC presents the original plan alongside the AI reviewer's feedback directly to the user.

The user must explicitly review this feedback and approve the plan before DC proceeds with execution.

Potential: This could provide a crucial safety net, since gemini would also have read capabilities in the codebase. The entire plan -> review -> approve -> act cycle stays within Desktop Commander.

Thoughts anyone? I already kind of do this with Ai-studio before I implement major changes.

[wonderwhy-er]

I kinda feel that we need modes of operation in that sense.
Would be cool if AI could choose them themselves.

1. Exploration - where AI just goes YOLO with goal of trying and testing and learning things
2. Planing - what you somewhat describe
3. Executing - where it works on part of the plan

Currently DC with Claude can go bit too far and start to try unplanned things.

But there is also alternative approach.
If you have well implemented ability to pause and roll back when AI went in wrong direction that feels better to me.
That is why I prefer DC with Claude to Windsurf or VS Code Copilot. Those are very careful and ask you a lot of things... That feels worse.

We have things to do and improve. Planing modes are not on short term roadmap in that sense.

[a-bonus]

Ya totally! In my experience DC works really well at implementation of a plan once it's already set out. And it's also really good at discovery and research of the codebase.