So, I use agents to write a ton of stuff now: e-mails, client messages, commit messages, comments, etc.
The problem is that AI writing is VERY easy to spot and most of the time I don't want it being sent under my name, since it does not sound like me at all.
After failing at trying to fix it with rulebooks or skills like humanizer, I decided it was better to try to "catch" every time the agents write public-facing text and force the agents to send it to a single queue and wait for my editing and approval.
Why this approach?
I still want the agents to draft the text for me, since it's time consuming to pull up the necessary information and create the initial structure of the message.
All I need to make the text sound like me is to lightly re-structure some sentences, remove the enthusiasm, trim it down, remove words I don't use, etc..
So I made this for myself and have used it for a while now, find it very useful, so I decided to make it public.
If offers a nice looking UI with a proper text editor that allows me to easily edit and approve/reject public facing text produced by any of my running agents. It's super simple, can be used locally or with cloud session agents (but hooks don't work there, so there's no way to enforce its use there).
Every message shows up as a draft that I can edit and it’s got all the context right there: what it is, who's gonna read it, and why the agent even wrote it.
Then, I can edit it, hit approve and the agent sends my version of the text.
I figured other people might also find it useful, so here it goes.
Any feedback is much appreciated.
Comments URL: [https://news.ycombinator.com/item?id=49962951](https://news.ycombinator.com/item?id=49962951)
Points: 1