AI Calls Tools Repeatedly Sometimes

[chubes4]

I have had a recurring bug where the AI keeps calling the same tool repeatedly.

This does not happen every time, but it happens often enough to be annoying, and to limit my willingness to set and forget this plugin.

I believe this is a directive problem, but i haven't quite figured it out yet.

The one thing i am trying to avoid is neutering the system by explicitly ending the turn after a successful handler tool call. I would like this to be completely fluid, so there will never be a need to end a turn programmatically, within reason.

The AI should know when it is time to stop.

Most of the time, it does. Sometimes, it still fucks up.

I have an 8 turn limit on conversations right now. This is probably too restrictive, but i am trying to perfect the directive system.

Building agents is all about the prompts.

[chubes4]

I ended up resolving this by ending the turn using PHP after a successful handler tool call.

For a while i wanted to achieve it without a hard stop, but for now, the hard stop works and stops duplicate publishing.

The duplicate publishing is a big problem in a system like this because it is designed to be run without supervision.

Duplicates would often go unnoticed causing user confusion and generally unprofessionalism.

If i can find out how to make it work more fluidly with the agent, i will try a different approach. But for now, I am closing this issue.