Continuous Self-Feedback Mode?

[cooleydw494]

I can't be the only person who wants this can I? Sure a lot of things don't require self-feedback on every action, but some things do, and other things run into errors so commonly without it that its worth spending the API request money on commands that don't need it just to make sure it stays aligned.

Is there already a way to do this? If there is I can't find out how.

[cooleydw494]

Like I've already spent a few hundred dollars developing strategies that almost work well enough but always encounter something that can't be automated continuously. If continuous self-feedback successfully kept a usually-predictable prompt from running off the rails and erroring into the API (spending money) over and over until it causes a fatal error, I think the results:cost ratio would get a lot better tbh.

[Boostrix]

There is the human feedback option being talked about and several ideas about implementing that - either way it could be be adapted to solicit self feedback based on the current situation / progress. You could file an issue / feature request

In fact, for starters we could use this to bail out of seemingly redundant looping.

This (PR #3668 ) is just based on hashing full thoughts + new decision (command + args) and incrementing a counter every time we get to see the same "situation":

[cooleydw494]

that's very welcome. Good to see it.