Anyway to work around this issue where the bot keeps doing stuff from prior tasks?

I am trying to incrementally give the bot things to do, and it keeps repeating stuff from the previous task. I feel like the bot has too many of its own ideas about what to do. How are you guys getting things done with this ai programmer?

Also what’s up with this behavior from the bot: When I ask it to do something, it goes and does whatever the hell it wants. For example I opened a new chat with the prompt,
“Create a new, blank, empty project, with a single document where you place all of the steps needed to create a React Webapp with tailwindcss and a set of libraries suitable for creating animations and sound effects.”

and the bot creates a basic project structure and all of that I told it not to do.

Using claude sonnect via openrouter. I get the same behavior at bolt.new. Just wondering if there are specific config options I can use to prevent the bot from flying off on its own plan all of the time, repeated prior artifacts, etc.

you could try to use the new experimental prompt and see if you get better results with it (if you not already did):

Is it documented somewhere, exactly what that feature is doing? Can I look at the ‘library’ myself?

As it is experimental, its not documented right now. You can take a look at the source code:

and compare it to the default:

1 Like

Just asked bolt / gemini to tell the difference. Think it covers it quite good :slight_smile:

this is a hallucination :sweat_smile: . bolt is not aware of the two prompts, only the one that is currently active. you can ask it to tell its system prompt maybe
but i guess that will break the ,parser as it has to write the artifact examples as well

1 Like

1 Like

Thanks for the pointers about the prompt and the example text with github link.

I am experimenting now with putting my custom rig for my project in that file under the introduction line and tweaking the included rules.

3 Likes