I’m just starting to try out Intern but I’m running into issues that other users probably also will or have.
My project: create a Dr. Who styled classic text adventure, AI-supported (dialogs and game to be generated).
The resulting code is a clear setup using multipe components. The issue I find is that class and method definitions and use are often not consistent, resulting in errors. It can be naming differences or using a class element instead of a method.
I’ve tried regenerating, but that typically results in a significantly different setup, introducing different inconsitencies and errors.
I’ve also tried giving Intern the whole project back and re-test and refactor. That typically fixes some stuff and breaks other stuff.
And of course, I’ve tried manually debugging, I typically fix a lot of stuff until I reach the limits of my (AI-supported) programming skills
What are good practices you use to get a working result?
I like to specify to have it ask me clarifying questions before proceeding. If you don’t specify that, sometimes it may but sometimes not, and in the latter, it’s the information vacuum that can often lead to incompatibilities or weirdness.
FWIW, we’ve been tracking a lot of bugs and improvements and many of these things should be ironed out over time
(P.S - I’m making a Doctor Who project right now on Intern too! Looking forward to new season this weekend)
I have some similar experiences with the rabbitOS intern. Some results are perfect and after a bugfix prompt, I get other results, layouts etc..
But after some experiences, how prompts and the AI work, maybe the following additional prompt can perform better results in your opinion.
If you have to bugfix, you can try to add the following sentence/prompt in your task/chat:
Fix [the button XYZ/element XYZ] to get [the result XYZ]. Keep the layout, the features, the design and all variables/elements. Do not change variables and the layout. All features have to be functionable. Finalize it with tests to maintain functionality of the software.
With this additional prompt, I could fix bugs without layout changes etc. in many cases. Not in all cases, but the probability was higher.
Maybe this can help you.
Also it is possible to add the following prompt in your first chat/task:
…
The functionality of the program has the highest priority. At the end of your tasks do tests and make sure the functionality of the program is given.
I will try another bugfixing prompts the next days.