Stephan Eberle
stevieecodes.bsky.social
Stephan Eberle
@stevieecodes.bsky.social
Ruby On Rails dev@NEW WORK, interested in Go/Flutter/Dart/ and building new things.
That's a good habit! ☺️
December 24, 2025 at 9:24 PM
Ahhhhh.... My son is killing me with that reference...
December 24, 2025 at 8:13 PM
Doing App Store images (or keeping them current). I never can convince myself to do it. Also my design-choices are borderline FDISK. 😬
December 24, 2025 at 10:00 AM
One project is live (Soothing Noise), a demo of the other one is in the "works". I can't type fast enough to keep up with the stream of thoughts...
December 22, 2025 at 8:59 AM
You're taking orders?
December 21, 2025 at 11:04 AM
Congratulations!
December 19, 2025 at 7:12 PM
I'm still reluctant of putting these on the website... 🤔
December 14, 2025 at 7:09 PM
"Das haben wir schon immer so gemacht!", wer kennt's nicht? :)
December 14, 2025 at 5:21 PM
That's why I use Claude Code, as the skills come in handy here.

From there is babysitting and a constant dialogue, checking and fixing and a back and forth.

Always remember: You OWN the code, therefore you OWN the problems it creates. Be mindful about what could go wrong later.

And have fun!
December 14, 2025 at 5:17 PM
I'll then go over each Github Issue and check for inconsistencies etc. I also use the LLM for that.

BEFORE creating the GH issues I decide on the exact tech stack, tell the LLM to use "idiomatic code of 2025" or something like that. I'll have it research what that means by web search.
December 14, 2025 at 5:17 PM
Usually I ask the LLM then to create a comprehensive list of Github Issues for me (on Github in a clean repo) and have every issue created will all the context the LLM would need to develop each issue. Only the common documentation and project goals, guidelines etc. are to be put into the README.md.
December 14, 2025 at 5:17 PM
From there I start with Claude Code (you could also use any other CLI tool) and tell it to read the PRD and and ASK ME QUESTIONS about it. That way I am forced to go into details. I don't let the LLM decide what tech to use (that's a no-go for me, as I'm responsible for whatever it spits out).
December 14, 2025 at 5:17 PM
..the guardrails. My workflow is usually to first brainstorm a PRD and have it cross-checked by another LLM (brainstorm with Claude and then have it checked by Gemini or vice versa).

Then I usually prepare the PRD as .md file and put it into my future project folder.
December 14, 2025 at 5:17 PM
If the desired outcome is below expectations it's oftentimes due to a prompt with incomplete or too wide a focus.

The prompt needs to be very specific and best don't let the tool do everything in one shot. Break down your idea and go the way step by step.

Be specific about your goals and...
December 14, 2025 at 5:17 PM