Need help transitioning Little Coder from benchmark testing to real-world web and app development. #20
-
|
I’ve reviewed the project and found several interesting aspects, but I’m unsure how to remove components that are primarily benchmark-related. A clearer explanation of Little Coder’s design would be helpful - for example, clarifying why skills are injected on demand by skill-inject extension. I understand the project evolved from an initial “pi” version through iterative benchmarking and debugging. But one can only guess why a particular feature was implemented and do I really need to use it or not. |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment
-
|
Hey! Thanks for taking interest in little-coder. Actually, I didn’t start from pi at all, but from a claude code replica. I had a strong hypothesis about a potential mismatch between the harness and the model underlying it. I wrote extensively about it here: https://open.substack.com/pub/itayinbarr/p/honey-i-shrunk-the-coding-agent Only later I rewrote it on top of pi because the community asked it dozens of times here: |
Beta Was this translation helpful? Give feedback.
Hey! Thanks for taking interest in little-coder. Actually, I didn’t start from pi at all, but from a claude code replica. I had a strong hypothesis about a potential mismatch between the harness and the model underlying it. I wrote extensively about it here: https://open.substack.com/pub/itayinbarr/p/honey-i-shrunk-the-coding-agent
Only later I rewrote it on top of pi because the community asked it dozens of times here:
https://www.reddit.com/r/LocalLLaMA/s/N4cv3u7Pyj