Clojurians Log v2

It definitely isn't part of the Aider prompts because the debugging process was basically "smash your code to bits and then tell you everything works perfectly". There was no gathering of information or forming hypotheses. Probably smart people have added this to other tools, I don't know.

I think all the others basically compete on just that. There's not much else they can compete on 🤣

user story or llm system prompt? 🤔

the latter probably

GPT-5?

Yep that, but also gemini-2.5-pro is pretty eager. I'm using gpt-5-codex and the gemini model the most at the moment.

I tried GPT-5 for a really short while. It was just too much fighting with the model. “No, this is the right way.” <proceeds with obviously incorrect solution>

I feed https://github.com/chr15m/ai-context/blob/main/AGENTS.md to every LLM (through aider) and a lot of it is spent just pleading with the LLM to chill out and not modify so many things.

Not sure about Aider, but I suspect something like that is part of most system prompt in most Agentic tools. Maybe I'm wrong, but I assumed so. I wonder if they expose it, so we don't have to duplicate things like that