I keep seeing these snippets of alleged "code" from the Claude Code leak, and thinking that if this is what top tier prompt engineering from one of the most successful companies in the field looks like, they can just have it, I'm not interested.

I already raised two toddlers.

@elb it's so fucking unmaintainable that the only way I can understand it as rational production code is that they believe that they'll fold before anyone (including an LLM) has to maintain it
@ricci @elb I still can't get over how sentiment analysis, **one of the few things where there is a legitimate argument for LLMs**, is outsourced to a hardcoded regex.
@elb I massively miscalculated how tissue-thin Anthropic are, and I consider myself exceptionally critical of them. I suppose I'll just have to learn to adjust