Show HN: Declaude

I had to develop a course related to quantum chemistry and unfortunately, Claude just kept writing it in its own very annoying claude-speak. It actually cost our team a lot of time and tokens to wrestle it to just speak like a normal human.

We used skills, as well as initial prompts and subagents to make sure that all of the tics were gone, but there was one too many times where it just wouldn't fix itself despite telling it in the prior prompt!

I made speak-english based off of https://github.com/gvzdv/claudish-to-english/commits/main/ and it runs a qwen model on my gcp servers.

It can convert documents or respond in-line in sessions with claude and I anticipate building a solution for prime-agent.

Show HN: Declaude
speak-english.tenken.co

Super cool! This has been annoying me for quite a long time. Feels like Claude models’ natural language ability is squeezed out for the improvements of coding and reasoning. I think that’s why a weaker model can possibly do better in English.

However, I was wondering if any skills or CLAUDE.md would help as a simpler alternative, by better context management and instructions?

Hey yeah, I have used skills and CLAUDE.md, but over time claude actually starts to remake the mistakes I told it not to make. It's like whack-a-mole, I want it to fix one issue then it creates another, so I run a separate server.