alt Hacker NewsI’ve asked it to use plain English, avoid jargon, explain terms before introducing them. Its first response was to make memory, it forgot after 5 prompts, then it suggested claude.md. Looked good at the start of a session, forgot after 6 interactions. Then it continued suggesting other locations, sometimes correcting its own corrections. Same effect, so today I got annoyed again. And then it suggested a new thing: add a script in a Stop hook in settings.json. I added another one to the PreToolUse, hoping to prevent it from running all kinds of experiments I didn’t ask for or approve of. I still have to see where this ends up.. Maybe this can help you as well.
I wrote a harness for running tabletop RPGs. One of the key things to keep the LLM behaving correctly (such as not controlling the PCs) was being able to inject instructions with every new prompt, which I don't include in the history.
It seems like all harnesses could benefit from something like this.