miniblog.

Playing with Hermes Agent and it's the first agentic tool where I feel I need to *justify* my changes. It's updating its notes over time. W: Split those two jobs, they're completely unrelated. H: Done. I also updated the skill to note that unrelated topics should be split.

It's confusing how 'REPL' refers to two different UIs: A CLI like IPython or Julia, where things like multiline editing and syntax highlighting need careful implementation. An editor protocol like SLIME or CIDER where multiline editing is a total non-issue.

I've realised that AGENTS.md / CLAUDE.md solve two distinct problems: * It tells the LLM the conventions of the project, avoiding follow-up prompts. * It precomputes a summary of the project, saving a bunch of searches at the start of a session.