Looks like a token-wasting plugin. But

[ashen sapphire]

Moving to a thread to reduce spam in the main channel.

Yes, a lot of the baseline functionality isn't token-optimized and I don't think I'm actually using any of the baseline functionality at this point. But the baseline structure of the various hooks and the index.js has been very useful for implementing my own development. It's a good springboard to demonstrate what is possible, imo, and helped me understand what was going on while I was first starting out.

And yes, most of my efforts recently are in optimizing AGENTS.md and other core markdown files to make them as concise as possible and optimized for LLM agents instead of for humans.

[gentle dome]

That is very true: one must explore and experiment, fail or succeed, but everything is useful 🙂

[ashen sapphire]

It's easy to keep adding and adding to the core markdown files and over time you can get a lot of bloat, contradictions, and obsolete stuff. Need a periodic review to prune and optimize, imo.

[gentle dome]

The bot is all about patchwork. I almost deleted it in anger multiple times due to this… It does not generalize right unless you painstakingly teach it to. That is probably one of the greatest challenges.

But sometimes you can just forcefully make it do what you want. My memory solution is not a fancy hook (although i have one for one particular fault case), but a strongly stated algorithm. Simply speaking: "your task is not complete unless it is remembered" (there are more nuances there, it is a 6-step checklist in total)

I have two such checklists with different goals: one in agents, one in soul. So far, they hold great