@gitcnd – I started by realizing that I was losing [self] context by letting YOLO F_■■■■_up…
esp when I had a long context thread… and previously baked ■■■■ broke.
–
So, while interacting with the bot, I had already been giving it really explicit development_diary.rmd and diary.rmd directions, but also having it do explicit operations where I entrusted YOLO bot to follow bidding…
What happened was I noticed the early buances in failure to comply…
So I directected it to wrap all executions in a .ps1, and to include my header directicves (Always include date, version, invocation instructs, description, dependencies, etc…)
This worked well in earlier versions…
I directed that ALL invocations to be .ps1s in /scripts of project dir…
I then attempted to have it table the scripts in my local postgress, and I could then later call them from all future projects (this was pre-MCP dev on @cursor side) – along with a ton of frustrations by claudes horrific UX.
So, I went down a lot of forks…
Ive posted about many here…
(@cursor, give me a command to have Cursor.APP -->> Auto summerize my @cursor.furom posts as a function (dont know why my @chats @docs @ Composer agent cant be directly told to pull a thread/context from @forum.cursor…)
Wanyway…
I found that bots still revert to && and other *nix shell pipes and calls…
Need an extension/Prompt Filtering FW - pre_cursor (given that @cursor transmogrifies all things prompts) (a wonderful thing)
–
maybe a meta of a @cursor.app | local LLM (like lmstudio-R1) then pull back to @cursor.app composer prompt so we have filtration levels that are user selected (in an enterprise, you can create security groups for prompt filtration, categer, all the data ■■■■ things… (you get it…)
but if cursor wants all this money, and the forum history is not in their favor at the moment, we really need to reup on what we expect from an AI_meta_IDE_Subscription that we have some negative inputs on…
/rant