Anti-slop
Out of the box, an assistant writes like an assistant. Anti-slop is a fixed set of rules you hand it so the tells go away.
Those tells are easy to spot once you know them: em-dashes, filler like robust and seamless, openers like it's worth noting that, the negate-then-affirm rhythm (this isn't X, it's Y). The rules ban them across every bit of prose the assistant produces, whether a vault note, a doc, a message, or a deliverable.
The tool is robust — seamless, even.
The tool holds up under load.
It's worth noting that this matters.
This matters.
This isn't a feature, it's a philosophy.
This is a philosophy.
What it bans
A behavior plugin: a fixed page of rules you wire into the assistant’s standing instructions, so it loads with every session. No code, no commands, no data, and it never touches the vault. The rules fall into four categories.
Typography that screams machine.
— em-dashFiller that pads instead of saying anything.
"robust"Throat-clearing openers and stock transitions.
"it's worth noting"Patterns and cadences that read as ChatGPT.
"not X, it's Y"Install
Wire the rules at one of two scopes. Project scope reaches this repo only. User scope reaches every imp session in any directory, while plain claude stays untouched. The user-scope add promotes the copy the project add installed, so run the project add first (or pass --from <dir>). The matching remove is the off switch.
imprnt plugin add anti-slop remove imprnt plugin rm anti-slop imprnt plugin add anti-slopimprnt global add anti-slop remove imprnt global rm anti-slop Make the rules your own
To run your own register and banned words on top, copy the rules file into the private _personal/ folder, extend it, and wire that copy instead. The _ prefix keeps it out of the public gallery (and add plugins/_personal/ to .gitignore if you ever git-init the project).