@ingin97/pi-dumb-zone
A deterministic Pi extension that warns about suspiciously agreeable assistant responses.
Package details
Install @ingin97/pi-dumb-zone from npm and Pi will load the resources declared by the package manifest.
$ pi install npm:@ingin97/pi-dumb-zone- Package
@ingin97/pi-dumb-zone- Version
0.1.4- Published
- Aug 31, 2026
- Downloads
- 405/mo · 13/wk
- Author
- ingin97
- License
- MIT
- Types
- extension
- Size
- 15 KB
- Dependencies
- 0 dependencies · 1 peer
Pi manifest JSON
{
"extensions": [
"./dumb-zone.ts"
],
"image": "https://raw.githubusercontent.com/ingin97/pi-dumb-zone/master/docs/assets/header.png"
}Security note
Pi packages can execute code and influence agent behavior. Review the source before installing third-party packages.
README
pi-dumb-zone
A minimal Pi extension that raises a red warning when an assistant response contains a targeted strong-agreement phrase such as “you are absolutely right”.
It is intentionally a deterministic lexical heuristic:
- no network calls
- no secondary model calls
- no persistence
- no semantic claims
Inspiration
This extension was inspired by Context Engineering with Dex Horthy, published by The Pragmatic Engineer. Thanks to Dex Horthy for the context-engineering perspective that prompted it.
Install
Install from npm with Pi:
pi install npm:@ingin97/pi-dumb-zone
For a project-local install, use:
pi install npm:@ingin97/pi-dumb-zone -l
Then restart Pi or reload with /reload.
Behavior
The extension appends a green status dot after the context indicator in Pi's default footer. At 100k context tokens, it changes that status to yellow, and shows one warning notification per session. A detected phrase takes priority and switches the status dot to red. Its notification displays the matched phrase, so the trigger is explicit.
Matching is case-insensitive and allows flexible whitespace. It recognizes targeted strong-agreement phrases such as you are absolutely right, you're right to question that, that's actually the better approach, exactly — the key thing is, 100%, I couldn't agree more, and that's an excellent point (including curly apostrophes). Only finalized assistant messages are inspected, so streamed text cannot produce duplicate alerts.
False positives are expected. This extension is an observability joke, not a measure of intelligence, alignment, or response quality.