Microaggression Identifier — Chrome extension

A Manifest V3 Chrome extension that warns you, in real time, when a message you are about to send may contain a microaggression, insult, or hate speech. It is the front end for the model in the microaggressionClassifier project, which serves predictions over a small local Flask API.

How the two projects fit together

 you type in X / Facebook / Instagram / Messenger / Discord
        │
        ▼
 content.js  ──CLASSIFY──▶  background.js  ──POST /print──▶  microaggressionClassifier
 (per-site UI)             (service worker)                  (RoBERTa Flask server, :5000)
        ▲                                                          │
        └──────────────── label + score ◀─────────────────────────┘

The classifier returns one of four labels (see microaggressionClassifier/RESULTS.md): general_insult, overt_microaggression, subtle_microaggression, hate_speech. extension/severity.js is the single place that maps those labels to colors and warning messages.

Layout

extension/
  manifest.json   MV3 manifest
  background.js    service worker — the only thing that talks to the backend
  content.js       one config-driven script for all five sites (PLATFORMS table)
  severity.js      shared label → severity mapping (used by content + popup)
  popup.html/css/js  manual checker + per-site settings + dark mode
  icons/logo.png
main.crx           packed + signed extension
main.pem           signing key (gitignored — keep a private backup)

Per-site modes

Set per site in the popup; stored in chrome.storage.local as <site>Setting:

Mode	Behavior
0	No identifying — do nothing
1	Warning text — color the draft by severity
def	Warning text and confirm before sending
2	Block sending — confirm before sending

Run it

1. Start the classifier backend:

   cd ../microaggressionClassifier
   pip install -r requirements.txt
   python TerminalHeroku.py        # serves http://127.0.0.1:5000/print

2. Load the extension — either:
   • Unpacked (for development): chrome://extensions → enable Developer mode → Load unpacked → select the extension/ folder, or
   • Packed: drag main.crx onto chrome://extensions.

Re-packing main.crx

Keep main.pem so the extension keeps the same ID across rebuilds:

chrome.exe --pack-extension="extension" --pack-extension-key="main.pem"
# produces extension.crx → rename to main.crx

Known limitation

The model has no "harmless" class — every message is scored against the four categories above. severity.js applies a confidence threshold (CONFIDENCE_THRESHOLD) so low-confidence predictions stay quiet, but the highest-leverage improvement is to add a neutral/harmless class to the training data (see microaggressionClassifier/RESULTS.md).