10 Mandatory Rules + streaming voice rewrite (WhisperTyping parity) #4013

Originally written by @ccantynz-alt on GitHub.
Imported from https://github.com/ccantynz-alt/48co.v1/pull/24

Summary

10 Mandatory Rules locked into CLAUDE.md

Every session must now: Scan → Detect → Fix → Research → Build → Test → Document. No exceptions.

Content.js COMPLETE REWRITE — streaming voice like WhisperTyping

The fundamental problem: our extension waited until you STOPPED talking to process text. WhisperTyping streams word-by-word AS you speak.

Before: Click → talk → stop → wait → text appears all at once (broken) After: Click → talk → text appears word by word in real-time (like WhisperTyping)

What changed:

• interimResults=true on SpeechRecognition — streams words as you speak
• Each new word IMMEDIATELY inserted into the focused text field
• ZERO popups, overlays, status bars — just text appearing
• 3 insertion methods for ProseMirror editors (Claude.ai, ChatGPT)
• Post-processing (punctuation, caps) only applied AFTER recording stops
• Mouse wheel click (auxclick) as trigger
• Remembers existing text — only replaces what was dictated

https://claude.ai/code/session_01PBeNmPHrD2W6A2YKZRC7V3