[FEATURE] Voice dictation tap mode: allow starting with text in input, and stop recording without auto-sending
Feature request
Voice dictation tap mode currently has two hardcoded behaviors that make it awkward to use for composing prompts. Requesting two changes (or settings to opt into them):
1. Allow starting recording when the input box already contains text
Current behavior: In tap mode, tapping the dictation key only starts recording when the prompt input is empty.
Requested: Allow the tap to start recording even when text is already typed, inserting the transcript at the cursor position (the same way hold mode already does). Dictation is often used to append to a partially typed prompt.
2. Option for the second tap to stop recording without auto-sending
Current behavior: In tap mode, the second tap stops recording and immediately submits the prompt whenever the transcript is at least 3 words long.
Requested: An option (e.g. extend the existing voice.autoSubmit setting to tap mode, defaulting to insert-only) so the second tap just stops recording and leaves the transcript in the input box for review/editing, with Enter being the only way to submit. Auto-sending gives no chance to fix transcription errors before the prompt goes to the model.
Workaround considered
Hold mode already provides both behaviors (works with existing text, release inserts without sending), but push-to-talk ergonomics aren't a substitute for tap-toggle: holding a key for the length of a long dictation is uncomfortable, and bare-letter keys in hold mode depend on key-repeat with a warmup delay.
Environment
- Claude Code CLI (TUI), Windows 11 Pro (10.0.22631), PowerShell
- Voice mode: tap, dictation language: en
🤖 Generated with Claude Code