Delete the two byte-identical local formatBytes copies from the chat-attachment
modules and repoint all call sites at the canonical lib/utils/format-bytes.ts,
which its own doc comment asks new code to use.
The copies had no upper clamp (>= 1 TB rendered "1 undefined") and no NaN or
negative guard (NaN rendered "NaN undefined"). Size labels now carry one
decimal ("10.0 MB" instead of "10 MB") because the canonical helper does not
strip a trailing zero; unit and e2e expectations updated to match.
* feat(web): add voice mode to task create dialog prompt
Wires the existing VoiceInputButton (introduced in #1159 for the chat
composer) into the task-create dialog's prompt toolbar so users can
dictate the initial task description.
- TaskFormInputs gains an internal insertAtCursor that splices the
transcript at the textarea caret with a single leading space when
it follows a non-whitespace character, matching the chat composer's
behaviour. A pending-cursor ref + useLayoutEffect restores the
caret position after React commits the new value.
- An optional onVoiceAutoSend prop is threaded from TaskCreateDialog
down through DialogPromptSection -> TaskFormInputs -> VoiceInputButton.
The dialog points it at formRef.current?.requestSubmit() so dictation
can create the task hands-free; every existing submit gate still
applies via the disabled submit button.
- Voice mode is gated globally on userSettings.voiceMode.enabled, so
the button self-hides for users who turned it off in settings and
self-greys on browsers without a usable engine.
* fix(web): invoke submit handler directly for voice auto-send
`HTMLFormElement.requestSubmit()` is missing on Safari < 16, so the
previous wiring silently failed with a TypeError on browsers where voice
mode itself works (Safari 14.5-15 supports Web Speech / Whisper Web).
Drop the form ref and call `setup.guardedHandleSubmit` with a synthetic
event instead. `guardedHandleSubmit` only reads `preventDefault` off the
event, and every validation gate (`validateForCreate`, session-mode
description/agent checks, edit-mode title check, `submitBlockedReason`)
lives inside the submit handler — so dictation with incomplete fields
still no-ops, identical to the prior requestSubmit-based behaviour.
This also matches the chat composer's pattern of calling the same submit
function the send button uses.
independent-review: voice auto-send Safari < 16 compat
* test(web): expand voice splice cases for task-create dialog
Adds adversarial coverage for the at-cursor splice helper that backs
voice-mode dictation in the task-create dialog:
- multi-line description with caret on a line boundary
- transcript with internal newlines preserved as-is
- existing tabs / newlines before the caret treated as whitespace
- session-mode renders the voice button
Splits the test file into a rendering-and-wiring `describe` and a
splice-behaviour `describe` so each arrow function stays under the
100-line `max-lines-per-function` lint cap.
qa: adversarial voice splice cases + session-mode parity
---------
Co-authored-by: Kandev Agent <agent@kandev.dev>