Migrates the boot.js mic-button behavior change from reverted fork commit
014b9eef into plugin/static/app.js. Replaces vanilla dictation-to-textbox
mic flow with voice-message mode: tap to record, tap again to stop → audio
File attached + sent automatically. Server-side audio_attachment_processor
in routes/transcribe.py transcribes for text-only models.
Implementation strategy:
- Own #btnMic onclick (boot.js sets it during init; plugin overrides via
MutationObserver after DOM mutations + idempotent dataset.svrntyVm flag)
- MediaRecorder with same mime-type fallbacks as the original
(webm/opus → webm → ogg/opus → mp4)
- DataTransfer to set #fileInput.files programmatically (cross-browser path)
- Dispatch 'change' so boot.js's fileInput.onchange → addFiles() runs
- Click #btnSend after 50ms tick so attachment registers before submit
Filename convention: voice-message-${timestamp}.{webm|ogg|mp4} — matches the
audio processor's name-prefix detection in _transcribe_audio_attachments.
Tests: 6 new JS-static checks in tests/unit/test_app_js.py (idempotent
guard, voice-message prefix, DataTransfer pattern, vault URL pin, mic
override). 26 tests total, all PASS.
Connection map: now 9 public API · 0 forced internal · 1 frontend.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
static/app.js now injects the "Vault connections" panel into Settings → System
via DOM mutation observer + populates from GET /api/vault/status (already
served by routes/vault_status.py). Mirrors the original behavior from
hermes-webui fork commit 3e2c74f3 without needing any upstream HTML/JS edit.
routes/transcribe.py documents the STT migration design choice (streaming_hook
public-API method vs forced-internal entries) — implementation lands as a
follow-on. Plugin still loads cleanly without STT (route disabled in
_phase2_routes() until Phase 2.1).
CONNECTION-MAP.md regenerated: still 0 forced internals.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>