feat(desktop): pre-capture HUD with audio level meters + webcam preview + presenter recording

[huntharo]

Summary

Add a richer pre-capture HUD that lets the user see what's about to be captured — system-audio level (RMS), microphone level (RMS), and a live webcam preview — before committing to record. Then record the webcam stream alongside the screen as a separate, repositionable picture-in-picture track.

This is the umbrella issue. The deeper data-model work (multi-channel video tracks, per-source audio tracks) and the in-flight UX (on-screen controls + capture confirmation) are likely follow-up issues; see Notes below.

Motivation

Today the user clicks record and hopes everything's wired up. Common failure modes that bite after the recording is done and unrecoverable:

• Mic muted at the OS level; ends up with a silent voice-over.
• Wrong input device selected (e.g. Bluetooth headset picked instead of the built-in mic).
• System audio toggle was off and the user didn't notice.
• No webcam to overlay because the picker didn't match.

A pre-capture monitoring HUD turns all of these into pre-flight problems instead of post-recording disappointments.

Scope

Pre-capture HUD additions

• System-audio RMS meter — live levels from the SCStream system-audio tap (if enabled in capabilities). Renders as a familiar VU/peak bar.
• Microphone RMS meter — live levels from the selected AVCaptureDevice. Includes a device picker (built-in vs. external mic vs. "none").
• Webcam preview — a small live thumbnail of the AVCaptureDevice video stream the user selected, so they can confirm framing/lighting before recording.
• Toggle switches for each: system audio on/off, mic on/off, webcam on/off. Persisted via Settings.

Webcam recording (multi-channel video)

• Record the webcam stream as a SECOND video track in the output container, NOT pre-composited into the screen frame.
• Independent control of position (e.g. corners), size, opacity, and visibility (show/hide) — both pre-record (defaults) and at edit time.
• Container: MP4 supports multi-track video. Editor needs to know about the second track and let the user move/hide it per-segment.

Required platform plumbing

• Mic + webcam permission handling (extends the existing recording-permissions.ts flow with a webcam permission).
• A device-enumeration command-bus handler ("list audio inputs", "list video inputs") so Settings + the pre-capture HUD share one source of truth.
• Sampling pipeline that taps the raw audio buffers for RMS WITHOUT going through the recorder's AVAssetWriter (the recorder is for the final file; the meters are pure UI).

Notes — likely splits

• Multi-channel video + per-source audio tracks — the AVAssetWriter setup for a second video track + storing system-audio and mic as separate tracks (rather than mixed into one) is detailed enough that it probably becomes its own issue once we start. Editor support also changes.
• In-capture on-screen controls + confirmation — separate issue: see feat(desktop): in-recording on-screen controls + capture-is-working confirmation #72 (once filed).

Out of scope

• AI-driven framing / virtual greenscreen / background blur for the webcam (separate, very different problem).
• Live cloud upload during capture.

🤖 Generated with Claude Code