Clawdy v0.0.2

Added

• Audio-synced pointing. When an answer names several places, the claw now visits each in turn (up to ~7), showing each place's name as it arrives. With an ElevenLabs key, each move is timed to the spoken audio and leads it by a beat so you see the target just before you hear it; on Apple TTS the claw visits them in order (untimed).
• Draggable overlays. Drag the research toast / idle mini-toast by its body to move it out of the way of whatever's behind it; the position persists across launches and is re-clamped to the current display.
• Overlays visible in screen recordings. Clawdy's claw cursor and on-screen annotations now appear in QuickTime/OBS by default (great for demos and tutorials), while still being excluded from the screenshots Clawdy sends to the model.

Fixed

• Legible input placeholders. Text-field placeholder text was dark-on-dark; it now uses a readable muted tone (with VoiceOver labels preserved).
• Faster, cleaner research. Removed a redundant per-image WebFetch pre-check that caused frequent HTTP 400s and slowed research; broken images are swapped for inline placeholders instead — on both the Claude and Codex research paths.