Roleplay workspace 1.19.4

Starting a new story now creates one transcript. The previous flow could create a default transcript while selecting the character or cast, then create the requested story.

• Voice and image generation default to off, with separate controls in Settings. Choices persist, voice playback stops when disabled, and disabled image generation blocks commands and automatic triggers.
• Desktop controls and spacing are more compact. Phones retain 44px touch targets. Dropdowns and settings headers use flat dark surfaces.
• Upstream GitHub App bot jobs skip on forks. This fixes the missing-credentials failures that caused notification emails.

Verified: ESLint, 411 unit tests, and 16 browser checks, including single-story creation, tool defaults and persistence, archive/delete behavior, and 320/390/768px layouts. Model, image, and voice integration checks use controlled fixtures.

Homelab image: ghcr.io/nulifyer/sillytavern:v1.19.4, or :latest for stable releases. The release workflow builds and starts native AMD64 and ARM64 images before publishing. Keep existing data and configuration volumes.