reactor-skills

Turn a loose scene idea (even a single short sentence, English or Chinese) into a production-ready LingBot-World 2 layered scene: a StructuredExample JSON — base / camera / movement layers with static+dynamic variants, stackable hold-key events (keys 1..9), and per-scene jump/crouch/stand prompts — that drops straight into the frontend's lib/lingbot-cases/, plus a derived seed-image prompt. Can also emit a single composed standalone prompt when the user wants just one prompt. Use whenever the user mentions LingBot 2, LingBot-World 2, lingbot-world-2, reactor/lingbot-world-2, layered scene, StructuredScene, composePrompt, isMoving, base/camera/ movement layers, hold keys, event keys, jumpPrompt/crouchPrompt, lib/lingbot-cases, a "playable character" video prompt, or gives a rough idea they want expanded into an interactive world-model prompt ("make a lingbot 2 scene for a knight in the desert"). Also handles requests phrased in the legacy actions.json / f-g slot export format (see references/legacy-sessions.md

Use whenever a user wants to author or refine a LingBot interactive video world, either a quick PROSE prompt or a full JSON virtual world. Covers all three viewpoints: third-person navigable-subject (driving / riding / chase), first-person traversal, and fixed spectator. Trigger on LingBot, LingBot-World, lingbot-world-fast-v1, "a prompt / base prompt / base + holds / hold-prompts" for a scene, "layered composition", a virtual world / scene file / world JSON, states / transitions / overlays / composeForRender, WASD-navigable scenes, or any "make me a prompt / scene / world for [subject/place]" request, even when the user doesn't say "LingBot" but is clearly describing an interactive video world or a steerable video-generation prompt.

Build real-time video AI applications with Reactor SDKs. Use when code imports `@reactor-team/js-sdk` or `reactor-sdk`; when connecting a web/mobile frontend to a GPU-powered Reactor model (Helios and others); when streaming video via WebRTC with command-based control; when publishing webcam input for video-to-video transformations; when handling JWT authentication for Reactor; or when debugging Reactor connection lifecycle (disconnected → connecting → waiting → ready). SKIP when project uses unrelated WebRTC (e.g., plain video calling), or video pipelines that don't touch Reactor.

Turn rough video-editing intent into precise, production-grade instructions for NVIDIA's SANA-Streaming real-time video-to-video editing model. Use whenever the user wants to edit, restyle, or transform an existing video with SANA-Streaming — swapping outfits, changing backgrounds, removing objects or watermarks, adding tracked overlays, applying an art style, re-rendering a whole scene, or restyling driving/robotics footage. Trigger on any mention of SANA, SANA-Streaming, "video edit prompt," or a request to write or improve a video-editing instruction, even when the user doesn't name the model.

Write effective prompt sequences for Reactor's Helios model, a real-time video frame generation API. Use when creating Helios prompts, video generation prompts, or when the user mentions Helios or Reactor. Covers start prompts, follow-ups, sequence planning, camera anchoring, lighting consistency, and progressive state tracking.