TL;DR: Fill the form, copy the generated prompt into Claude / Gemini / ChatGPT / your local model. The output always includes the docs URL so the AI knows where to fetch fresh API info.
Goal — one sentence describing what the script should do.Required behaviour — bullets the AI must obey (Ctrl/Cmd-click for multi-select).public partial class extending WasmScripting.WasmBehaviour All UI-event targets carry [ExternallyVisible] Cross-client state goes through Networking.SendMessage with byte-tag protocol Receivers validate sender before applying state No reflection, no Activator.CreateInstance No System.IO, System.Net, System.Threading worker threads Use CVR.CVRInput instead of UnityEngine.Input Persistence via WasmScripting.FileStorage (gated by WorldPermissions) [WasmSerialized] only on save/load fields, plain fields for runtime state No emojis in code, comments, or output Comments are single-line; no docstrings or XML doc blocks No VRChat / UdonSharp idioms (no [UdonSynced], no Networking.LocalPlayer.IsUserInVR, no override Interact) Lifecycle/game events use enum-name scan — no [ExternallyVisible] on those Cite the docs page or CCK stub for every API call Before writing, enumerate which bindings + events the script will use and confirm Target tool Claude Gemini ChatGPT Ollama / local Generic / not listed Output Copy
One-time system prompt for your tool (project instructions / custom instructions / Modelfile SYSTEM) → see AI-assisted Authoring .
Per-task user message for each script → generated above.
Bindings reference the AI fetches when unsure → Bindings Surface (versioned, currently CCK Release-WASM.28).CVR-specific surface for the things that aren’t standard Unity → AI-Authoring → CVR-specific binding surface .