Using This Documentation with AI Assistants
If you use an AI coding assistant (such as Cursor, GitHub Copilot, Claude, ChatGPT, or Antigravity) while developing mods for Shape of Dreams, this page explains how to get the most accurate code without polluting your context window.
1. The Core Problem with API Documentation
Shape of Dreams contains 897 types and thousands of methods in a single assembly.
If you try to paste whole web pages or entire folders of documentation into an LLM:
- You burn tens of thousands of tokens per prompt.
- The model suffers from "attention degradation" (forgetting details in the middle).
- The model often hallucinates method names that exist in standard Unity or other games rather than Shape of Dreams.
2. Recommended Workflows
Option A: Querying the SQLite Database (api.db) — Best for Coding Agents
For tool-assisted agents (Cursor with terminal/agent mode, Antigravity, Cline):
- Download
api.dbandLLM_GUIDE.mdfrom the docs-data repository. - Place them in your mod workspace.
- Prompt your assistant:
"Check
LLM_GUIDE.mdand queryapi.dbusing sqlite3 to find the exact method signatures and parameter types before writing patches." - The assistant executes pinpoint SQL queries that cost under 50 tokens each and returns 100% verified C# signatures.
Option B: Web & Chat-Based Assistants (ChatGPT, Claude Web)
If you are chatting with an AI in a browser:
- Navigate to the class you need in this portal (e.g.
AbilityTriggerorActor). - Copy only the specific method or field declaration you intend to hook or modify.
- Provide that targeted signature in your prompt:
"Here is the exact method signature from Shape of Dreams:
public virtual bool CanBeCast(). Write a HarmonyPrefix patch in BepInEx 5 that checks my custom condition before calling this method."
3. Need the Source Data?
All raw JSON models, the knowledge graph, and the SQLite database are open and maintained in the community repository: 👉 DocsHub-SoD/docs-data on GitHub