Skip to main content
How to get the most out of prompting GladeKit

​
1. Use a clear prompt structure

Structure your prompt so the assistant knows the situation, the goal, and any constraints:
  • Current state: What exists now (scene, objects, scripts, or “empty scene”).
  • Desired state: What you want (behavior, objects, or changes).
  • Constraints (optional): e.g. “use the selected GameObject,” “don’t create new scripts, only modify existing ones,” “use the New Input System.”

​
2. Be specific about names and targets

  • GameObjects: Use exact names when it matters (e.g. “add a Rigidbody to Player,” “move SpawnPoint to 0, 2, 0”).
  • Scripts: Prefer paths or unique names (e.g. “in Assets/Scripts/PlayerMovement.cs” or “the script named PlayerMovement”).
  • Selection: If you’re referring to what’s selected, say so (e.g. “use the selected prefab,” “apply this to the selected GameObject”). The assistant receives selection from Unity and can use it when you make it explicit.
  • Errors: If you’re reporting a problem, paste the exact error message or a short description (e.g. “NullReferenceException in PlayerMovement.Update()”). Just saying “Fix the errors” without the error text is vague.
Avoid vague phrasing like “the script,” “the object,” or “fix it” when you mean a specific script, object, or error.

​
3. Align with how Glade works

  • One main ask per message: Complex requests are handled in phases (research → create → configure → verify). You can still describe the full goal in one prompt (e.g. “I want a player that moves and jumps on the platform”); the assistant will break it down. For very big tasks, splitting into steps (e.g. “first add movement,” then “now add jump”) can help.
  • Scene vs assets: The assistant can distinguish scene objects (in the hierarchy) from prefabs/assets (in the Project). If you want something in the scene, say “in the scene” or “in the hierarchy.” If you want a prefab created or modified, say “create a prefab” or “update the prefab at Assets/….”
  • Render pipeline and Input System: The assistant is told your project’s render pipeline (URP, HDRP, or Built-in) and input system (New vs Old). You usually don’t need to repeat that unless you’re overriding (e.g. “use the Old Input Manager for this script”).
  • After script creation: The assistant waits for compilation. If you see “script created, waiting for compile,” let it finish before asking for the next change so it can add components or modify the script correctly.

​
4. When the response isn’t what you want

Glade uses the whole thread (your messages and its replies) as context. If it gives a wrong or unwanted response and you reply with “fix it” or “do it again,” it may keep building on that bad response. Better approaches:
  • Improve the original prompt: Edit your last message and add what was missing (e.g. object name, script path, exact error, or “don’t create a new script, only modify X”).
  • Start a new thread with a clean summary: In a new chat, write a short summary of the goal and any important details (e.g. “I need a WASD + jump script on the Character using the New Input System. Last time the script was added to the wrong object.”). Then send the request again. That avoids reinforcing the previous mistake.
Avoid long chains of “no, not that - do this instead” on the same thread when the root cause was an unclear or incomplete first prompt.

​
Example Prompts

The more precisely you describe “what is” and “what you want,” the less the assistant has to guess. Example 1: Building a New Mechanic “I have a scene with a floor and a Character. I want the Character to move with WASD and jump with Space. Create a C# script that does this and add it to the Character. Use the New Input System.” Example 2: Modifying Existing Code “I have a PlayerMovement script on the Player GameObject. Add a sprint that doubles speed while Left Shift is held. Keep the existing movement logic.” Example 3: UI Updates “I have a Canvas with a Button. When the button is clicked, call a method on the GameManager in the scene. Create the script and wire the button.”