Copy/Paste the contents of the training file (don’t just attach the file, but open it in a text editor and copy the contents to the prompt, attaching a file behaves different in the context window of the chatbot)
Run the prompt
Gemini will respond “I’m Ready”
Describe the script or automation you need
Copy the resulting script to Excalidraw/Scripts/Your New Script.md or to where ever you’ve defined your SCRIPT folder, and run the script in Excalidraw using the Obsidian Command Panel (CTRL/CMD+P and type in the title of your script)
since publishing the video, the training file has grown quite a bit, right? When pasting it in AI Studio it shows me 500k tokens used right away. With two prompts and a short script I already seem to reach the overall token limit - even with Gemini 3. Do you have any idea how to work around that issue? Can you think of parts of the library that are not essential to the training and could be left out?
The single biggest chunk of code that could be deleted from the training file is the MindMap Builder Source code. You can manually remove that from the training file, though then you lose the example implementation for using the new side panel API.
To remove MindMap Builder open a text editor and delete the section between
<!-- Source: ea-scripts/Mindmap Builder.js -->
and
<!-- Source: ea-scripts/Mindmap connector.md -->
If you find the time, I’d welcome collaboration on the training file generation script. Should be converted into a collection of skills for better agent efficiency.
Okay, I’m going to try with the reduced file first. When the proof of concept is done, I’ll think about adding the sidepanel – would definitely make sense for scripts like the layer manager.
Also, I am going to check Google AI Studio again. I looked like it cut the output at about 513k tokens but that could’ve been an error rather than a general limitation.
I will have a look at the generating script and let you know in case a clever thought crosses my mind