For security reason, in frontmatter script we coud define:
Index.json and readme.md will list script files with those data. They will be exposed too GitHub owner repository.
A GitHub repository template will be provided including tools to quick start SB libraries repo.
During import process, It informs user of rights, It coud be interesting to:
Sounds good, but IMHO it should be simple as possible for the user and developer too.
And not too flood either of them with workflows and instructions and steps to follow.
Let’s see what Dr. @Zef comes up with 
Let’s leave him today to work in peace & undisturbed
I have created silverbullet-libraries/index.json at main · malys/silverbullet-libraries · GitHub. It’s the index of my repository.
Tooling provides automatic process based on GitHub actions:
@Mr.Red @ChenZhu-Xie You are using many times LLM to generate scripts.
What do you thing to create a md files in a library with community optimized prompt?
I could be a big plus to perform plugs development
Based on my current experience, it seems that LLMs — particularly Claude Opus, Sonnet 4.0, and GPT-5 Thinking — rank in their familiarity with practical scripting languages roughly as follows: JavaScript ≳ CSS > Lua > Space Lua.
When it comes to CSS, for example, I’ve been working on modifying the responsive behaviour of the TreeView Plug. The LLM can infer part of the DOM structure from existing space-style plug for TreeView, yet it still lacks a big map of the DOM hierarchy governed by TreeView.plug.js. As a result, it’s quite common to F12 open DevTools, use Ctrl + Shift + C to inspect elements, and then manually describe the surrounding DOM context to the LLM.
As for Space Lua, in many cases the LLM still relies on syntax from the 1.x or even 0.x space-script era, rather than incorporating the updated 2.x documentation now available on silverbullet.md. Human intervention (e.g., Ctrl + Shift + F on silverbullet.md) therefore remains necessary — though, identifying and correcting such errors is usually straightforward.
I believe that @Mr.Red and many others possess stronger “hard coding skills” than I do. For me, the LLM functions as a kind of courage buff — it grants me the audacity to gradually familiarize myself with other people’s plugins, with SilverBullet itself, and with the related languages, toolchains, debugging methods, and problem-solving feedback patterns. This process, in a sense, mirrors how the LLM itself learns.
Exploring SilverBullet with the aid of an LLM has taught me a great deal about programming, at least on the applied rather than the foundational level. Coming from a physics background, my coding skills are relatively modest, yet my imagination is vivid; coupled with this newfound boldness, my exploration is steadily becoming broader and deeper.
yeah same here… its not a simple magical prompt to code using LLM, and usually it takes 10+ back and forth until it gets something done, and then it’s only half baked code, you need to then go in deep in the code do some manual modifications, and then go back to the LLM, and Copy/Paste your edited code, ask for some modifications. then it f*ks up all the previous correct code, you then need to go 1-2 versions back when it was still working. so it’s more like a process than justt a prompt.
like Chén-Zhú said, it helps if you give the correct link to the silverbullet.md/API/... you want it to use, or even copy paste an example code for what you want to achieve and tell it to us the function from the example, but do “this” instead.
So sadly there is no secret recipe, and definitely still do some stuff manually so you need at least some basic programing skill, even in another programming language.
I will share my method. It’s never a magic prompt but initial instructions help.
Check silverbullet-libraries/src/LLMInstructions.md at main · malys/silverbullet-libraries · GitHub