Agents Template Repository?

Renoise Tool Development
1

Since around November 2025, LLMs have gotten so good that my full-time job is mostly babysitting Claude now.

For over 30 years, being a (web) programmer meant actually typing code. Now, I set up and maintain a strongly typed template, a test harness, and some reusable UI components. It’s all version-controlled so I can review the diffs, but what used to take a week now takes me a day.

Caveats: The frontier AI plans I use aren’t free, my own skills will inevitably atrophy, and there’s the whole ethical yada yada. But pretending this isn’t happening? I’m witnessing the end of programming in a way someone back in the 1960s punching holes in paper cards for giant mainframes was probably thinking “WHAT ABOUT MY PUNCHCARD JOB?! NOOOOOO!”

Here is some brainstorming on what an agent starter template for Renoise Lua scripts could look like:

API Knowledge: The agent needs to understand the Renoise API. We can easily solve this by pointing it to the renoise/definitions repository.

Automated Testing: The agent needs a reliable way to run its own unit tests.

Linting: Code formatting, type checking, enforced.

Workspace Visibility: The agent needs a way to “see” what’s happening inside Renoise. We could cobble a connection together over TCP, but ideally, we need the tracker equivalent of the Claude Chrome extension, a direct bridge that lets Claude look around inside Renoise natively: LSP (language Server Protocol), ability take screenshots and click on things, etc…

A CLAUDE.md (or AGENTS.md) file. Maybe skills or whatever…

Thoughts?

I know some variation this is already possible, and some are already doing it with their own Renoise scripts…

What I’m going for is a thread about getting ideas on what Renoise needs, built-in, to have a better experience. Then, community efforts could be consolidated into a “Use this template” starter kit on GitHub.

2

I dunno what it all means, but remembered this thread and wonder if it ties in with connecting Renoise to your agents?

For me, after years of bugging the forum with script/tool requests I’m happy to be able to create something from curiosity in Claude. Right now I just direct it to the api info and that seems sufficient?

3

Oh wow, an MCP server, OK then! That fulfills, to a high degree, my ask of being able to “see” Renoise. Thanks for the link.

For web dev, I often take screenshots of what I’m working on, annotate with red arrows, feed it back and prompt with words like “I don’t like this, change it to X.” Claude goes “I agree” and does it, but instead of doing it blindly, it validates by looking in Chrome. AFAIK that MCP project won’t fill that visual gap but for controlling Renoise directly, yahoo.

Right now I just direct it to the api info and that seems sufficient?

Yes, evidently, it’s good enough. The extra tooling I am proposing simply increases the quality control, reliability, and dev tooling around the output. It forces the AI to make less mistakes. It’s the difference between “vibe coding” and “vibe engineering” I guess.

My ideal is: Go to a GitHub repo, click “Use this template” to fork a starter kit, and be in a good spot to work right away.

Aside: Are you using “Claude Desktop” or “Claude Code”? Myself, I use Claude Code. Maybe Claude Desktop can see Renoise, like interact with it?

4

at least on macOS, claude code can see renoise by simply using AppleScript to take a screenshot.

5

Same on windows, you can snip in a picture and ask stuff based on it.

Haven’t gone into deep yet (for it to see Renoise other then the odd picture uploads for giving direction) , was just using the free thing in the browser until yesterday where I now have the most basic subscription. I’ve been using the app on my phone as well at work lol.

Perhaps all the AI discussion and or AI based tools could be moved to a AI tools subforum not to clutter up this one?