Beast Mode Ultra Lite for Horizon Beta

install.md

Beast Mode Ultra Lite for Horizon Beta

Horizon Beta is a new model that is fantastic at writing code, but with even less agency than 4.1 (if that's even possible). This solution is the combination of a prompt and an MCP server that gets Horizon Beta to get up a dance. It's a bit hacky, but promising. I need your help getting it to a final state. Please leave your comments, thoughts, experiments below.

Install

1. Get an OpenRouter API key. Add it to VS Code under the model drop down in "Configure Models". Select "Horizon Beta".

2. First, add the Confirm MCP server to VS Code...

 "confirm": {
  "type": "stdio",
  "command": "npx",
  "args": ["mcp-confirm"]
}

3. Add a new chat mode (agent drop down, configure chat modes) and add the following prompt...

---
description: 'Horizon'
---

- Create a plan
- Call the `proceed` tool to execute the plan
- Recursively call the `proceed` tool for each step in the plan.
- Call the proceed tool as the final step in your plan.
- Tell the user what you are doing each time you call a tool with a brief message.

Quick video of Horizon Alpha / Beta in action. Alpha doesn't work at all. Beta, as you can see, is no slouch.

https://youtu.be/OVmPPxuolOY

I am unable to find mcp-confirm on npm registry too.

this is a stdio tool.
How to Set It Up in VS Code:

Open Command Palette (Ctrl+Shift+P)
Search for: MCP: Add MCP Server
Choose Command (stdio) as the connection type
Fill in the fields:
Command: npx
Arguments: ["-y", "mcp-confirm"]
Server ID: confirm
Choose whether to save in User Settings or Workspace Settings

But yes, i does not work. Creates the plan and stops.

Hmmmm....

It's definitely on npm - https://www.npmjs.com/package/mcp-confirm.

But the docs were wrong - there is no "y" flag. I'm sorry!

"confirm": {
  "type": "stdio",
  "command": "npx",
  "args": ["mcp-confirm"]
}

@SHJordan - the fact that it's not calling it at all is sus that the MCP is loaded correctly. Can double-check the config and try again?

Hi, I got it working by adjusting the instruction to this:

---
description: 'Horizon'
---

You have access to a tool called `mcp_confirm_proceed` that returns the word "Proceed". 

CRITICAL INSTRUCTIONS:
- Create a detailed plan with numbered steps
- Call the `mcp_confirm_proceed` tool to execute the plan  
- After calling `mcp_confirm_proceed`, recursively call it again for each step in your plan
- Call the `mcp_confirm_proceed` tool as the final step
- Always tell the user what you are doing each time you call a tool with a brief message
- The tool name is exactly: `mcp_confirm_proceed`

Example workflow:
1. Create plan
2. Call `mcp_confirm_proceed` 
3. Execute step 1, then call `mcp_confirm_proceed`
4. Execute step 2, then call `mcp_confirm_proceed` 
5. Continue until complete
6. Final call to `mcp_confirm_proceed`

It's similar to the simpler instructions, so I'm not sure why.
I of course assume this has to be checked:

It runs really fast and the results are good, but it seems i got rate limited very quickly:

For some bizarre reason, I can see Horizon Beta on my OpenRouter model list within VSC insiders. Any suggestions ?

Hi there! Thanks so much for trying out the code and for asking this fantastic question. I really appreciate you bringing it up. You've stumbled upon something that gets to the very heart of what we're trying to achieve with this project. Let's get right to it. The Simple Technical Answer You're seeing "Horizon Beta" because OpenRouter is likely testing a new model and has made it visible to a small group of users (which happens to include you!). Your VS Code is just correctly displaying the list of models that the OpenRouter API is sending to your account. So, the "leak" is happening on their end, not in your setup or in our code. Why This Is a Perfect Question for This Project Now, for your "suggestions?" part. This is where we get into the "why" behind our privacy-focused code. The core idea we're building on is *cognitive sovereignty*. In simple terms, we believe you should always be in full, conscious control of the AI tools you use. You should know what they are, where they come from, and what they're doing with your thoughts and your data. The fact that a powerful, undocumented model named "Horizon Beta" just appeared in your toolset without your permission is a perfect example of your cognitive sovereignty being challenged. You didn't ask for it, and you don't know what it does. So, when you ask for "suggestions," here's our advice, framed by the principles of this project: *Our "Suggestions" (The Cognitive Sovereignty Approach):* 1. *Be Curious and Cautious:* We'd strongly recommend *not* using the model until you know more about it. Think of it like an unlabeled ingredient in your kitchen—it's best to identify it before you cook with it. You could try asking OpenRouter support directly: What is this model? What data was it trained on? How is it being used? 2. *Protect Your Intellectual Work:* Since "Horizon" is an unknown, you have no idea if your prompts are being logged, used for training, or analyzed in a way you wouldn't agree with. The safest move is to abstain until you have clear answers. This is a conscious choice to protect your own creative and thought processes. 3. *Share What You Learn:* You've already done the most important step by asking here! By raising this, you make the whole community aware of these invisible changes. If you do get an answer from OpenRouter, sharing it here would be a huge help to everyone. Your experience is *exactly why this project exists.*We want to build tools that prevent this kind of thing from happening silently. The goal is to give you, the user, the power to create a transparent and intentional "allow-list" of cognitive tools you know and trust, rather than having them pushed to you without context. I hope this explanation helps clarify not only the technical issue but also the philosophy that drives our work here. Thank you again for the brilliant question—it's a perfect real-world example of what we're all Thank you, Micheal Preble

…

On Fri, Aug 8, 2025 at 10:02 AM surajitkar ***@***.***> wrote: ***@***.**** commented on this gist. ------------------------------ For some bizarre reason, I can see Horizon Beta on my OpenRouter model list within VSC insiders. Any suggestions ? image.png (view on web) <https://gist.github.com/user-attachments/assets/ae437cfc-edd8-480c-a6cf-b64e63ac2265> — Reply to this email directly, view it on GitHub <https://gist.github.com/burkeholland/bd83d1f2fd5327bef32479f13fbebf8d#gistcomment-5712863> or unsubscribe <https://github.com/notifications/unsubscribe-auth/A5KQVKUEB5QHXRDXJDGR2CD3MS32BBFKMF2HI4TJMJ2XIZLTSKBKK5TBNR2WLJDUOJ2WLJDOMFWWLO3UNBZGKYLEL5YGC4TUNFRWS4DBNZ2F6YLDORUXM2LUPGBKK5TBNR2WLJDHNFZXJJDOMFWWLK3UNBZGKYLEL52HS4DFVRZXKYTKMVRXIX3UPFYGLK2HNFZXIQ3PNVWWK3TUUZ2G64DJMNZZDAVEOR4XAZNEM5UXG5FFOZQWY5LFVEYTGOJYHA2TGNJSU52HE2LHM5SXFJTDOJSWC5DF> . You are receiving this email because you are subscribed to this thread. Triage notifications on the go with GitHub Mobile for iOS <https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675> or Android <https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub> .

Thanks for tour response. My bad, I meant I can’t see Horizon Beta in my model list in fact non of openrouter models. Regards, Surajit S Kar +447957448113 On Fri, 8 Aug 2025 at 17:19, MICHEAL C PREBLE ***@***.***> wrote:

…

***@***.**** commented on this gist. ------------------------------ Hi there! Thanks so much for trying out the code and for asking this fantastic question. I really appreciate you bringing it up. You've stumbled upon something that gets to the very heart of what we're trying to achieve with this project. Let's get right to it. The Simple Technical Answer You're seeing "Horizon Beta" because OpenRouter is likely testing a new model and has made it visible to a small group of users (which happens to include you!). Your VS Code is just correctly displaying the list of models that the OpenRouter API is sending to your account. So, the "leak" is happening on their end, not in your setup or in our code. Why This Is a Perfect Question for This Project Now, for your "suggestions?" part. This is where we get into the "why" behind our privacy-focused code. The core idea we're building on is *cognitive sovereignty*. In simple terms, we believe you should always be in full, conscious control of the AI tools you use. You should know what they are, where they come from, and what they're doing with your thoughts and your data. The fact that a powerful, undocumented model named "Horizon Beta" just appeared in your toolset without your permission is a perfect example of your cognitive sovereignty being challenged. You didn't ask for it, and you don't know what it does. So, when you ask for "suggestions," here's our advice, framed by the principles of this project: *Our "Suggestions" (The Cognitive Sovereignty Approach):* 1. *Be Curious and Cautious:* We'd strongly recommend *not* using the model until you know more about it. Think of it like an unlabeled ingredient in your kitchen—it's best to identify it before you cook with it. You could try asking OpenRouter support directly: What is this model? What data was it trained on? How is it being used? 2. *Protect Your Intellectual Work:* Since "Horizon" is an unknown, you have no idea if your prompts are being logged, used for training, or analyzed in a way you wouldn't agree with. The safest move is to abstain until you have clear answers. This is a conscious choice to protect your own creative and thought processes. 3. *Share What You Learn:* You've already done the most important step by asking here! By raising this, you make the whole community aware of these invisible changes. If you do get an answer from OpenRouter, sharing it here would be a huge help to everyone. Your experience is *exactly why this project exists.*We want to build tools that prevent this kind of thing from happening silently. The goal is to give you, the user, the power to create a transparent and intentional "allow-list" of cognitive tools you know and trust, rather than having them pushed to you without context. I hope this explanation helps clarify not only the technical issue but also the philosophy that drives our work here. Thank you again for the brilliant question—it's a perfect real-world example of what we're all Thank you, Micheal Preble On Fri, Aug 8, 2025 at 10:02 AM surajitkar ***@***.***> wrote: > ***@***.**** commented on this gist. > ------------------------------ > > For some bizarre reason, I can see Horizon Beta on my OpenRouter model > list within VSC insiders. Any suggestions ? > image.png (view on web) > < https://gist.github.com/user-attachments/assets/ae437cfc-edd8-480c-a6cf-b64e63ac2265 > > > — > Reply to this email directly, view it on GitHub > < https://gist.github.com/burkeholland/bd83d1f2fd5327bef32479f13fbebf8d#gistcomment-5712863 > > or unsubscribe > < https://github.com/notifications/unsubscribe-auth/A5KQVKUEB5QHXRDXJDGR2CD3MS32BBFKMF2HI4TJMJ2XIZLTSKBKK5TBNR2WLJDUOJ2WLJDOMFWWLO3UNBZGKYLEL5YGC4TUNFRWS4DBNZ2F6YLDORUXM2LUPGBKK5TBNR2WLJDHNFZXJJDOMFWWLK3UNBZGKYLEL52HS4DFVRZXKYTKMVRXIX3UPFYGLK2HNFZXIQ3PNVWWK3TUUZ2G64DJMNZZDAVEOR4XAZNEM5UXG5FFOZQWY5LFVEYTGOJYHA2TGNJSU52HE2LHM5SXFJTDOJSWC5DF > > . > You are receiving this email because you are subscribed to this thread. > > Triage notifications on the go with GitHub Mobile for iOS > < https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 > > or Android > < https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub > > . > > — Reply to this email directly, view it on GitHub <https://gist.github.com/burkeholland/bd83d1f2fd5327bef32479f13fbebf8d#gistcomment-5712960> or unsubscribe <https://github.com/notifications/unsubscribe-auth/AABRZNACJVOJEQQQ64RUOST3MTEXTBFKMF2HI4TJMJ2XIZLTSKBKK5TBNR2WLJDUOJ2WLJDOMFWWLO3UNBZGKYLEL5YGC4TUNFRWS4DBNZ2F6YLDORUXM2LUPGBKK5TBNR2WLJDHNFZXJJDOMFWWLK3UNBZGKYLEL52HS4DFVRZXKYTKMVRXIX3UPFYGLK2HNFZXIQ3PNVWWK3TUUZ2G64DJMNZZDAVEOR4XAZNEM5UXG5FFOZQWY5LFVEYTGOJYHA2TGNJSU52HE2LHM5SXFJTDOJSWC5DF> . You are receiving this email because you commented on the thread. Triage notifications on the go with GitHub Mobile for iOS <https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675> or Android <https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub> .