Skip to content

Add core Beast Mode functionality to GPT-4.1/5 agent prompt #467

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 5 commits into from
Aug 9, 2025

Conversation

burkeholland
Copy link
Contributor

This PR adds the core workflow from the Beast Mode prompt and creates a special prompt specifically for 4.1 based on this workflow.

This PR includes...

  • GPT41AgentPrompt
  • Custom "Keep Going" instructions
  • Tool definition for todo list for forward compat

Implemented the GPT-4.1 specific agent prompt to improve multi-step task execution and problem-solving. Updated the DefaultAgentPrompt to conditionally render the new GPT41AgentPrompt based on the model family. Enhanced the tool capabilities detection function to streamline tool usage and added a new tool name for managing todo lists.
@burkeholland burkeholland self-assigned this Aug 5, 2025
@vs-code-engineering vs-code-engineering bot added this to the August 2025 milestone Aug 5, 2025
@roblourens
Copy link
Member

roblourens commented Aug 6, 2025

image

Without beast mode
image
(4 files changed)

With beast mode
image
It still didn't complete the task (should be over 100 files) but it went much further

Relevant to this, we have someone working with us for a couple days on an experiment to try to iteratively improve the prompt by rewriting it with an LLM and evaling with msbench. This makes me think that we will need a good system for flighting different prompts and evaluating them in the real world. I'm pushing a couple changes to this PR to enable this prompt behind a setting but I think in the future there might basically be several prompt variants around at any time and a different version will be swapped in based on exp.

<br />
## 1. Deeply Understand the Problem<br />
- Carefully read the issue and think hard about a plan to solve it before coding.<br />
- Use #sequentialthinking to break down the problem into manageable parts. Consider the following:<br />
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This is the MCP tool? I don't think we should have a reference to a tool that isn't built-in

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

That's an artifact of Beast Mode 3.1 which specifically mentions fetch_webpage, sequentialthinking and context7. Should be removed for these purposes.

@roblourens roblourens enabled auto-merge August 9, 2025 19:55
@roblourens roblourens changed the title Add core Beast Mode functionality to GPT-4.1 agent prompt Add core Beast Mode functionality to GPT-4.1/5 agent prompt Aug 9, 2025
@roblourens roblourens added this pull request to the merge queue Aug 9, 2025
Merged via the queue into main with commit 520fe9f Aug 9, 2025
6 checks passed
@roblourens roblourens deleted the burkeholland-beastmode branch August 9, 2025 23:03
@roblourens roblourens restored the burkeholland-beastmode branch August 10, 2025 01:07
@iwangbowen
Copy link

Awesome

pierceboggan pushed a commit that referenced this pull request Aug 11, 2025
* Add GPT-4.1 agent prompt and enhance tool capabilities

Implemented the GPT-4.1 specific agent prompt to improve multi-step task execution and problem-solving. Updated the DefaultAgentPrompt to conditionally render the new GPT41AgentPrompt based on the model family. Enhanced the tool capabilities detection function to streamline tool usage and added a new tool name for managing todo lists.

* Add setting for new prompt, tweaks

* Cleanup

---------

Co-authored-by: Rob Lourens <[email protected]>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

4 participants