Beast Mode 3.1: AI-Powered VS Code Chat Agent

The Problem

GPT-4.1 has two critical weaknesses in its default coding behavior:

1. Lack of agency - The model discusses actions without executing them
2. Lack of accuracy - It works too quickly, skipping crucial thinking steps

The Solution

Beast Mode is a carefully engineered system prompt that addresses these issues through:

• Todo lists - Forces continuous status updates, improving task completion
• Sequential thinking prompts - Questions like "What are the edge cases?" encourage deeper analysis
• Web research requirements - Leverages VS Code's fetch tool to gather current documentation

Key Features in v3.1

Memory System

Users can instruct the model to remember project-specific rules, saved to .github/instructions/memory.instructions.md. Similar to ChatGPT's memory feature but scoped to projects.

File Management

Reduces redundant file reading by providing explicit guidance on when re-reading is necessary.

Git Control

Prevents autonomous staging/committing without explicit user instruction—a safety guardrail against unwanted changes.

Installation

Retrieve the prompt from Holland's GitHub gist, then create a custom chat mode in VS Code's agent dropdown. Select "User Data Folder" for global access across projects.

Connections

• opus-4-5-is-going-to-change-everything - Same author's broader thesis on AI coding. Beast Mode represents the practical tooling that enables the AI-first workflow he advocates.
• writing-a-good-claude-md - Both address the same challenge: crafting effective instruction files for AI coding agents. Beast Mode's memory system mirrors CLAUDE.md's role as persistent project context.
• context-engineering-guide-vscode - Microsoft's official guide to VS Code context engineering. Beast Mode's .github/instructions/ approach aligns with the .github/copilot-instructions.md pattern described here.