AI-Berkshire Agent Setup Documentation: Complete Installation Guide

AI-Berkshire agent setup documentation is located in the repository's README.md and automated installer scripts, supporting both Claude Code commands and OpenAI Codex skills through a language-agnostic workflow. The xbtlin/ai-berkshire repository provides platform-specific shell scripts that handle the entire installation process, from CLI prerequisites to skill synchronization.

Prerequisites and Installation Options

AI-Berkshire supports two primary LLM platforms. Choose the installation path that matches your preferred coding assistant.

Claude Code CLI Installation

Before running the AI-Berkshire setup, install the Anthropic CLI globally:

npm install -g @anthropic-ai/claude-code

This command installs the claude binary required to execute the repository's custom commands.

OpenAI Codex CLI Installation

For Codex users, install the CLI using one of three methods supported by the documentation:

# Option 1: curl script

curl -fsSL https://chatgpt.com/codex/install.sh | sh

# Option 2: npm

npm install -g @openai/codex

# Option 3: Homebrew

brew install openai-codex

Verify installation with codex --version before proceeding to the repository setup.

Step-by-Step AI-Berkshire Agent Setup

Once the base CLI is installed, configure the agent capabilities through the repository's automation scripts.

Clone the Repository

Clone the canonical source and navigate to the project root:

git clone https://github.com/xbtlin/ai-berkshire.git
cd ai-berkshire

The skills/*.md files in this directory serve as the canonical source definitions for all agent behaviors.

Run the Platform-Specific Installer

Execute the appropriate installer script based on your CLI choice:

For Claude Code users:

./scripts/install-claude-commands.sh

This script copies the Skill Markdown files into the global Claude Code commands directory, making them available as slash commands.

For Codex users:

./scripts/install-codex-skills.sh

Optionally, add slash-prompt compatibility:

./scripts/install-codex-prompts.sh

The scripts/install-codex-skills.sh generates and installs the Codex skill packages, while scripts/install-codex-prompts.sh creates wrappers for slash-prompt compatibility.

Synchronize Generated Artifacts

After editing any skills/*.md file or pulling updates, regenerate the platform-specific artifacts:

# Refresh Codex skill packages

python3 scripts/sync-codex-skills.py

# Refresh slash-prompt wrappers (if using optional compatibility layer)

python3 scripts/sync-codex-prompts.py

These Python scripts treat the Markdown skill definitions as the single source of truth, ensuring consistency across both platforms.

Verifying Your Installation

Confirm the setup completed successfully:

• Claude Code: Run claude --version to verify the CLI version, then list available commands to see the installed skills.
• Codex: Run codex --version, then restart the Codex application to load the newly installed skills into the prompt context.

Both platforms should now recognize the 18+ research skills defined in the skills/ directory.

Running Your First Agent

Invoke agents using platform-specific syntax after completing the AI-Berkshire agent setup.

Claude Code syntax:

/investment-research 腾讯
/investment-team 美团
/dyp-ask "拼多多的护城河到底在哪里？"

Codex syntax:

After restarting the application, type the skill name directly:

investment-research 腾讯
earnings-review PDD 2025Q4
news-pulse 腾讯

Each command automatically spins up four parallel sub-agents that perform distinct analyses—business model, valuation, moat, and risk—then aggregate the results into a unified response.

Understanding the Agent Architecture

The underlying architecture is documented in assets/architecture.mmd, a Mermaid diagram illustrating how the coordinator agent dispatches parallel sub-agents. When you invoke a skill, the system leverages tools/financial_rigor.py for financial validation utilities that ensure consistent, quantitative analysis across all sub-agents.

This multi-agent design, implemented in the xbtlin/ai-berkshire source code, ensures comprehensive coverage of investment research dimensions without manual prompt engineering.

Summary

• AI-Berkshire agent setup documentation resides primarily in README.md and the scripts/ directory.
• Installers at scripts/install-claude-commands.sh and scripts/install-codex-skills.sh automate platform-specific setup.
• Canonical skill definitions live in skills/*.md and are synchronized via scripts/sync-codex-skills.py.
• The architecture uses four parallel sub-agents (business, valuation, moat, risk) coordinated through a multi-agent system.
• Verification requires CLI version checks and a restart to load newly installed skills.

Frequently Asked Questions

Where is the official AI-Berkshire agent setup documentation located?

The official documentation is located in the repository's README.md file at the root of the xbtlin/ai-berkshire repository. Additional implementation details are found in the shell scripts within the scripts/ directory, including install-claude-commands.sh and install-codex-skills.sh, which contain the executable setup instructions.

Can I use AI-Berkshire with both Claude Code and OpenAI Codex simultaneously?

Yes. The repository supports both platforms through separate installer scripts. Run ./scripts/install-claude-commands.sh for Claude Code integration and ./scripts/install-codex-skills.sh for Codex. The skills/*.md files serve as the single source of truth, so you can maintain both installations and synchronize them using the respective Python sync scripts.

What happens if I modify a skill definition after installation?

After editing any file in skills/*.md, you must regenerate the platform artifacts. Run python3 scripts/sync-codex-skills.py to refresh Codex skill packages, and for Codex slash-prompt compatibility, run python3 scripts/sync-codex-prompts.py. Claude Code users typically do not need synchronization as the installer copies Markdown files directly.

How do the AI-Berkshire agents handle complex financial analysis?

According to the assets/architecture.mmd diagram and tools/financial_rigor.py implementation, each agent invocation spawns four parallel sub-agents specializing in business model analysis, valuation, competitive moat assessment, and risk evaluation. These sub-agents execute simultaneously and aggregate their findings, with financial rigor enforced through shared validation utilities in the tools/ directory.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}