Uncle Frank's Bootstrap Core

Minimal bootstrap for the autonomous LLM development platform. No BS, just the core flow.

Setup

# Install dependencies
npm install

# Set environment variables
export CLAUDE_API_KEY="your-anthropic-api-key"
export TERRAGON_AUTH="JTgr3pSvWUN2bNmaO66GnTGo2wrk1zFf.fW4Qo8gvM1lTf%2Fis9Ss%2FJOdlSKJrnLR0CapMdm%2Bcy0U%3D"

# Copy sacred documents to your project
cp /path/to/CLAUDE.md .

Usage

# Run a task through the system
npm run dev task "Add Discord bot integration"

# Initialize a new project
npm run dev init

Core Flow

1. Request → Classification → Validation
2. Task decomposition into checkpoints
3. Human approval gate
4. Checkpoint execution via Claude
5. Pass/fail testing
6. Merge to production state

What's Included

• Meta-Agent: Request classification and task decomposition
• Document Manager: Project.md, Task.md, Checkpoint.md handling
• Claude Proxy: Integration with Claude execution engine
• Principle Enforcer: Sacred document validation
• Simple CLI: Human approval gates

What's NOT Included

• All 72 helper agents (add them using the system itself)
• Fancy UI (build it through the system)
• Complex dependency graphs
• Automated retries and escalations
• Discord bot, voice input, etc.

First Task

Use the bootstrap to add its first enhancement:

npm run dev task "Add validation system for cross-document coherence"

This will create the task, decompose it into checkpoints, and execute through Claude.

Philosophy

Every feature gets added through the system itself. No cowboy coding.