Skip to content

digital-rain-tech/harness-lab

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

9 Commits
 
 
 
 
 
 
 
 

Repository files navigation

Harness Lab

Research repository for analyzing Claude Code's architecture, patterns, and implementation details.

Disclaimer

This repository contains code extracted from the leaked Claude Code source for research and educational purposes only.

  • No proprietary claims: We make no claims of ownership over this code
  • Academic use: Intended for studying agent harness engineering, tool orchestration, and AI system architecture
  • Not for distribution: Code is preserved as-is for analysis; do not use in commercial products
  • Credit: All code belongs to the original authors at Anthropic

This project is not affiliated with, endorsed by, or maintained by Anthropic or Claude Code.

What's Here

  • src/commands/ — CLI command implementations (~70+ commands)
  • src/tools/ — Tool definitions and executors (~40+ tools)
  • src/hooks/ — Pre/post tool use hooks
  • src/plugins/ — Plugin system
  • src/skills/ — Skill loading and bundled skills
  • src/buddy/ — Companion sprite feature
  • src/services/autoDream/ — Background memory consolidation
  • src/tasks/DreamTask/ — Dream task implementation
  • src/utils/ — Utility functions including messages.ts, messagePredicates.ts, messageQueueManager.ts
  • src/constants/messages.ts — Message constants
  • src/components/messageActions.tsx — Message action components
  • Various other utilities, components, and state management

Missing / Excluded Files

The following files appear to be excluded from the leak (original TS source not present):

  • types/message.ts — Core message type definitions (imported by 184 files but not in source)
  • These are likely original TypeScript files that were filtered out

Research Focus

  • Agent execution systems and tool orchestration patterns
  • CLI-based AI workflow design
  • Memory management and consolidation strategies
  • Feature flagging and phased rollout systems
  • Multi-agent coordination

Motivation

Claude Code represents a significant engineering effort in AI agent systems. This repo exists to study how those systems work — from the tool execution pipeline to the user interface — without the noise of the original repository.

Related

  • claw-code — Clean-room rewrite in Python/Rust

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors