claude-code-agentic-framework

内容来源:README.md(说明文档) · 原始地址 · 查看安装指南

原始内容

Codebase Singularity

A Claude Code plugin implementing the "Codebase Singularity" concept by IndyDevDan. This framework creates an agentic layer around your codebase, enabling AI agents to operate autonomously through planning, building, reviewing, and fixing workflows.

What is the Codebase Singularity?

The Codebase Singularity is the point where AI agents can run your codebase better than you can. The framework provides:

  • Structured Workflows: Plan-Build-Review-Fix cycle for all development
  • Specialized Agents: 9 agents for different tasks (planning, building, reviewing, etc.)
  • Quality Assurance: Automated code review with scoring and feedback
  • Multi-Agent Orchestration: Complex tasks coordinated across agents
  • Feedback Loops: Self-reviewing, self-correcting automation

Classification

This implementation is Class 3, Grade 1+ - the highest level of the agentic layer framework:

Class Description Status
Class 1 Foundation (memory, agents, tools, hooks) ✓ Complete
Class 2 Workflows (portable layers, E2E flows) ✓ Complete
Class 3 Orchestration (multi-agent coordination) ✓ Complete

Quick Start

In This Project

  1. Open Claude Code in this directory
  2. Activate with context:
    /prime
    
  3. Start developing:

In Other Projects

  1. Install globally (one-time):

    # Copy commands and agents to ~/.claude/
    cp -r .claude/commands/project ~/.claude/commands/singularity
    cp -r agents/*.md ~/.claude/agents/
    cp -r skills/* ~/.claude/skills/
    
  2. Initialize in new project:

    /setup
    
  3. Start developing:

    /plan Add user authentication
    /build
    /review
    /fix
    

    Or use the complete cycle:

    /cycle Add user authentication
    

Commands

Core Workflow

Command Purpose
/prime Activate with full context
/plan Create implementation plan
/build Implement from plan
/review Quality analysis
/fix Address issues

E2E Workflows

Command Purpose
/cycle Complete Plan-Build-Review-Fix
/feature Full feature development
/bugfix Bug investigation and fix

Orchestration

Command Purpose
/orchestrate Multi-agent coordination
/delegate Direct agent invocation

Utility

Command Purpose
/setup Initialize in new project

Agents

Agent Model Specialty
orchestrator opus Multi-agent coordination
planner opus Planning, architecture
builder sonnet Code implementation
reviewer opus Quality analysis
fixer sonnet Issue resolution
test-writer sonnet Test creation
doc-fetcher haiku Documentation research
security-auditor opus Security analysis
refactorer sonnet Code improvement

Structure

├── .claude-plugin/plugin.json   # Plugin manifest
├── commands/                     # 11 workflow commands
├── agents/                       # 9 specialized agents
├── skills/                       # 4 skill modules
├── hooks/hooks.json             # Lifecycle hooks
├── scripts/                      # Hook scripts
├── templates/settings.json      # Project settings
├── specs/                        # Plans and reviews
├── ai_docs/                      # AI reference docs
├── CLAUDE.md                     # Claude-specific memory
├── AGENTS.md                     # Universal agent instructions
├── .mcp.json                     # MCP configuration
└── README.md                     # This file

Core Workflow

The Plan-Build-Review-Fix cycle is the foundation:

    ┌─────────┐    ┌─────────┐    ┌─────────┐    ┌─────────┐
    │  PLAN   │───►│  BUILD  │───►│ REVIEW  │───►│   FIX   │
    └─────────┘    └─────────┘    └─────────┘    └────┬────┘
         ▲                                            │
         └────────── Loop until quality passes ◄──────┘

Documentation

Document Purpose
CLAUDE.md Claude-specific memory - read first
AGENTS.md Universal AI agent instructions
docs/QUICKSTART.md 20 hands-on use cases
docs/USER_GUIDE.md How to use the framework
docs/DEVELOPER_GUIDE.md Technical deep-dive
docs/AGENTIC_LAYER.md Core concept explanation
docs/CLASSIFICATION.md Class/Grade system reference
docs/HOOKS_GUIDE.md Hooks and feedback loops
docs/CONFIGURATION_GUIDE.md All settings and options

Using in New Projects

Prerequisites

Install commands globally (one-time setup):

# From this project directory:
cp -r .claude/commands/project ~/.claude/commands/singularity
cp -r agents/*.md ~/.claude/agents/
cp -r skills/* ~/.claude/skills/

Initialize New Project

cd /path/to/your/project
claude

# Run setup command
/setup

This creates:

  • CLAUDE.md - Project memory
  • AGENTS.md - Agent instructions
  • commands/ - Workflow commands
  • hooks/ - Automation hooks
  • specs/ - Plans and reviews

Configuration

Settings (templates/settings.json)

Configure agents, commands, hooks, and preferences.

MCP Servers (.mcp.json)

External tool integrations (filesystem, GitHub, memory).

Hooks (hooks/hooks.json)

Lifecycle automation (auto-lint, file protection, logging).

Acknowledgements

This project is inspired by the YouTube video:

The Codebase Singularity: "My agents run my codebase better than I can" by IndyDevDan

All code and documentation were generated by Claude Code powered by Claude Opus 4.5.

License

MIT