Codev

Overview#

Codev is a software development operating system designed for human-agent collaboration, with the core philosophy of Context Drives Code. It introduces the SPIR protocol (Specify-Plan-Implement-Review), structuring the software development process into four stages: specification definition, plan decomposition, code implementation, and review improvement.

SPIR Protocol#

A structured development protocol with four phases:

• Specify: Define what to build in clear, unambiguous language
• Plan: Break down specifications into executable phases
• Implement: Build code, write tests, and verify requirements for each phase
• Review: Capture lessons learned and improve methodology

Key Features#

• 📄 Natural Language as Primary Programming Language: Specifications and plans drive implementation, all decisions are version-controlled
• 🤖 AI-Native Workflow: Structured formats that AI agents can understand; multi-agent consultation support (GPT-5, Gemini Pro, etc.)
• 🔄 Continuous Improvement: Each project improves methodology, lessons learned feed back into the process
• 📋 Standard Compatibility: Supports AGENTS.md standard (Cursor, Copilot, etc.) and CLAUDE.md (Claude Code)

Agent Farm#

Optional component providing a web-based dashboard for parallel management of multiple AI agents:

• Protocol-aware, understands specifications, plans, and Codev conventions
• Git worktree management, isolating each builder's changes
• Auto-prompting, builders start with their assigned specification tasks
• Supports Architect-Builder pattern

Project Structure#

your-project/
├── codev/
│   ├── specs/              # Feature specifications
│   ├── plans/              # Implementation plans
│   ├── reviews/            # Reviews and lessons learned
│   └── resources/          # Reference materials
├── AGENTS.md               # AI agent instructions
├── CLAUDE.md               # AI agent instructions
└── [your code]

Installation & Usage#

Requirements: Node.js 18+, Git 2.5+

# Global installation
npm install -g @cluesmith/codev

# Initialize project
mkdir my-project && cd my-project
codev init

# Verify installation
codev doctor

# Start Agent Farm dashboard
af dash start

CLI Commands#

• codev init - Initialize Codev project structure
• codev adopt - Add Codev to existing project
• codev doctor - Check dependencies and environment
• af spawn <spec_number> - Spawn builder for specified spec
• af status - Check running status
• consult - Multi-model consultation

Configuration Example#

Configure AI Shell in af-config.json:

{
  "shell": {
    "architect": "claude --dangerously-skip-permissions",
    "builder": "claude --dangerously-skip-permissions"
  }
}

Performance Data#

According to project case studies:

• SPIR Method Score: 92-95 points
• VIBE (Conversational) Method Score: 12-15 points
• Reduces back-and-forth messages from dozens to 3-4 document reviews

Project Info#

• License: Apache-2.0
• Primary Language: TypeScript (90.8%)
• Current Version: v2.1.1
• Releases: 39 versions
• Commits: 2,408

Pending Verification#

• Commercial support: No explicit mention of enterprise or commercial support services
• Performance benchmarks: Comparison scores lack detailed testing methodology
• Security audit: No mention of code security audit or third-party security assessment
• Large-scale adoption: No enterprise production environment case studies found