agentic-sdlc 1.0.0

AI-powered SDLC framework with self-learning brain, automated workflows, and intelligent knowledge management

Agentic SDLC

Transform your IDE into a complete Software Development Lifecycle team with AI-powered agents, automated workflows, and intelligent knowledge management.

🎯 What is Agentic SDLC?

Agentic SDLC is an AI-powered development framework that simulates a complete software development team within your IDE. It provides:

• 17 Specialized AI Roles - PM, BA, SA, UI/UX, QA, Security, Dev, DevOps, Tester, Reporter, and more
• 23 Automated Workflows - From planning to deployment with /slash commands
• Reinforced Brain System - 21 Intelligence sub-agents including HITL, Sandbox, and Self-Healing
• Multi-Agent Teams - AutoGen-powered autonomous agent collaboration
• Cross-IDE Compatibility - Works with Cursor, Windsurf, Cline, Aider, Gemini, and any AI-powered IDE
• Monorepo Architecture - Shared brain system across multiple projects

🏗️ Architecture & Flows

3-Layer Concentric Architecture

The system is built on a robust 3-layer architecture where dependencies flow inward (Layer 3 → Layer 2 → Layer 1):

Layer 1 (Core) - The stable foundation that rarely changes:

• GEMINI.md - Universal guide and single source of truth
• Skills - 17 AI role definitions
• Templates - 20+ document templates
• Rules - 8 rule files for enforcement
• Workflows - 23 workflow definitions

Layer 2 (Intelligence) - The brain system with 21 sub-agents:

• Monitoring & Compliance: Observer, Monitor, Workflow Validator
• Quality & Scoring: Judge, Scorer, Evaluation
• Learning & Optimization: Self-Learning, DSPy, A/B Test
• Execution & Safety: HITL, Sandbox, Self-Healing
• Intelligence & Routing: Proxy, Router, Task Manager, Research
• Generation & Tracking: Artifact Gen, Cost, Performance

Layer 3 (Infrastructure) - External interfaces and tools:

• CLI, MCP Connectors, GitHub Integration
• Neo4j Knowledge Graph, Documentation
• Workflow Scripts, Communication Tools

Orchestrator Workflow with HITL Gates

The orchestrator workflow manages the complete SDLC lifecycle with mandatory human approval gates:

Key Features:

• 11 Phases: From Planning to Self-Learning
• 3 HITL Gates: Design Approval, Code Review, Deployment Approval
• 4 Checkpoints: State persistence at critical phases
• Self-Healing Loop: Automatic fix/retry on test failures
• Brain Status Checks: Validation at start and end

Flow:

1. Planning (@PM) → Create project plan
2. Requirements (@BA) → Define user stories
3. Design (@SA + @UIUX) → Architecture & UI/UX specs
4. 🛑 HITL Gate 1 → Human approval required
5. Verification (@TESTER + @SECA) → Quality & security review
6. Development (@DEV + @DEVOPS) → Feature implementation
7. 🛑 HITL Gate 2 → Code review & PR approval
8. Testing (@TESTER) → E2E testing with self-healing
9. Bug Fixing (@DEV) → Issue resolution
10. Deployment (@DEVOPS) → Production deployment
11. 🛑 HITL Gate 3 → Production approval
12. Reporting (@PM) → CHANGELOG & review
13. Self-Learning (@BRAIN) → KB sync & archive

Brain Intelligence Sub-Agents Network

The brain system consists of 21 specialized sub-agents working in harmony:

Data Flow Patterns:

• Compliance Feedback Loop: Observer → Judge → Self-Learning
• Quality Improvement Loop: A/B Test → Judge → Self-Learning
• Persistence: All agents → State Manager
• Learning: All agents → Knowledge Graph (Neo4j)
• Approval Gates: HITL ↔ Critical phases
• Auto-Fix: Self-Healing ↔ Testing
• Model Routing: Proxy → All agents
• Cost Tracking: Cost → All agents

Brain Learning Loop

Every task execution improves the system through compound learning:

8-Step Learning Cycle:

1. Execute Task - Any SDLC phase
2. Observer Monitors - Track actions & violations
3. Judge Scores - Quality assessment (0-100)
4. A/B Testing - Generate alternatives
5. Self-Learning - Extract patterns:
   • Observer violations → New rules
   • Judge scores → Quality patterns
   • A/B results → Best solutions
   • Completed tasks → Reusable solutions
   • Fixed bugs → Anti-patterns
6. Knowledge Storage - Neo4j Graph + SQLite State + LEANN Vector Search
7. Context-Aware Suggestions - Smart recommendations
8. DSPy Optimization - Improve prompts

Side Flows:

• Error Path: Self-Healing → Fix → Back to execution
• Cost Path: Cost Monitor → Budget alerts
• State Path: State Manager → Checkpoints

SDLC State Machine

Complete state machine showing all transitions and error handling:

States & Transitions:

• IDLE → brain init → PLANNING
• PLANNING (@PM, @BA, @PO) → User Approval → DESIGN
• DESIGN (@SA, @UIUX) → HITL Approval → VERIFICATION
• VERIFICATION (@TESTER, @SECA) → Passed → DEVELOPMENT
• DEVELOPMENT (@DEV, @DEVOPS) → HITL Code Review → TESTING
• TESTING (@TESTER) → Tests Passed → DEPLOYMENT
  • Tests Failed → Self-Healing → Back to DEVELOPMENT
• DEPLOYMENT (@DEVOPS) → HITL Production Approval → REPORTING
• REPORTING (@PM, @REPORTER) → Complete → LEARNING
• LEARNING (@BRAIN) → Done → IDLE

Error Handling:

• Any State → ERROR → HALTED
• HALTED → Fix Issue → Resume → Previous State

Checkpoints:

• Planning, Design, Development, Deployment

🧠 The Brain System

At the core of Agentic SDLC is the Brain - an intelligent knowledge management system that:

• Learns from every task - Automatically captures patterns from bugs, features, and solutions
• Provides recommendations - Suggests approaches based on past successes
• Builds knowledge graphs - Maps relationships between skills, technologies, and solutions
• Enables compound intelligence - Each project's knowledge benefits all others

→ See GEMINI.md for complete Brain documentation

✨ Installation

🚀 Quick Install

Install directly from GitHub:

pip install git+https://github.com/truongnat/agentic-sdlc.git

📦 Initialize in Your Project

Navigate to your project and initialize:

cd your-project
agentic-sdlc init

This will:

• Create .agent/ directory with workflows, skills, and templates
• Set up brain system configuration
• Initialize knowledge base
• Create .env.template for API keys

⚙️ Configuration

After initialization, configure your API keys (optional):

cp .env.template .env
# Edit .env with your credentials

🔧 Development Setup (From Source)

For contributing or development:

git clone https://github.com/truongnat/agentic-sdlc.git
cd agentic-sdlc

# Run setup script
./bin/setup.sh  # Linux/macOS
.\bin\setup.ps1  # Windows

# Or install in editable mode
pip install -e .

🧠 System Commands

All operations are centralized through the asdlc script.

Windows (PowerShell)

.\bin\asdlc.ps1 <command>

Linux / macOS (Bash)

./bin/asdlc.sh <command>

📊 System Dashboard

Monitor agents, costs, and approvals via the real-time dashboard:

python asdlc.py dashboard

🚀 Reinforced Intelligence Features

The system has been enhanced with enterprise-grade reliability:

• 🛡️ Sandboxing: Securely execute agent-generated code in isolated Docker containers.
• 🛑 HITL (Human-in-the-Loop): Mandatory approval gates for critical phases (Deploy, Security, Code Review).
• 🔄 Persistence & Recovery: Workflow session state management with SQLite-based checkpointing.
• 🩹 Self-Healing: Automated QA→DEV feedback loops that learn from error patterns.
• 💰 Cost Monitoring: Real-time token tracking and budget alerts per model/task.
• 🏆 Evaluation: Robust benchmarking framework to measure and improve agent performance.
• 🏠 Local LLM Support: Privacy-first execution using Ollama for local model hosting.

🚀 Core Features

1. AI Role System (17 Roles)

Specialized AI agents for every SDLC phase:

Planning    → @PM, @BA, @PO
Design      → @SA, @UIUX
Review      → @QA, @SECA
Development → @DEV, @DEVOPS
Testing     → @TESTER
Delivery    → @REPORTER, @STAKEHOLDER
Meta        → @BRAIN, @ORCHESTRATOR

2. Slash Commands (23 Workflows)

Execute complete workflows with simple commands (mapped to asdlc workflow <name>):

/brain           # Brain system management (asdlc brain status)
/cycle           # Complete task lifecycle
/explore         # Deep investigation
/orchestrator    # Full SDLC automation
/sprint          # Sprint management
/monitor         # System dashboard
/validate        # System validation
/metrics         # View metrics dashboard
/release         # Release management
/emergency       # Critical incident response
/housekeeping    # Cleanup & maintenance

3. Monorepo Architecture

agentic-sdlc/              # 🧠 Brain (Root)
├── .agent/                # AI workflows, skills, KB
├── tools/                 # Neo4j, research, utilities
├── docs/                  # Documentation
└── projects/              # Your projects
    ├── project-1/
    ├── project-2/
    └── [add-yours]/

Benefits:

• ✅ Shared brain across all projects
• ✅ Compound learning from every solution
• ✅ Consistent workflows and quality
• ✅ Centralized knowledge management

4. Knowledge Management

Automated Learning:

• Records error patterns and solutions
• Captures successful implementation approaches
• Builds skill and technology graphs
• Provides context-aware recommendations

Three-Layer System:

1. LEANN - Vector-based semantic search
2. Neo4j - Knowledge graph with relationships
3. File-based KB - Categorized markdown entries

📖 Documentation

Getting Started

• GEMINI.md - Complete brain system guide (IDE-agnostic)
• Quick Start - 5-minute setup guide
• CLI Examples - Command usage examples

Architecture

• Monorepo Architecture - System design
• Project Structure - Directory organization
• Documentation Index - All docs

Tools & Setup

• Neo4j Tools - Knowledge graph system
• Research Agent - Automated research
• MCP Setup - Model Context Protocol

🎯 Use Cases

Solo Developer

/auto Create a SaaS platform with authentication and billing
# Complete automation from planning to deployment

Team Development

# Each team member uses the same brain
python asdlc.py brain sync
git pull  # Share knowledge base
/pm Start Sprint 3

Existing Large Project

python asdlc.py setup
/brain  # Index and analyze codebase
/pm Migrate authentication to OAuth2

🔧 Available Commands

# System Dashboard
python asdlc.py dashboard       # Start the UI monitoring

# Brain & Intelligence
python asdlc.py brain status    # Check system state
python asdlc.py brain health    # Full health check
python asdlc.py brain sync      # Sync knowledge graph

# Workflows
python asdlc.py workflow cycle  # Run task lifecycle
python asdlc.py workflow orchestrator  # Full automation

# Release Management
python asdlc.py release preview # Preview changes
python asdlc.py release release # Full release cycle

🌟 Why Agentic SDLC?

Traditional Development	With Agentic SDLC
Manual planning	Automated with @PM
Ad-hoc architecture	Structured with @SA, @UIUX
Inconsistent code quality	Enforced by @QA, @SECA
Lost knowledge	Compound learning brain
Repetitive tasks	Automated with @AUTO
Single-agent limits	Multi-agent teams with AutoGen
Solo problem-solving	17+ AI experts available

🔗 Links

• Repository: https://github.com/truongnat/agentic-sdlc
• NPM Package: https://www.npmjs.com/package/agentic-sdlc
• Issues: https://github.com/truongnat/agentic-sdlc/issues
• Documentation: docs/

📄 License

MIT License - See LICENSE for details

---

Next Steps:

1. Read GEMINI.md to understand the brain system
2. Follow Quick Start to get started
3. Explore workflows to see available automations

Questions? Check the documentation or open an issue.