SecureFlow: AI-Powered Security & Development Crew

SecureFlow is a comprehensive AI-powered system for security assessment and application development using CrewAI's multi-agent orchestration framework.

Status: ✅ Production-Ready | Version: 0.1.0

Features

🔒 Security Assessment Crew

• Recon Agent (Gemini) - Network scanning, service enumeration, CVE discovery
• Analyst Agent (Claude/OpenCode) - Vulnerability analysis, risk assessment, exploitation chains
• Reporter Agent (Ollama) - Executive reporting, remediation roadmap, CVSS scoring

💻 Application Development Crew

• Architect Agent (Claude/OpenCode) - System design, tech stack selection, project structure
• Developer Agent (Gemini) - Code implementation, feature building, file creation
• Reviewer Agent (Ollama) - Code quality review, bug detection, improvement suggestions

🎯 Key Capabilities

• Multi-agent collaborative workflows
• Shared context and inter-agent communication
• FastMCP server with SSE transport
• CLI interface with full command support
• Comprehensive test suite with pytest
• Production-ready deployment

Quick Start

Installation

# Clone or download SecureFlow
cd /home/sky/secureflow

# Create and activate virtual environment
python3 -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

# Install with development dependencies
pip install -e ".[dev]"

First Run

# Check system status
secureflow status

# View available commands
secureflow --help

Usage

Security Assessment

# Run full security assessment on target
secureflow scan scanme.nmap.org

# The assessment will:
# 1. Run network reconnaissance (nmap, CVE lookup)
# 2. Perform vulnerability analysis
# 3. Generate executive report with remediation plan

Application Development

# Build an application
secureflow build "Create a REST API for a todo application" \
  --language python \
  --output /tmp/todo-api

# The development crew will:
# 1. Design system architecture
# 2. Generate production code
# 3. Review code quality

MCP Server

# Start the FastMCP server
secureflow server

# Server will listen on port 5000 with SSE transport
# Available tools:
# - run_security_crew(target)
# - run_dev_crew(task, language, output_dir)
# - run_code_review(code, language)
# - crew_status()

Configuration

Environment Variables

Create a .env file (copy from .env.example):

cp .env.example .env

Required variables:

• MCP_SECRET - Bearer token for authentication (generate: openssl rand -hex 32)
• GEMINI_API_KEY - Google Gemini API key (optional, for Recon agent)
• OPENCODE_URL - OpenCode server URL (default: http://localhost:4096)

Optional:

• OLLAMA_BASE_URL - Local Ollama server (default: http://localhost:11434)
• TELEGRAM_BOT_TOKEN - For Telegram notifications
• TELEGRAM_CHAT_ID - For Telegram notifications

Testing

# Run all tests
pytest tests/ -v

# Run with coverage
pytest tests/ --cov

# Run specific test file
pytest tests/test_cli.py -v

# Run specific test
pytest tests/test_cli.py::test_cli_version -v

Architecture

Security Crew Workflow

Phase 1: RECONNAISSANCE
  Recon Agent (Gemini)
  - nmap scan for open ports
  - Service version detection
  - CVE lookup for each service
  └─> Writes findings to SharedContext

Phase 2: ANALYSIS
  Analyst Agent (Claude/OpenCode)
  - Reads recon findings
  - Correlates CVEs
  - Assesses risk severity
  - Identifies exploitation chains
  └─> Writes analysis to SharedContext

Phase 3: REPORTING
  Reporter Agent (Ollama)
  - Reads all findings
  - Creates executive summary
  - Plans remediation roadmap
  - Generates HTML report
  └─> Returns final report

Development Crew Workflow

Phase 1: ARCHITECTURE
  Architect Agent (Claude/OpenCode)
  - Analyzes requirements
  - Designs system architecture
  - Recommends tech stack
  - Plans project structure
  └─> Writes design to SharedContext

Phase 2: IMPLEMENTATION
  Developer Agent (Gemini)
  - Reads architecture design
  - Generates code modules
  - Creates project files
  - Implements core features
  └─> Writes code to SharedContext

Phase 3: REVIEW
  Reviewer Agent (Ollama)
  - Reads generated code
  - Checks code quality
  - Identifies bugs
  - Suggests improvements
  └─> Returns review report

Project Structure

secureflow/
├── secureflow/           # Main package
│   ├── cli.py           # CLI interface
│   ├── server.py        # FastMCP server
│   ├── config.py        # Configuration
│   └── crew/            # Multi-agent system
│       ├── agents.py    # Agent definitions
│       ├── tasks.py     # Task definitions
│       ├── tools.py     # Tools for agents
│       ├── orchestrator.py      # Security workflow
│       ├── dev_orchestrator.py  # Dev workflow
│       ├── memory.py    # Shared context (SQLite)
│       └── chat.py      # Inter-agent messaging
├── tests/               # Test suite
├── docs/                # Documentation
└── pyproject.toml       # Package metadata

Development

Code Style

# Format code with black
black secureflow/ tests/

# Check with flake8
flake8 secureflow/ tests/

# Type checking with mypy
mypy secureflow/

# Organize imports with isort
isort secureflow/ tests/

Adding New Tools

1. Create tool function in secureflow/crew/tools.py:

@tool("Tool Name")
def my_tool(param: str) -> str:
    """Tool description."""
    result = {"status": "success", "data": "..."}
    return json.dumps(result, indent=2)

1. Add to agent:

from secureflow.crew.tools import my_tool

@staticmethod
def agent_with_tool():
    return Agent(
        role="Agent Role",
        tools=[my_tool],
        # ... other config
    )

Deployment

Production with systemd

# Copy to production location
sudo cp -r secureflow /opt/

# Create service file
sudo cp secureflow.service /etc/systemd/system/

# Enable and start
sudo systemctl enable secureflow
sudo systemctl start secureflow

# Check status
sudo systemctl status secureflow

Docker

FROM python:3.12-slim
WORKDIR /app
COPY . .
RUN pip install -e ".[dev]"
CMD ["secureflow", "server"]

Troubleshooting

Import Errors

# Verify environment
secureflow status

# Reinstall package
pip install -e ".[dev]" --force-reinstall

Server Not Starting

# Check port availability
lsof -i :5000

# View logs
tail -f /tmp/crew_session.log

Tests Failing

# Run with verbose output
pytest tests/ -vv

# Show print statements
pytest tests/ -s

# Run single test
pytest tests/test_cli.py::test_cli_version -vv

Performance

• Recon Phase: 30-120 seconds (network dependent)
• Analysis Phase: 60-180 seconds (LLM response time)
• Reporting Phase: 30-90 seconds (report generation)
• Total: 2-5 minutes per security assessment

Security

• ✅ Bearer token authentication on MCP server
• ✅ No credentials in logs or session files
• ✅ Local SQLite storage for findings
• ✅ Optional Telegram notifications
• ✅ No external data sharing by default

Limitations

• Requires network access for NVD CVE lookups
• OpenCode server must be running for Claude-based agents
• Large codebases may timeout during analysis

Future Improvements

• [ ] Persistent result storage
• [ ] Web UI dashboard
• [ ] Real-time progress streaming
• [ ] Scheduled assessments
• [ ] Integration with SIEM systems
• [ ] Custom tool marketplace
• [ ] Plugin system for extensions

Contributing

Contributions welcome! Please:

1. Fork the repository
2. Create feature branch
3. Add tests for new code
4. Submit pull request

License

MIT License - see LICENSE file for details

Support

• Issues: Report bugs on GitHub
• Documentation: See docs/ directory
• Examples: Check tests/ for usage patterns

Citation

If you use SecureFlow in research, please cite:

@software{secureflow2026,
  title={SecureFlow: AI-Powered Security Assessment and Development},
  author={SecureFlow Team},
  year={2026},
  url={https://github.com/secureflow/secureflow}
}

---

Built with ❤️ using CrewAI and FastMCP

Status: ✅ Production-Ready | Version 0.1.0 | Last Updated: May 2026