How splx-ai/agentic-radar works
Main Components
• CLI Interface: The entry point (agentic_radar/__main__.py and cli.py) where users pass input directories and command options.
• Core Analysis Engine: The central “agentic_radar/analysis” module which subdivides into multiple submodules:
• CrewAI Analysis (agentic_radar/analysis/crewai) – parsing, modeling, and converting graphs from CrewAI workflows.
• LangGraph Analysis (agentic_radar/analysis/langgraph) – analyzing LangGraph-specific structures and handling custom/predefined tools.
• n8n Analysis (agentic_radar/analysis/n8n) – converting n8n JSON configurations and parsing workflow details.
• OpenAI Agents Analysis (agentic_radar/analysis/openai_agents) – including parsing, error handling and categorization for OpenAI Agents workflows.
• Utility Modules: These include general helpers (agentic_radar/utils.py and analysis/utils.py) as well as tool categorizer and parsing helpers that support multiple framework modules.
• Report Generation: The “agentic_radar/report” submodule constructs the final comprehensive HTML security report that includes workflow graphs, lists of tools, and vulnerability mapping.
• Mapper: The module (agentic_radar/mapper) that appears to handle the mapping of identified components to known vulnerabilities (utilizing a JSON file that lists vulnerabilities).
• Utility modules to each analysis submodule, as these provide shared code (helpers, parsers, configuration loaders) required for processing.
• Mapper as a component that takes the analysis output (tools, workflow components) and cross-references them with vulnerability data (such as vulnerabilities.json or tool_categories.json) to create a vulnerability map.
Architectural Patterns and Design Principles
• Modularity: Each framework (CrewAI, LangGraph, n8n, OpenAI Agents) is handled via its own module within a larger analysis package. This makes the system extensible (evidenced by support for future frameworks like LlamaIndex, Swarm, etc.).
• Separation of Concerns: The CLI, Analysis, Mapping, and Report Generation layers are clearly separated to keep responsibilities distinct.
• Reusability: Utility modules and shared parsers allow code reuse across multiple framework-specific analyses.
Relevant Technologies and Frameworks
• Python as the primary language.
• Integration with external security standards (as mentioned in the README, such as OWASP guidelines).
• Use of Jinja templates (found inside agentic_radar/report/templates) for generating HTML reports.
• CI/CD tools (GitHub workflows) for automated tests and releases.
• Pre-commit and configuration for code quality.
• Input (code directory) → CLI interface → individual analysis submodules → mapping module → report generation.