The workflow for extracting and analyzing AI intelligences using Agent-Wiz is divided into three main steps: extraction, visualization, and threat modeling. Here are the specifics of how to use it:
1. Extraction workflow
pass (a bill or inspection etc)extractcommand to statically parse the code and extract the workflow. For example, for a CrewAI-based project, run the following command:
agent-wiz extract --framework crewai --directory ./my_project --output workflow.json
--framework: Specify the frame type (supported)autogen,crewai(etc.).--directory: Specifies the project folder path.--output: Specifies the path of the output JSON file.
The generated JSON file contains nodes (intelligences, tools, etc.) and edges (interactions) for subsequent analysis and visualization.
2. Generation of visualization charts
utilizationvisualizecommand converts a JSON file into an interactive chart:
agent-wiz visualize --input workflow.json --open
--input: Specifies the input JSON file.--open(Optional): Automatically opens the chart in the browser.
The chart is based on D3.js, supports zooming and dragging, and is able to visualize the interaction between intelligences.
3. Threat modeling analysis
utilizationanalyzecommands can perform security risk assessments based on the MAESTRO framework:
agent-wiz analyze --input workflow.json
Once the analysis is complete, the tool generates a report in Markdown format that includes:
- List of system objectives and assets
- Attack entry points and potential threats
- Risk assessment and security recommendations
This answer comes from the articleAgent-Wiz: Analyzing AI Intelligentsia Workflows and Security RisksThe
English 
简体中文 
日本語 
Deutsch 
Português do Brasil