Introduction
There are plenty of tools that promise to make coding faster. Fewer actually help you think better.
That difference matters. Most mistakes in software don’t come from typing speed or syntax. They come from unclear thinking, rushed decisions, or blind spots that only show up later when things break.
This article is about a setup that tries to address that problem directly: combining BMAD as a structured thinking framework with Qwen Code through its CLI. The goal is not just to generate code, but to guide how decisions are made before and during implementation.
To make that practical, I will use the Six Thinking Hats approach as the backbone for brainstorming. It forces you to look at a problem from multiple angles, instead of defaulting to the first idea that “seems good enough.” When paired with a CLI-driven coding assistant, it becomes a workflow: think deliberately, then execute quickly.
This is not a theoretical piece. It’s about how to actually use these tools together to reduce bad decisions. While my example will focus on software development, the tools and techniques presented here are broadly applicable to many contexts and industries, including decision-making, business plans, presentations, what-if scenarios, etc.
The Squad
This setup uses two components:
- Qwen Code (CLI) to write and run code from the terminal
- BMAD framework (v6) to structure thinking and guide decisions The BMAD framework is available for all common AI tools (Claude Code, Codex, etc.)
Qwen Code
https://github.com/QwenLM/qwen-code
Installing Qwen Code is straightforward, follow the official setup instructions. Using npm is recommended, as it is the standard package manager for most AI-related tooling. You must install NodeJS v20+, we will also need it for BMAD in a moment.
You will need a Qwen account to use the CLI. Notably, Qwen provides a very generous free-tier quota. It is possible to run full sessions at no cost that would otherwise cost tens of dollars on comparable plans such as Claude. You can even use a local model (Qwen models are opensource) if you want a 100% standalone and free version (but be aware BMAD may not work properly if model capabilities are too tight), check my article on this: https://medium.com/@kevin.drapel/your-local-qwen-with-qwen-cli-and-lm-studio-564ffb4c1e9e
Qwen can be invoked from the commandline by simply launching “qwen”. You will need to authenticate the first time (/auth command in case of issues), choose “Qwen OAuth” which is the free plan.
BMAD 6
The acronym stands for “Build More Architect Dreams” It is a comprehensive Agile development framework, but in this context only a small subset is used, specifically the brainstorming capability.
bmad-code-org
/
BMAD-METHOD
Breakthrough Method for Agile Ai Driven Development
Build More Architect Dreams — An AI-driven agile development module for the BMad Method Module Ecosystem, the best and most comprehensive Agile AI Driven Development framework that has true scale-adaptive intelligence that adjusts from bug fixes to enterprise systems.
100% free and open source. No paywalls. No gated content. No gated Discord. We believe in empowering everyone, not just those who can pay for a gated community or courses.
Why the BMad Method?
Traditional AI tools do the thinking for you, producing average results. BMad agents and facilitated workflows act as expert collaborators who guide you through a structured process to bring out your best thinking in partnership with the AI.
-
AI Intelligent Help — Invoke the
bmad-helpskill anytime for guidance on what's next - Scale-Domain-Adaptive — Automatically adjusts planning depth based on project complexity
- Structured Workflows — Grounded in agile best practices across analysis, planning, architecture, and implementation
- Specialized…
In practice, the framework installs a collection of agents and skills into a local folder. These can then be invoked by an AI coding agent, such as Qwen Code in this setup.
Preparing our Project for BMAD
Create an empty folder on your disk, this is the place where the project will be stored. BMAD will create a few folders next to the one of Qwen (.qwen).
Launch a terminal, navigate to the folder, then launch BMAD setup:
npx bmad-method install
First question in the installation process will be the folder, you can just press Enter to install it in the current one (and Enter a second time to confirm).
When asked for the modules to install, pick the 3 first modules (use arrows/space to enable them), then Enter to confirm.
BMAD will ask if you want to add custom modules or agents, you can answer No and continue. Next step is the integration with tools, here we need to select “QwenCoder”.
Next questions are about the interaction, how you want the agent to call you and in which language. The choice is up to you but we will continue in English. We also choose the default output folder.
The module configuration is then set on “Express Setup”.
BMAD finalizes the installation and we are ready to go.
Invoking The Brainstormer
With the setup complete, we can now use BMAD inside Qwen.
Launch Qwen and enter “/skills” command. You should see a long list of “bmad-xxxxx” skills indicating your installation was successful. We are going to use bmad-brainstorming.
Before going into the skill, let’s set the scenario for our brainstorming demonstration: we aim to develop a web application for gardeners that assists them in planning their next plantings. The app will help them decide what to plant, when to plant it, and the optimal locations for each plant in their garden.
We now need to expand this idea but we are not sure what the best options are, so we will use the bmad-brainstorming skill to help us. In Qwen, we simply type “bmad-brainstorming” as a prompt (in Claude Code and other tools, the skills are exposed as real commands).
bmad-brainstorming
Note that during the different steps, BMAD (through Qwen) will ask different permissions to write files into its folders. “It keeps track of the intermediate steps as markdown files.
The skill asks us what is the vision and what are the expected outcomes.
This is our first answer:
1) Make a simple web tool for gardeners that tells them what veggies to plant and where to put them. It should give seasonal tips based on where they live and let them map out their plants in basic beds or pots. Keep it simple, don’t ask for too much info, and focus on being clear rather than super precise. 2) I want to have a specification that can be used for the implementation.
BMAD summarizes the key aspects of the vision and wants us to confirm if we are aligned on the goals. It also proposes the approach for the next steps. There are different brainstorming methods available and depending on the kind of goal to achieve, you can let it propose the most appropriate choice. You can also enforce a given methodology.
We prompt “1” for the user-selected techniques, BMAD shows a list of categories. You could choose one of the main category or you can enforce a specific approach here: “Six Thinking Hats” by De Bono (https://en.wikipedia.org/wiki/Six_Thinking_Hats).
This method involves adopting multiple perspectives to examine goals and challenges. It is especially valuable in software design, as it clearly separates emotional, analytical, and creative thinking, helping reduce bias in decision-making. Each “hat” is associated with a specific colour and represents a distinct mindset.
I want to use Six Thinking Hats
BMAD invites us to either continue with this or add another methodology into our process. As we want to keep it simple for the sake of the demonstration, we just continue.
Going through the Hats
BMAD starts with the blue hat that is focused on the thinking process.
One important point: you are not required to write everything in detail, you can complete your ideas iteratively. This is where AI is extremely powerful as it can easily pivot. If a request is unclear or you can’t make up your mind, you can also say you don’t know, ask for some suggestions or simply skip an answer.
Coming back to the blue hat, we are asked what the key components are, logical flow, constraints. Here I skip the answer for question 2, I will only answer for 1 and 3.
- I need functional specifications, as well as technical specifications.
- It must be simple for a gardener, audience is not expected to be familiar with computers We get a summary of the blue hat and BMAD switches to the white hat which is focusing on facts. This is one of the most important hats because it builds the pillar leading to a robust analysis.
The questions are targeting the needs and required data to achieve the goals. We give some answers based on our expertise:
- When to plant, what kind of soil, sunny vs shady area, when is the harvest.
- location, season, space between plants.
- we need the location of the garden to figure out if the area is cold, mild, etc. we also need the spacing information and some plants cannot be planted together, you also need to consider rotations.
- watering needs
White hat is summarized and then the red hat is presented with a perspective based on the gut feeling.
The questions focus on the feeling of the users, what they like and what is frustrating them, as well as our own feeling. This is our answer:
- they should feel that the tool is reliable, they must feel relieved that they do not have to keep in mind all the scheduling
- missing planting at the right time + missing watering
- it is important to have a calendar and reminder of regular tasks (watering)
- users would love to be guided
I will skip the next hats (yellow, black, green) but the process is similar, you just answer to refine the AI’s understanding. The black hat is particularly interesting as it forces you to adopt a negative mindset, what could go wrong. The green one takes the opposite path, how to get out of the negative aspects. The yellow one focuses on the benefits.
At the end, the blue hat is presented again if we want to refine the mindset. We just ask BMAD to go forward with a “let’s continue”.
A summary of the hats is shown and again BMAD asks you what you want to do next. You can add more ideas, dive deeper into a particular topic or just continue. To keep the demo concise, we proceed directly to specification generation.
Generated Specification
A specification is quickly generated. It covers all sections you could expect from such document: overview, functional specification with features and user stories.
The technical aspects are covered as well as it was part of the information we requested: architecture, technical stack, data sources, APIs, etc.
Further Refinements
At this point, it is strongly recommended to read the first version of the generated document. You will probably have some changes to perform in it. You may have new ideas that come up during the review. Instead of simply changing manually the .md file, you can make a new round of brainstorming.
This is what we are going to do by invoking the “green hat” (the optimistic view) again
Can we make a run with a green hat, I would like to improve a few things
As we had made a first full pass of Six Thinking Hats, the green hat is now presented with some additional ideas that you can explore.
Green hat comes up with some ideas such as a webcam in the garden.
We propose a few new ideas but also ask for some advice:
I think it would be nice to have an idea of the incoming weather in that area. I think users would also like to have the schedule that is printable (PDF?) . Can you also propose me some ideas going into those directions?
The green hat comes up with an extensive set of areas to explore and summarize them. Our proposal is the first one in the list, but BMAD also brought some new ideas such as a “harvest goals calculator” or a gamification with “streaks & badges”. We decide to go for the practical additions (B).
B is nice, proceed with that
BMAD proceeds with the necessary changes in our specification.
Jumping to another Brainstorming Method
Maybe you feel the Six Thinking Hats is not sufficient or appropriate for your usage, you can instruct BMAD to propose you an alternative:
Can you suggest another brainstorming method that would be appropriate? We would continue with it
It recommends a few options, the top 3 is:
Question Storming: You have a solid spec now identify what you don’t know yet.
Assumption Reversal: Challenge the design decisions you’ve made to find better alternatives.
Constraint Mapping: Before implementation, know which constraints are real vs. imagined
We go with the Assumption Reversal.
We get challenged on the location, what happens if the user NEVER enters his location. This removes a lot of information that could be useful for the garden management.
The user will not have access to weather information. It will not be possible to give advices on the right time for planting. An alternative could be to allow the user to choose a hardiness zone if he does not want to disclose his exact location. If none of this is possible, the suggestions will be disabled.
If we stop the process at that point, we get a conclusion with our reversal thinking. After each major milestone, we are proposed to continue with brainstorming, adapt the plan or proceed with the implementation.
Conclusion
The brainstorming tool in BMAD is not just a way to generate ideas, it is a way to structure thinking.
By combining perspective-based reasoning and exploration, it transforms what is usually an unstructured and sometimes inconsistent process into something repeatable and reliable. Ambiguities are reduced early, hidden constraints surface naturally, and ideas evolve into more coherent solutions.
The time invested in this phase is not wasted. It directly impacts the quality of the outcome by strengthening the foundations of the solution before any implementation begins.
Used properly, BMAD does not just help you think of better ideas, it helps you design better systems.
Original article:
Top comments (0)