BMAD-METHOD V2 in an Evolution IMO - The POWER of Custom Agents, Smaller Docs, and CHECKLISTS!

How To
1

If you’ve been trying to build with AI in Cursor but running into noisy outputs, repeated context loss, or burning through credits while iterating, I wanted to share a system I’ve been refining for the past month, now at V2.

The BMAD Method V2 is a full AI dev workflow that uses custom agent modes to guide each phase of a project from brainstorming to architecture to dev handoff.

The real shift came when I started doing all planning outside the IDE using Gemini Web UI with Custom GEM and 2.5 Pro Thinking. The agent modes (shared in the repo) are optimized to work inside Gemini or ChatGPT, where long context windows can be used to full advantage. Then, when you bring the output into Cursor, it’s already streamlined and structured for dev.

What it includes:

  • Custom agent modes for BA, PM, Architect, PO, Scrum Master, and Dev
  • Prompts that pause for feedback before moving on
  • The POWER of Checklists at the end of each agent phase that audits across all new and existing artifacts - this is HOLY ■■■■ next level game changer
  • Stories generated one at a time, fully aligned with what the Dev agent expects
  • Instead of Giant PRD and Architecture Docs that repeat info and overload the dev context - files are no many and single purpose and do not repeat. And the Scrum Master will pull together just the info the dev agent mode needs so it does not need to load a lot of extra files. This really has made a huge difference.
  • Significant cost savings — users are reporting hundreds per month saved in LLM and Cursor usage by front-loading the planning
  • No rules required, but works with them if you have a setup already

Everything is documented and open source, including a full video walkthrough, prompt configs, and example outputs.

Video walkthrough:

BMAD V2 GitHub repo (prompts, instructions, examples):

Let me know if you’re experimenting with something similar in Cursor or want help getting started with the prompts.

3 Likes
2

Yesterday I saw the repository, but I was a little confused

I’ll test right now!

1 Like
3

Thanks - yeah there are a lot of docs, but it could be overwhemling - let me know how well it works for you!

4

in V2 there are 7 agent persona’s but the 8 limit in cursor, and the default 3 mean I can’t set this up to test. is there a workaround?

5

V3 I am resolving this issue - since Cursor does not seem to be raising the limit anytime soon.

But I still highly recommend if using all agents to create the brief, prd, architecture, design docs, prompts for UI tools like v0 or lovable - really recommend using Gemini 2.5 Pro in the Web - and throw in its deep research capabilities, and super powerful depending on what task is at hand in your planning phases - 1M context window and quality of that model is unbeatable at this time compared to what the lower context windows will allow and cost in cursor (IMO) - have tried both and much better in the web.

V3 beta of the repo has fixed the issue of clean exports of the docs from gemini.

Doing this, only really need SM and dev agent - other useful tasks are being moved to task files and checklists in v3 that any agent can run also reducing the overall ide version of the agents.