MBAD + Claude Code IDE is unstoppable.

[Jmjcoke]

Are you sick of starting new chat sessions all the time to stop the dumbing down of the model. Claude Code is immune to this problem. It compresses the chat at intervals and keeps a running % until compression under the input box. I installed Claude code into cursor (Mac) and with the built in tools had it read the MBAD GitHub and copy over all the files and kept one chat going for 15 hours.

I also installed Claude code into cursor on a windows machine, you have to go the WLS Ubuntu route. But with the WLS extension in cursor it is great once you go thru all the setup.

[bmadcode]

That is wild - I just read up on Claude Code, and it does sound amazing! Do you do most of your work in Cursor with its agents and then when you have a large complex task, kick it over to claude code? Curious how the cost is racking up or if you hit usage limits on one of the plans such as the 5x or 20x?

I am tempted to try it out - I read up on its memory optimizations, and also that it does not charge to relook up previously looked up content, so this could save quite a bit - all with the power of claud 4 opus - wild!

Please share more details if you dont mind @Jmjcoke ! For your 15 hour chat - do you know what the total cost was - and what that like active marathon 15 hours of coding? (or I guess spread out, but net 15 hours total in one chat)?

Has the future already arrived and context windows are a thing of the past?? haha dang.

[cabinlab]

'adopt different mindsets for intentional friction.'

Yeah, in some sessions Claude would try to abstract out and significantly reduce the BMAD parts. It would say things like, "I'm already good at planning and breaking things down into discreet tasks. and looking at problems from multiple angles."

At some point I had it read the full BMAD context and analyze for commonalities with its strengths, then inverted the question to ultrathink its weaknesses, and added I share those weaknesses. Then we were vibing and it framed BMAD as useful "guardrails" and made the friction comment.

Since then it's less likely to reframe BMAD as 'stuff I do anyway'.

[dracic]

'adopt different mindsets for intentional friction.'

Yeah, in some sessions Claude would try to abstract out and significantly reduce the BMAD parts. It would say things like, "I'm already good at planning and breaking things down into discreet tasks. and looking at problems from multiple angles."

At some point I had it read the full BMAD context and analyze for commonalities with its strengths, then inverted the question to ultrathink its weaknesses, and added I share those weaknesses. Then we were vibing and it framed BMAD as useful "guardrails" and made the friction comment.

Since then it's less likely to reframe BMAD as 'stuff I do anyway'.

Oh, Master, bestow upon us yet more wisdom concerning Claude Code, the secret blade of the development engineers. Ere we forge the pact for the Max Concession...
👍

[cabinlab]

Oh, Master, bestow upon us yet more wisdom concerning Claude Code, the secret blade of the development engineers. Ere we forge the pact for the Max Concession... 👍

Careful! I've been talking so much to Claude the last couple days that I may overinterpret any human interaction and invite myself to dinner 🤣

[ksylvan]

'adopt different mindsets for intentional friction.'

Yeah, in some sessions Claude would try to abstract out and significantly reduce the BMAD parts. It would say things like, "I'm already good at planning and breaking things down into discreet tasks. and looking at problems from multiple angles."

At some point I had it read the full BMAD context and analyze for commonalities with its strengths, then inverted the question to ultrathink its weaknesses, and added I share those weaknesses. Then we were vibing and it framed BMAD as useful "guardrails" and made the friction comment.

Since then it's less likely to reframe BMAD as 'stuff I do anyway'.

I kind of love this interaction of Claude acting like an arrogant junior Dev and having you reframe its view.

[ksylvan]

also @cabinlab - this is a great line of instruction I will add into the orchestrators! 'adopt different mindsets for intentional friction.'

I have seen that emergent behavior actually with /party-mode in the web where I had it mock a post planning 'retro' where it was really good at giving different perspectives and one agent said something about the another agent dropping the ball LOL

WOW!!! Super funny. Everything we've all seen in real human agile teams, enacted by the LLM agents! 🤯🤣

[dracic]

I was playing yesterday with the similar setup. I don't have any subscription, just using prepaid API. Spent couple euros, it works amazing. BMAD commands don't work though, it only recognizes claude code commands, but everything else works amazing.

[dracic]

Your implementation in CLAUDE.md, please explain further, where did you put the other files, did you just copy the Agent Orchestrator into the CLAUDE.md file? It seems like this is a great way to extend the personality profiles into the Claude Code Environment - then you can create your own. - Thanks for sharing

Yes, just copied the Agent Orchestrator to CLAUDE.md. bmad-agent and docs folder as usual.

[kaefcode]

Amazing! Thanks for sharing! Will try it out!

[bmadcode]

@dracic @fabb - I just created an issue for this to replace slash commands with another character - will have to see what is easy to type and will work across most instances. Bummer slash will not work as its so common.

[bmadcode]

Closed - I set up claud code and saw all of the starter characters they use - so slash commands replaced with * which seems like a save choice, and still easy to type/remember IMO @dracic @fabb

[ksylvan]

* worked for me in VS Code GitHub Copilot too

[ksylvan]

I just loaded up the BMAD IDE Agent in VSCode just using GitHub Copilot. I am using Claude Sonnet 4 as my agent.

I love it!

I started a project and used the Web agent to re-architect it. Then I started a chat with the scrum master persona.

I asked it about the state of the project. A real open-ended question: "Let's see the state of the current story. What's happening with it?"

It looked at my story file (which was all about project scaffolding and setup, and I had manually completed it and marked it as done) and it came back with

I then asked "Can you double check all the AC's are actually met?"

And it meticulously checked all the acceptance criteria. This is so awesome!

[ksylvan]

@bmadcode Brian, one more note: when the SM agent went to create the next story, it looked for the epic files where it expects them (right in docs/) and they weren't there because I have moved those files around with my simplified exploded docs/ setup, but that didn't stop it!

It read the docs/index.md and searched around till it found the Epic 1 file and then it continued.

It generated the 1.2.story.md file, and because I have the markdownlint extension installed, it did one more surprising thing:

And then it double-checked its work against the checklist. Beautiful!!!

[bmadcode]

very nice! I wonder if it will reliable always do that, what I can do to simplify and not have a breaking change is instruct the SM to check both places, might optimize the performance a bit.

I am working on a massive template standardization - once its out and applied and non breaking, going to roll in the doc exploder.

Going with this for MD templating for LLM standard:

• {{placeholder}} = Simple text replacement placeholder
• [[LLM: instruction]] = Instructions for the LLM (not included in output)
• <<REPEAT: section_name>> ... <> = Repeating section
• ^^CONDITION: condition_name^^ ... ^^/CONDITION: condition_name^^ = Conditional section that will render if the condition_name logically applies
• example = Example single line or multiline content for llm single or multishot guidance - do not render