THIS IS EPIC Game Changer for Gemini - BMad Agent with Canvas!

[bmadcode]

Would love some feedback on this exciting idea - The most annoying thing about the GEM is it feels Google is going to abandon it or keeps removing or forgetting to add features to it - unfortunately as its one of my fav's features from early on. But I realize, its unnecessary now that Gemini Pro support Github Attachment - its now easy to just use a github project as your GEM and get full access to all of the full gemini chat features, such as deep research and Canvas!

I am going to streamline this - but all I did was take the web agent build output - create a new github repo with nothing but those files in it, and prompted Gemini (non gem) chat with 'You are the BMad Agent - read and embody the instructions in the github projects agent-prompt.md - and load all other files from the repo into your core memory for full operational instructions'

This needs more testing of course - but this is really exciting. And what is also great about this is that updating the web agent is as easy as updating the github repo of the web build

[getpat]

That is excellent, gemini is a good partner but the best would be SuperGrok, I will have to give this a try with it. It is a much better writer than Gemini.

[bmadcode]

Have not tried it, ill add it to the backlog haha! If you can get similar working and have good luck and the context is good - please let us know @getpat !

[bmadcode]

I just had the analyst simulate a random project, assume collaboration etc, just to get the document out - this is gonna be epic for working through the larger documents! I envision a whole new workflow based around this new ability to use canvas - such as:
First output with as much is known as possible the first document, the allow the user to select sections they want to iterate on potentially by selecting or annotating in canvas (need to experiment here with what's possible).

[getpat]

Something else to consider in this workflow, specifically with Augment - Augment agents have been announced these subagents are task bots controlled by the main bot, with proper modification BMad could attach these agents to the project task list and it could agentically complete the tasks for approval within the workflow

[bmadcode]

@getpat do you mean the remote agent feature thats coming, or is this something else? If you have a link I'll take a look - I did find this, not sure if its what you are referring to https://www.augmentcode.com/blog/introducing-remote-agent

Cursor also has added a new similar background agent and i think has cloud agents also coming.

If I understand your idea - you could be talking to the orchestrating agent, and have it dispatch the SM agent to draft the story and similar for the dev all from a single chat? and potentially even put it a bit more on autopilot if you want to see how far it can get through the whole iterative cycles on its own potentially - with the integration of some QA and testing agents into the mix!

[getpat]

@bmadcode You just got email from me on another subject. Yes, is the answer augment.new waitlist for augment remote - sounds like a perfect match.

[getpat]

In general I would say you could generate a PRD pass it to the project manager they get an approved task list and the tasks get passed to the coder, the coder creates separate branches for each task and build the task, they get merged back after the workflow - potentially including a code review and testing - gets folded by human in the loop back to main. You have a agentic workflow runner.

[bmadcode]

Thanks @getpat Ill take a look later today for the email! What you are describing is a very similar pipeline I have been experimenting a bit with outside of this speciific project for work using some different technologies, its showing promise. Basically multipe agent orchestrating workflows, wether local or remote hosted. Combined with cloud coding agents that are emerging from basically every provider over the last 2 weeks (codex, gh copilot, google's, augments, cursors, a few others), this is gonna get powerful fast. I think using them with the power of having agents before them break down the tasks simply to what BMAD Method can do will really empower them (or generally just providing at least for now simple coding tasks to the final coding agents.)

[dracic]

I was thinking in that direction when we brainstormed #28. My prompt worked with the BMAD repo attached to a gem. Of course, I experimented with this setup too, i just pasted my prompt to a chat with your repo attached. I liked the Canvas feature with the BMAD method. Some things are missing now when initiating chat from a gem, but I have a good feeling that will change soon. In my humble opinion, a GitHub repo in a gem is too significant a feature to make me feel that gems are obsolete for Google. Maybe we should just wait some time before making fundamental changes? The best setup for the future would be a minimal setup prompt in a gem that would load the orchestrator from attached github repo. That way, you would have the latest version of BMAD on each chat initiation, whether it's Brian's or your own customized version pushed to your own repo. It is also best, context-wise. Activating an agent would load only what needs to be loaded from a repository.

[bmadcode]

@dracic good point - or at least Ill say who knows with google haha - im still salty they got rid of Google Reader about 15 years ago that I put tons of time into LOL.

But - I agree, I will not make it is not GEM optimized, was planning to streamline some parts - what I can do it make it canvas optional - so if user indicates canvas is avaialble it will still work in sections (or not based on yolo preference) but do it either as chat output or in the canvas.

[dracic]

We can even leverage GitHub MCP for ide workflow, online and offline, there is no need for some exclusive BMAD MCP. I'll make a showcase.

[fabb]

And who knows, maybe gemini web will get support for MCP too at some point

[bmadcode]

They announced it for their API on the 20th, hope it comes to the web version soon - wonder if that would require a desktop app, which would be nice.

[bmadcode]

@dracic interesting idea - I have not tried the github mcp yet since I have been able to just ask the agent to commit for me and it does fine without it, along with other basic commands.

[fabb]

I saw that you can add custom tools in google AI Studio