PM can RUIN EVERYTHING

[igorshunov]

U had an incident just now. Ask PM about inserting a story in an epic. PM just recreated all epic and story structure from scratch and thus I have losted in the project, because it is impossible to reconcile the real status of what was done with the new structure.
Given that the framework requires WAY more work from a human than is needed (operating the implementation cycle, for example - switching between agents etc) this means couple of a human (not AI) life is wasted!

[igorshunov]

Sorry I mean "I had an incident"

[igorshunov]

I really think that BMAD would be invaluable with an orchestrator who never forgets the general context of the projects and is responsible for transmitting artifacts / requests / escalations between subagents

[dracic]

I bet this was in Gemini CLI? It is not PM's fault if a ide and model of your choice deals with tool calling fails in a strange way. But we could give some instructions on not doing risky stuff with existing documentation.

[igorshunov]

It was Qwen :) I got your point, and but on the other side as far as I understand the framework is intended to be agnostic, so would it be nice to have integrity protection measures which will work everywhere like automatic git commit after every change of any artifact?

[dracic]

So i't practically Gemini CLI. i had the same experience. When the Gemini CLI has a failed attempt at using the edit tool, it forces itself to use sed and various workarounds, eventually trying to recreate the entire document, which regularly results in overwriting half of it. I tried providing instructions in GEMINI.md to stop this behavior, but it seems "baked into" how the tool itself utilizes its functions. The only safety measures are checkpoints and commits. We are not at the stage where we can trust LLMs unconditionally.

[igorshunov]

@dracic thank you for the insight!
What is your preferred IDE then (apart from Clode Code)?