Guys, so, while specs and steering are pretty damn good and mimic best practices that the community ended up gravitating towards, there is an issue in which, from my understanding, the bigger the context is, the less likely it is for Kiro to actually obey your rules. I'm using "rules" as the collective of documents from specs, hooks and steering.
So, here comes a few thoughts in regards to how things work or what could be useful as further practices and tools, and I may be wrong, this is intended as a discussion:
-
Is there a way for us to define the importance of each element inside our rules? Let's say that I have some good coding practices, and also the rule of "before creating a new file, ask for my permission", by definition the latter rule is a much higher level of importance, but it can get ignored due to the context size;
-
Does LLM always use the whole context of the rules for everything it does, or it can ignore something if it judges it as not relevant?
-
Wouldn't it be better if we could create a "Director" markdown whose overseers which markdown file ruleset should be used in the context? Let's say that I want it to simply commit to GitHub, it doesn't need to see the whole ruleset, it just needs to see the file structure and the conversation context. Or, for example, I'm just working specifically in the layout/css part, it doesn't need the whole coding practices, just the design document...
What are your guys thoughts on this? Are those issues that have already been solved otherwise or are they indeed hurdles that require solutions? Do you have any suggestions for those problems?