I was developing a new project using kiro and start experimenting on steering. now here is that I notice
-
when you mention the steering like make them understand ofcourse they need to read it right but if kiro failed to generate like select model feature and lets say I click the same model. kiro then proceed to re read the steering file which is a good thing. but would that be possible to let kiro know at the first of conversation ? because repetitng reading the steering use also token especially when the steering is long, kiro need to re understand it again which burn token and re read the task which burn token and then read what its current taks which burn token and start doing its taks which is sometime it burns a lot of token
-
when you have existing steering file and you decide to update it, kiro will become wild card more where it vibe code every thing dont follow the steering which is frustrating because every conversation. kiro need to read the steering files. im against to the thing that you need to click the refine just to make kiro understand that we have new steering which infact kiro needds to understand that we change something on it and we start the conversation it should read the file since kiro promote automation
-
only specs read the steering file but if you do the vibe code it will never read the steering unless you told so.
-
Agent hook is kinda meh for me. not gonna lie, there is no repetive task if I combine specs and steering. also the most scary part for this agent hook is the file on save, its a token eater, you see I have this existing giant project and I use the agent hook to hear every file save, its say I ask kiro to do prettier and linting check so there is no need to do the husky work at pushing the branch. but after I just ask kiro to change stuff after the command I reach the limit for today. so the big projects are frustrating to kiro because there is a lot of happening, Agent hook is like a token eater that eats token