You can’t unit test for all the aspects that make up taste, it’s true. But if you break off parts of that - eg. by looking at what is codified out there as “go

You can’t unit test for all the aspects that make up taste, it’s true. But if you break off parts of that - eg. by looking at what is codified out there as “good” design, what’s considered best practice etc - you can create tools the agent can call on that let it get critiques of its own work. What’s really cool about this is those tools can be code, written by agents and committed to your repo. Put together a script that for example makes sure your brand colors are enforced (eg. https://github.com/cadamsdotcom/CodeLeash/blob/main/scripts/check_brand_colors.py https://github.com/cadamsdotcom/CodeLeash/blob/main/scripts/...) and then put it in your pre-commit checks (https://github.com/cadamsdotcom/CodeLeash/blob/main/.pre-commit-config.yaml#L40 https://github.com/cadamsdotcom/CodeLeash/blob/main/.pre-com...), and the agent will get feedback on its use of tasteless defaults and adjust accordingly (partly because you blocked commits that contain said tasteless defaults!)