Design System Consistency
The problem with consistency is that no single violation is worth stopping for. A colour two hex values off is invisible in isolation. A one-off spacing value saves you four minutes today. But a year of those and the system is a suggestion rather than a system, and every new project starts with an archaeology dig to figure out which grey is the real grey.
Skynet holds your system in memory — tokens, type scale, component rules, the brand guidelines PDF nobody reads — and checks work against it on request. It is a reviewer, not an autocorrect. It surfaces the near-miss and the reason it matters, and you decide whether it is a mistake or a deliberate exception. Some exceptions are the good part of the work.
How it works
Teach it your system
Connect the brand guidelines, the token list, and the component documentation. Skynet builds one memory of your system, including the parts that only exist as a paragraph in a doc rather than a variable in a file.
Review new work against it
Point it at a file, a page, or a deck. It reports what does not match: colours off-token, type outside the scale, spacing that breaks the rhythm, components used against their documented rules.
Separate drift from intent
Every flag comes with the rule it broke and the source of that rule. You mark the deliberate ones as exceptions, and the agent remembers, so it stops re-flagging your good decisions.
Keep the docs honest
When the system genuinely changes, the agent drafts the update to the documentation so the written system and the built system stop diverging.
Build it from a prompt
Describe your system once and give it something to check.
You get a system that stays a system without one person having to be the full-time enforcer of it. The taste calls remain yours — the agent is only responsible for noticing, which is the part humans are worst at and machines are good at.