Adom UI Design System

@@ -1,6 +1,6 @@ {   "slug": "adom-ui-design",-  "version": "2.1.4",+  "version": "2.1.5",   "type": "skill",   "description": "The Adom UI Design System plus a bundle of focused UI skills (tooltips, floating-windows, toasts, responsive-ui, app-shell, ai-driven-apps, clickable-paths, dark-default, input-feedback, settings-layout). One install drops the whole set of skill files onto your container, plus shared icon assets (the settings cog) so every Adom app draws the same glyph.",   "dependencies": {},

@@ -217,6 +217,28 @@ Deleting your hand-rolled panel is the part that breaks pages, not mounting the    happened to be right while the install path was broken. `curl` your own route back and    compare its sha256 to the pack's. +### ⛔ A GREEN CHECK IS WORTH NOTHING UNTIL YOU HAVE SEEN IT GO RED++Before you trust any check you just wrote, **break the thing it is checking and watch it+fail.** If it still passes, it was never testing what you thought.++This caught two instruments in one day, both belonging to careful people:++- lbr wrote a regression check for a panel-state fix. It went green. The negative control+  exposed why: with a fresh profile its graduation policy kept the panel open regardless, so+  the check passed for a reason unrelated to the fix, and **reverting the fix still passed.**+  Rewritten to graduate the user first, so the policy actively wants the panel folded and only+  a durable recall can keep it open, the control fails cleanly and the real build passes.+- lbr's own `js-call-check.py` needed two rounds of its own false positives fixed before it+  was usable, and its adopters are told to run the negative controls rather than trust a green+  first run.++The same shape sank three other things here: a truncation instrument that measured the wrong+element and reported 0 of 6 truncating on a panel visibly cutting words off; a conformance+checker that failed every correct panel; and a fleet audit that reported an app had no panel+when it had one. **Every one of them ran cleanly and returned a confident answer.** A check+that has never been observed to fail is a claim, not evidence.+ ### ⛔ A PRE-CHECK ON YOUR OWN MACHINE IS A DIAGNOSIS, NOT THE PROOF  Run selfCheck wherever you like while you work. It only counts from **inside ah, on the