Screen Auditor — Pixel-Level iOS QA Loop for Claude Code

Proof — with vs without (real finds from the build session, all from rendered pixels): (1) the app's swipe shell was showing a DIFFERENT screen than the selected tab — the nav underline said "feed" while chats rendered. Code review had passed it; one screenshot pair exposed it, the agent root-caused it to a layout-centering bug in the pager math, replaced the pager, and a re-sweep confirmed the fix on screen. (2) A header row was wrapping words mid-word ("schoo l / peopl e") — caught in a routine sweep, fixed with a no-wrap rule. (3) A full-bleed photo hero was silently clipped to the safe area by its container. None of these are visible in source code; all of them are one glance at a screenshot. That's the product: the agent takes that glance, on every screen, after every build. Screen Auditor is a complete visual-QA loop for iOS apps developed with Claude Code: a sweep script that builds your app, installs it on the simulator, walks every screen via deep links and screenshots each one; a session prompt that makes the agent READ every screenshot against a 15-point judging rubric (nav↔content match, overlap, clipping, wrapping, safe areas, dead space, empty states), fix confirmed defects, rebuild, and re-sweep until two consecutive sweeps come back clean; plus pointer rules for tap-driving (calibration, settle-click, when NOT to touch the mouse) and an XCUITest skeleton for pointer-free flow driving. For iOS developers using Claude Code who are tired of "looks fine to me" reviews shipping rendered bugs. Requires a Mac with Xcode + a simulator.