What problem does it solve? Golden image tests in Flutter fail whenever rendering legitimately changes, and blindly regenerating reference PNGs destroys the visual evidence needed to review whether the change is acceptable. This Skill provides a controlled, manual-only runbook for re-baselining Shed Book's eight golden images without losing the diff evidence or corrupting baselines with wrong fonts or platforms. ## Core Features & Use Cases - Verify-before-regenerate workflow: Runs make goldens first to produce four failure images per changed golden (master, test, isolated diff, masked diff) before any regeneration, since --update-goldens overwrites unconditionally. - Environment precondition checks: Confirms macOS host, exact Flutter version from .fvmrc, real bundled font loading via FontLoader in test/flutter_test_config.dart, and the TolerantFileComparator at 0.005 tolerance. - Review and commit discipline: Requires inspecting every changed image for legibility (no Ahem boxes, readable tag numbers), deleting test/features/failures/, and landing the PNGs alone in their own commit and PR. - Use Case: After a design-system change moves several golden images, use this Skill to verify the diffs are intended, regenerate the baselines on the pinned Flutter version, and ship the new PNGs as a standalone reviewed commit. ## Quick Start Ask the agent to re-baseline the golden images after confirming which intended change moved them, and it will verify, inspect, regenerate, and commit the PNGs separately.
