The visible feature was only the surface
Visual inconsistencies are often reported from one screenshot without provenance, comparison points, responsive states, or a systemic root cause.
The system also had to respect concrete constraints. Screenshots must be tied to route, viewport, time/build, and state. Responsive/a11y behavior needs multiple widths and interaction states. An audit request alone does not authorize implementation.
My role covered agent workflow design and automation engineering, with responsibility for agent workflow, validation automation, and packaging.
I turned the constraints into boundaries
Agent audit/repair workflow; consistency reference defining evidence, severity, dependency order, and verification expectations. Packaging supplies portable host installation.
Compare equivalent elements side-by-side before generalizing.
Audit shared tokens/primitives before local pages.
Repair shared causes before adding component overrides, trading a broader root fix for less design drift.
The key decision: Compare equivalent elements side-by-side before generalizing.
The implementation had to prove the model
I delivered the work across the full path: Defined screenshot provenance and viewport/state coverage. Codified severity and finding fields. Designed dependency-ordered audit and repair sequencing. Packaged cross-host installation and contract tests.
The result is concrete: Agents can produce reproducible UI findings. Distinguish observed defects from inferred causes. Propose the smallest systemic fix. Recheck affected viewports, states, motion, focus, tests, and builds when implementation is requested.
The senior engineering lesson was that solving the visible workflow is only half the job. The architecture must also make constraints, failure modes, evidence, and ownership explicit enough for the next change to remain safe.