What problem does it solve? When replacing an application without changing its specification, teams need to verify that the new implementation matches the legacy one beyond what test suites cover—spacing, colors, fonts, borders, and content parity. This Skill detects those differences with deterministic tools and uses the LLM only to classify each detected diff, avoiding unreliable visual eyeballing. ## Core Features & Use Cases - Deterministic three-path detection: Compares current and new baselines via pixel diffing, computed-style trait comparison, and aria snapshot structural comparison, with no LLM involvement in detection. - LLM triage of detected diffs only: Presents cropped diff-region image pairs one at a time for classification into actionable, acceptable, or environmental noise, with user approval required before marking anything acceptable. - Convergence gating and handback: Judges convergence as zero unexplained diffs plus zero unfixed regressions, attributes diffs blocked by unimplemented sibling features, and hands actionable diffs back to parity-replace or files issues via issue-create. - Use Case: After parity-replace makes the new side green for a feature, run parity-diff against a target environment to catch visual regressions the suite cannot see, classify them, and iterate until convergence. ## Quick Start Ask the AI to run parity-diff for a feature slug against a chosen new-side target environment, for example by requesting to detect and classify current-versus-new differences for the order-list feature on the preview target.
