albertirgi/replica is a coding-agent skill that reconstructs an interface from a reference site, a screenshot, or a screen recording. It can be used in Claude Code and other coding agents. According to the repository description, it scores the replica on four axes — layout, style, pixels, and motion — and provides a workflow for improving it until it passes the criteria. Release information for a specific version is not in the given material.
The tool converts a rendered page into a specification containing design tokens, per-element boxes, and computed styles, then compares it with the replica. Scoring results are split across the four axes, and the impact of the items to fix along with the expected score gains are recorded in a report and in machine-readable material. The comparison command ends with a failure code if it does not reach the configured criteria. The README states that the comparison uses a virtual clock and fixed execution conditions, and that it terminates when the criteria are met or after at most 8 iterations. The structure helps an agent repeatedly inspect and fix the differences between the reference screen and the replica.
The repository contains test material for the replica and the reference, along with its own test scripts, so the basic validation flow can be checked. However, how accurate the general visual comparison really is cannot be independently verified from the README description alone. What can therefore be confirmed from this repository is the procedure for comparison and iterative improvement; the given material offers no basis for asserting that the results are accurate across a wide range of screens.