replica-diff measures a clone against the original with two standard-library Python tools. parity.py reads the feature matrix and weights must, should and could rows; imgdiff.py compares two PNG screenshots by layout and ignores colour, so a rebrand does not count against you.
It then adds a behaviour comparison of the flows and writes a report with a verdict: not shippable, shippable, or better than the original.
What You Get
- A feature score with must-haves done of total, areas weakest first and a missing list in build order
- A layout score per screen, with the regions that differ in the original’s pixel coordinates
- A diff image per screen and optional combined score of 80 percent features and 20 percent layout
- A behaviour comparison of clicks, error handling, remembered state and emails per flow
- A verdict of not shippable, shippable or better than the original
Common Use Cases
- Finding which must-have features of a clone are still missing
- Comparing a clone’s screens to the original’s after a rebrand changed every colour
- Setting a pass mark for CI with the fail-under option
What Makes It Unique
It defines parity as doing the same job, and both scripts leave out what would distort the number: skipped rows and features you added are not scored, and the layout diff works on edge maps so colour changes do not count. Pixel mode is for your own regressions only.
Under The Hood
Both scripts are standard-library Python with no network calls. parity.py reads the CSV and optional JSON diff results; imgdiff.py decodes PNG files itself and can write a diff image. The skill’s instructions have the agent run them on your project files, collect screenshots and write the parity report, so it runs commands and writes files. The original’s screenshots come from public pages or your own account.