Guide · Agent skill
How to use the replica-diff agent skill: clone parity checker
Updated · written with AI from the project's own documentation and checked against our catalogue data
replica-diff is an agent skill, a folder of instructions and small Python tools that an AI agent such as Claude can load and follow. It measures how close your clone of an existing app is to the original, using a feature checklist score and a screenshot layout comparison. This guide explains what the skill does, how to install it, how to ask for it in plain language, and what to check before you trust the results.
Replica Diff: Clone Parity CheckerJakeschincariol/replica-skill · MIT Download the skill Quick start
- 1Download and place the skill. Download the ZIP from the item page and unzip it into ~/.claude/skills/ for all projects, or into .claude/skills/ inside one project. You can also upload it in the Claude apps under Settings, Capabilities, Skills.
- 2Fill in the feature matrix. Create replica/features.csv and mark every row yes, partial with a note, no, or skip with a reason.
- 3Capture matching screenshots. Save the original screens in replica/screens/ and the clone screens in replica/clone-screens/, at 1440x900 or 390x844 in the same state.
- 4Ask the agent for a parity check. Say something like 'how close is my clone to the original' or 'what's missing'. The agent runs parity.py and imgdiff.py.
- 5Review the report and fix the top gaps. Open replica/parity.md, check the verdict and the top five things to build next, then work through the missing must-haves first.
What the replica-diff skill makes the agent do
The skill checks parity, meaning whether your clone does the same job as the original. It is not a visual copy check. The main number is a feature parity score: can a user do everything they could do in the original.
It also runs a layout comparison that checks whether information sits in the same places and hierarchy. It deliberately ignores colour, so a rebrand does not count against the clone. After scoring, the agent writes a report at replica/parity.md with a verdict and the top next steps.
- Feature parity score from replica/features.csv, weighted by must (3), should (2) and could (1)
- Missing features listed in build order, with each area shown weakest first
- Layout score per screen from screenshot comparison, ignoring colour
- Behaviour differences between the two apps, written as flow, original does, clone does, fix or keep
- A verdict: not shippable, shippable, or better than the original
Install the replica-diff skill
Download the ZIP for the skill from the item page, then choose one of these routes. For a personal skill available in all your projects, unzip it into ~/.claude/skills/. For a skill that only applies to one project, unzip it into the .claude/skills/ folder inside that project.
If you use the Claude apps instead, open Settings, go to Capabilities, then Skills, and upload the skill there.
The skill folder contains three files: imgdiff.py, parity.py and SKILL.md. The Python tools need Python 3 only. The source says they use the standard library and need no pip installs.
- Requires Python 3 on your machine
- No pip installs are needed
- Runs scripts: yes. Uses network: no
Prepare your files before you ask for a check
The skill expects a feature matrix at replica/features.csv. You fill it in. Each row's clone column should be yes, partial (with a note), no, or skip (with a reason). Rows you added that the original does not have are left out of the score.
For the layout check, you need screenshots of the original and the clone at matching viewport sizes and in the same state: same data shape, same tab open, logged in the same way. Use 1440x900 for desktop and 390x844 for mobile. Original screenshots go in replica/screens/ and clone screenshots go in replica/clone-screens/. Only use originals from public pages or your own account.
- replica/features.csv: your feature matrix
- replica/screens/: original screenshots
- replica/clone-screens/: clone screenshots at the same viewport and state
Trigger the skill with natural requests
You do not need to type a command to use the skill. The skill's description lists phrases that should trigger it, so you can ask in ordinary words. Phrases named in the skill include 'how close is my clone', 'compare it to the original', 'what's missing', 'parity check', 'diff the screens' and 'is it ready'.
If you want the tools run directly, the skill lists these commands. The first gives the feature score and missing list. The second compares one pair of screenshots and writes a diff image.
- Ask: 'How close is my clone to the original?'
- Ask: 'What's missing from the clone? Give me the missing list in build order.'
- Ask: 'Diff the screens for S07 and tell me where the layout differs.'
python3 parity.py replica/features.csv
python3 imgdiff.py replica/screens/S07.png replica/clone-screens/S07.png --out replica/diffs/S07.pngRead the scores and the verdict
The layout score uses bands: matches at 90 and above, close at 75 and above, partly at 50 and above, and different below that. The layout mode reports the regions that differ, in the original's pixel coordinates, biggest first. It also reports a height difference if the clone's page is much longer or shorter.
The verdict follows fixed rules. Not shippable means a must-have is missing or an S1 bug is open. Shippable needs all must-haves done, a feature score of 80 or higher, and no open S1 or S2 bugs. Better than the original means shippable plus fixes from replica-entrepreneur. The skill says to give honest numbers, so a clone at 62% is reported as 62%.
- Layout bands: matches 90+, close 75+, partly 50+, different
- Pixel mode is for your own regressions, such as clone today against clone last week, not against the original
Safety notes and limits
The skill runs scripts, so read the files before you run them if you are unsure what they do. It does not use the network, according to the source. The tools only read the files you give them and write outputs such as the report and diff images.
The skill is not for pixel-perfect copying, and it is not for judging visual quality or brand design. Colour is ignored on purpose, and the tool measures structure only. Do not use it on apps you have no legitimate access to.
- Runs scripts: yes. Uses network: no
- Do not compare apps you have no legitimate access to
- Measures structure and feature coverage, not visual quality
What people use it for
- A developer rebuilding an existing app who wants an honest completeness score before deciding what to build next.
- A founder checking whether a clone is ready to ship by looking for missing must-have features and open S1 or S2 bugs.
- A designer comparing whether a rebuilt screen keeps the original's layout and information hierarchy while ignoring a new colour scheme.
- Regression checks of your own clone over time, using pixel mode to compare today's screenshots with last week's.
Frequently asked questions
How do I install the replica-diff agent skill?
Download the ZIP from the item page and unzip it into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder. In the Claude apps, you can instead upload it under Settings, Capabilities, Skills.
What do I need to run replica-diff?
You need Python 3, since the tools use only the standard library and no pip installs. You also need a replica/features.csv file you fill in, and screenshots of the original and clone at matching viewport sizes.
Is replica-diff free to use?
The item is listed under the MIT licence. Check the licence file in the download for the full terms before you use or redistribute it.
Does replica-diff work with a pixel-perfect copy?
It is not designed for that. The skill says not to chase the original's exact look, and its layout mode ignores colour. Pixel mode exists only for your own regression checks, not for comparing against the original.
Why does my layout score ignore colour?
The layout comparison looks at edge maps and where things sit on the page. Colour is removed on purpose so that a rebrand does not lower your score.
Why is my clone marked not shippable?
The verdict is not shippable if any must-have feature is missing or an S1 bug is open. Check the missing list in build order, which shows the must-haves first within each area.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
replica-diff suits developers, founders and designers who are rebuilding an existing app and want an honest, structure-based completeness score. It is a poor fit if you want pixel-perfect copying, visual or brand judgement, or if you cannot legitimately capture screenshots of the original.
Download the skill