replica — clean-room cloning pipeline

Upstream: Jakeschincariol/replica-skill, MIT, commit 77c9436, installed 2026-10-22. License text: LICENSE.txt. Tools are standard-library Python, invoked as python. All eleven skills and their tools were verified at install time against the upstream test suite (57 tests, green); that suite asserts the upstream repository layout and is not shipped here.

Choosing a part

Stage	Part	It writes
1. Reverse-engineer the target: screens, flows, components, inferred data model, feature matrix	parts/replica-recon/	replica/recon.md, replica/features.csv
2. Stack, database schema, routes, build order	parts/replica-architect/	replica/architecture.md
3. Design system as tokens and component specs, with original assets	parts/replica-design/	replica/design/tokens.json
4. Rebuild screen by screen from the recon map	parts/replica-build/	the app itself
5. Auth, migrations, payments, email, jobs, integrations	parts/replica-backend/	backend code + security checklist
6. Test every flow, log bugs by severity	parts/replica-test/	replica/test-plan.md, replica/bugs.md
7. Score parity and diff screenshots	parts/replica-diff/	parity score, missing list, diff.png
8. Read real reviews, rank complaints, pick a positioning angle	parts/replica-entrepreneur/	replica/reviews.csv, fix plan
9. New name, palette, logo brief, voice, sweep the old name out	parts/replica-brand/	brand kit, sweep report
10. Landing page, pricing, store listing	parts/replica-launch/	replica/launch/
11. Preflight, production, DNS, ship	parts/replica-deploy/	replica/deploy/

Order

recon → architect → design → build → backend → test → diff → entrepreneur → brand → launch → deploy

Each part reads the file the previous one wrote from replica/ in the project root. Any part also runs on its own: replica-entrepreneur on an app you have not started is a cheap way to decide whether cloning it is worth it.

Tools

python skills/replica/parts/replica-diff/imgdiff.py original.png clone.png --out diff.png
python skills/replica/parts/replica-diff/parity.py replica/features.csv
python skills/replica/parts/replica-entrepreneur/reviews.py replica/reviews.csv
python skills/replica/parts/replica-design/contrast.py replica/design/tokens.json
python skills/replica/parts/replica-brand/sweep.py . --avoid "Original App"
python skills/replica/parts/replica-launch/listing.py replica/launch/listing.json

imgdiff.py compares edge maps, so a rebrand's colours do not count as a diff; it reports which regions moved, in the original's pixels. parity.py weights must/should/could, counts partials as half, and refuses to call a clone shippable while a must-have is missing. sweep.py blocks a deploy that still carries the original's name, domain or colours — including inside identifiers.

Rules

• Features and flows, clean-room. Never the target's source, proprietary assets, logo, trademark, copy or private APIs. Read the artifact; write everything fresh.
• Only what you are allowed to read: public pages, store listings, help docs and the owner's own account. No scraping behind a login, no paywalls, no one else's account. If the account's terms forbid building a competitor, stay on public sources.
• Rebrand before launch. New name, palette, logo, words; sweep.py must be clean before replica-deploy ships.
• Numbers, not vibes. Parity comes from parity.py, layout from imgdiff.py; reviews only from rows with links, and thin themes are labelled thin.
• Sizing first. replica-recon states honestly how big the clone is before the build starts; open-ended scope here is a fork worth ask_questions.
• In this kit, the visual and device sides stay with the existing skills: reverse-engineering for APK/decompilation work, browser for phone and panel capture, design for tokens, ship for deploys.

Related

reverse-engineering (artifact analysis), browser (capture on device and panel), design (design systems), review (critique), ship (release mechanics).