← Back to list
⚡

replica-launch Checked

Writes the launch for an app clone: the landing page built on the positioning angle, pricing set …
⚡ Productivity skills By Jakeschincariol Version v1.0.0 Added 2026-10-12 Updated 2026-10-12 Source ↗
9.1Overall rating
0Views
0Downloads
1.9kStars

Installation

🤖 Install via AI

Copy the prompt below and send it to your AI assistant (e.g. Claude Code) — it will follow the instructions and install automatically.

Install the "replica-launch" skill by following the instructions at https://skill123.me/install/replica-launch.
⌨️ Command line install

Run in your terminal — downloads and installs to ~/.claude/skills/.

curl -fsSL https://skill123.me/install/replica-launch.sh | bash
📦 Download ZIP

Download the zip and extract it into your skills directory (e.g. ~/.claude/skills/), then restart your session.

⬇ Download v1.0.0 · 6 KB
📊 Static rating
9.1/10
Trigger
7.5
Structure
8.0
Workflow
9.6
Content
10.0
Engineering
10.0
Security
10.0

Six-dimension rubric with reviewer rationale.

View scorecard ↓
🛡 Dynamic test
—

Not live-tested yet — this skill is rated on static analysis only.

About this skill

Writes the launch for an app clone: the landing page built on the positioning angle, pricing set against the original's public pricing and its users' complaints, and the App Store and Google Play listing linted against the stores' limits and copycat rules. Use when the user says "landing page", "pricing", "how much should I charge", "App Store listing", "store screenshots", "launch plan", "Product Hunt", or after /replica-brand.

replica-launch

Reads replica/fixes.md (the angle and the evidence), replica/brand.md,

and replica/brand.json. Writes replica/launch/: landing.md,

pricing.md, listing.json, launch-plan.md.


python3 listing.py replica/launch/listing.json      # limits, claims, the original's name

Start from listing.example.json in this folder.

Documentation

Writes the launch for an app clone: the landing page built on the positioning angle, pricing set against the original's public pricing and its users' complaints, and the App Store and Google Play listing linted against the stores' limits and copycat rules. Use when the user says "landing page", "pricing", "how much should I charge", "App Store listing", "store screenshots", "launch plan", "Product Hunt", or after /replica-brand.

replica-launch

Reads replica/fixes.md (the angle and the evidence), replica/brand.md,

and replica/brand.json. Writes replica/launch/: landing.md,

pricing.md, listing.json, launch-plan.md.


python3 listing.py replica/launch/listing.json      # limits, claims, the original's name

Start from listing.example.json in this folder.

The rules

  • **No fake proof.** No invented testimonials, user counts, star ratings,

press logos or "trusted by" rows. An empty proof section is better than a

fake one. Real beta users, with permission, are fine.

  • **No reviewer quotes as testimonials.** The original's reviews are research.
  • **The original's name stays out** of your app name, store listing, keywords

and ads. Apple and Google reject it, and it invites a takedown.

Step 1: landing page

landing.md, section by section, then build it in the project:

1. **Hero**: the angle as a headline (what you do, for whom), one line under

it, one button. Screenshot of the real product.

2. **The problem**: the top two "hate" themes, in plain words. Paraphrase,

do not quote reviewers.

3. **How it works**: three steps, from the core flow.

4. **Features**: each tied to a fix from fixes.md. Lead with the fixes,

not the parity features. Parity is the price of entry.

5. **Pricing**: the table from step 2.

6. **FAQ**: the real objections, including "can I import from {{category}}

tools?" if you built an importer.

7. **Final call to action.**

Copy in the brand voice. Run replica-design's contrast check on the page.

Step 2: pricing

pricing.md:

  • The original's public pricing page and 2 or 3 alternatives, in one table,

with the URL and the date read. Prices change, so the date matters.

  • What reviewers said about price and billing, from feedback.md, with counts.
  • The model: free plan or trial, per seat or flat, monthly and annual

(annual usually 2 months free).

  • Three tiers at most. Name them by who they are for.
  • Fix the billing complaints in the product: one-click cancel, clear renewal

emails, no surprise per-seat jumps.

  • The Stripe products and prices to create (the user creates them).

Step 3: store listing

Fill listing.json, then:


python3 listing.py replica/launch/listing.json
View full documentation

It checks App Store limits (name 30, subtitle 30, promotional text 170,

keywords 100, description 4000) and Google Play's (title 30, short

description 80, full 4000), counts characters the way a person does, flags

ranking and price claims in short fields, emoji, wasted keyword characters,

and **any name from avoid** anywhere in the listing. Exit 1 on an error.

Apple's App Review Guideline 4.1 (Copycats) rejects apps that simply copy a

popular app or make minor changes to another app's name or UI. Your fixes

and your own brand are what get you through review, so make them visible in

the screenshots and the first lines of the description.

Also prepare: screenshots at the current required sizes (check App Store

Connect and Play Console when you upload, they change), privacy labels and

the data safety form, privacy policy and support URLs, age rating, and review

notes with a demo account the reviewer can use.

Step 4: launch plan

launch-plan.md: a waitlist or beta list before launch, analytics and error

tracking live, where the original's unhappy users talk (the Reddit threads

and communities from the research), a Product Hunt or Hacker News post that

leads with the fix, and the first 10 users to talk to by hand.

Output

replica/launch/ complete, the landing page built, listing.py passing.

Next: /replica-deploy.

Score breakdown

Trigger
7.5
What (landing page, pricing, store listing, launch plan) and When (eight quoted phrasings plus 'after /replica-brand') clear; good keyword spread (landing/pricing/charge/store screenshots/Product Hunt); no explicit When-Not.
Structure
8.0
101-line SKILL.md plus listing.py and listing.example.json; script and example template at skill root rather than scripts/ and assets/ (role mixing); no redundant docs; example file usage explained.
Workflow
9.6
Four ordered steps with hard integrity guards (no fake proof, no reviewer-quote testimonials, original's name excluded from name/keywords/ads) and a deterministic linter usable as a gate (exit 1, trademark detection with word-boundary matching, claim regexes, keyword waste analysis); pricing step requires dated sources; verification closed by 'listing.py passing' plus replica-design's contrast check on the landing page; user creates all Stripe products (credential safety).
Content
10.0
Tight imperative prose; copyable commands with arguments; concrete store limits listed inline; section-by-section landing structure directly actionable.
Engineering
10.0
Valid frontmatter; script has shebang, argparse usage, stdlib only; listing.example.json referenced and present; sibling skills (replica-brand, replica-design, replica-deploy) exist in the family.
Security
10.0
Capability manifest: reads replica/fixes.md, brand.md, brand.json and listing.json; writes replica/launch/* and landing/pricing copy; agent may read the original's public pricing page (declared scope); script is pure local linting with no network, no credential access, no persistence, no dynamic execution. Capabilities match declared scope.
Launch-prep stage of the replica-* suite: landing page on the positioning angle, pricing benchmarked against the original's public pricing, and an App Store / Google Play listing linted by a bundled stdlib Python linter (store field limits, human-style character counting, ranking/price claims, emoji, wasted keywords, and any 'avoid' name - i.e. the original's trademark - anywhere in the listing, exit 1 on error). Strong anti-fabrication rules (no invented testimonials or 'trusted by' rows, reviewers' words are research not copy) and real regulatory awareness (Apple Guideline 4.1 copycats, Google claim bans). Same structural nit: listing.py and listing.example.json at skill root instead of scripts/ and assets/; no When-Not in description.