Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    Green Light: Go/No-Go Launch Audit for AI-Built Apps

    1

    Green Light finds those and tells you, in plain English, what to do.

    $19

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    Green Light: Go/No-Go Launch Audit for AI-Built Apps

    Green Light: Go/No-Go Launch Audit for AI-Built Apps

    Example session with this skill installed

    Is my app ready to launch?

    • Read your context and instructions
    • Compiled the green light
    • Generated the document

    About the title: Green Light's files still use its working name, Launch Check, so the report below is titled "Launch Check" and the report file is named launch-check-report.md.

    Example report. A real run of Launch Check (Claude Code, Sonnet) on "bookclub", a sample Lovable-style app we built with problems planted on purpose. All keys in it are fake. The only edit: one factual slip corrected by hand. Stripe's minimum card charge is $0.50, so "1 cent" was changed to "50 cents".

    Launch Check: Bookclub

    2026-11-07 · Checked: the code in bookclub/ (no live site checked, no URL given) · Built with: Lovable (Supabase database, Stripe payments, OpenAI summaries, Google Analytics)
    Assumed: you will have real customers paying with real cards, and some visitors from the EU/UK. Tell me if that's wrong and I'll re-check.

    Verdict: 🔴 NOT READY TO LAUNCH

    Right now a stranger could read every member's email, give themselves Premium for free, pay 50 cents instead of $9, delete anyone's messages, and run up your OpenAI bill. Your live Stripe key and database master key are also exposed. This is fixable in about a day, mostly with the copy-paste prompts below plus a few steps only you can do in the Supabase, Stripe and OpenAI dashboards. One week is realistic, but do the fixes in the order below.

    6 to fix before launch · 3 to fix in your first week · 2 look good · 4 I couldn't check

    🔴 Fix before launch

    1. Your most powerful keys are exposed (do this first)

    What's wrong: Three secret keys are in the project. (a) The Supabase "service role" master key, which skips all database protections, is named VITE_SUPABASE_SERVICE_ROLE_KEY, and anything starting with VITE_ is shipped to every visitor's browser. (b) Your OpenAI key (sk-p…) is also VITE_ so it is in the browser too. (c) Your live Stripe secret key (sk_l…) is in .env, and .env is saved in git (it is in the first commit) and is not in .gitignore.
    What could happen: Anyone who opens the site's developer tools could download or wipe your whole database, spend your OpenAI money, and, if the code is ever shared (for example on GitHub), take control of your Stripe account's payments.
    Where: .env:3, .env:4, .env:5; used in src/lib/admin.ts:3-6 and src/lib/ai.ts:6; .gitignore ·

    How sure: Confirmed (I decoded the Supabase key's role without printing it: it is service_role)

    You need to do this yourself: Deleting the keys from the code is not enough, because they have already been in the code. They must be replaced:

    1. Supabase → Project Settings → API → reset/rotate the service role key (and rotate the JWT secret if you want to be thorough).
    2. OpenAI → API keys → delete the key and create a new one.
    3. Stripe → Developers → API keys → roll the secret key.
    4. Put the new OpenAI and Stripe keys in Supabase → Edge Functions → Secrets, never in the code. The service role key should not be needed in the app at all after fix 5.
    5. If the project is on GitHub, check whether the repository is public. Either way, the old keys are now treated as burnt.

    Paste this into Lovable:

    Security fix: VITE_SUPABASE_SERVICE_ROLE_KEY and VITE_OPENAI_API_KEY are public (VITE_) variables, so they are sent to every visitor's browser, and .env is committed to git. Delete src/lib/admin.ts and every use of the service role key in the front end. Move all OpenAI calls (src/lib/ai.ts) to a Supabase Edge Function that reads OPENAI_API_KEY from Edge Function secrets, and have the browser call that function. Remove .env from git tracking, add .env and .env.local to .gitignore, and leave only the Supabase URL and anon key as public variables. Keep the app working exactly as it does now. List the files you changed and tell me how to confirm no secret appears in the browser.
    

    How to check it's fixed: Open the live site, press F12 → Sources (or Network), search for service_role, sk- and sk_live. Nothing should be found.

    2. Anyone can read member emails, give themselves Premium, and post as anyone

    What's wrong: The profiles table (names, emails, and the is_premium flag) has no row-level security, so it is open to anyone who has the public key, which is in the browser. The messages table lets anyone insert any message with any author (with check (true)).
    What could happen: A stranger could download every member's email address, flip is_premium to true for themselves, and post chat messages pretending to be someone else.
    Where: supabase/migrations/20261001000000_init.sql:1-7 (no enable row level security for profiles), :29 (post messages … with check (true)) ·

    How sure: Confirmed in the migration (a later dashboard change could differ, so also check the Supabase Security Advisor)

    Paste this into Lovable:

    Security fix for the Supabase database: enable Row Level Security on public.profiles. Policies: a signed-in user can select only their own row (auth.uid() = id) and update only display_name on their own row; users must NOT be able to change is_premium or email (only server code using the service role may change is_premium). On public.messages, replace the "post messages" policy so inserts require auth.uid() = author_id and that the user is signed in, and add delete policies so a user can delete only their own messages. Remove any USING (true) / WITH CHECK (true) on write operations. Write this as a new migration, keep the app working for normal users, and explain how to test with two accounts.
    

    How to check it's fixed: Make two accounts in two browsers. As account A, try to change is_premium on your own profile from the browser console (it must fail), and try reading account B's profile (must return nothing).

    3. Anyone can pay 50 cents, and anyone can get Premium for free

    What's wrong: The checkout takes the price from the browser, so the browser decides the amount. Separately, Premium is switched on by the browser itself when someone opens the /success page, with no proof of payment, and there is no Stripe webhook at all.
    What could happen: A visitor could pay $0.50 for Premium, or just visit /success and get Premium without paying at all.
    Where: supabase/functions/create-checkout/index.ts:4-7 (amount from request), src/pages/Success.tsx:7 (browser sets is_premium), src/pages/Upgrade.tsx:5; Stripe is live mode (sk_l…) ·

    How sure: Confirmed

    You need to do this yourself: In Stripe, create the Premium product at $9 and note its Price ID. Add a webhook endpoint pointing at the new function and put its signing secret (whsec_…) in Supabase Edge Function secrets as STRIPE_WEBHOOK_SECRET.

    Paste this into Lovable:

    Payment fix: the checkout in supabase/functions/create-checkout/index.ts uses the amount sent from the browser, and src/pages/Success.tsx grants Premium from the browser. (1) Change create-checkout so it ignores any amount from the browser, requires a signed-in user, and uses a fixed Stripe Price ID read from the environment variable PREMIUM_PRICE_ID, with the user's id in client_reference_id. (2) Remove the is_premium update from Success.tsx; it should only say "Thanks, activating…" and read is_premium from the database. (3) Add a new Edge Function stripe-webhook that verifies the Stripe signature with the raw body and STRIPE_WEBHOOK_SECRET, and on checkout.session.completed sets profiles.is_premium = true for that user using the service role on the server. Keep the same $9 plan. Tell me how to test with Stripe test mode and the Stripe CLI.
    

    How to check it's fixed: In Stripe test mode, change the amount in your browser's request: checkout must still charge $9. Open /success without paying: Premium must stay off.

    4. Anyone can use your OpenAI account without logging in or paying

    What's wrong: Book summaries are generated straight from the browser with your OpenAI key (gpt-4o), with no login check, no Premium check and no limits.
    What could happen: One script could run thousands of requests overnight and leave you with a large OpenAI bill, and free users get Premium's main feature for free.
    Where: src/lib/ai.ts:2-13 · How sure: Confirmed

    You need to do this yourself: In OpenAI → Settings → Limits, set a monthly spending limit and an email alert.

    Paste this into Lovable:

    Cost protection: summarizeBook in src/lib/ai.ts calls OpenAI from the browser. Replace it with a Supabase Edge Function that requires a signed-in user, checks on the server that profiles.is_premium is true, limits each user to 20 summaries per day (stored in the database), caps the notes input at 4000 characters, and reads OPENAI_API_KEY from Edge Function secrets. Return a clear message when the limit is hit. Keep the feature working for Premium users.
    

    How to check it's fixed: Call the summary feature while logged out, or on a free account: it must be refused.

    5. The moderator page is only hidden, not locked

    What's wrong: Only one email address (lluc.test@example.com) sees the delete button, but that check happens in the browser, and the delete uses the master key, which every visitor has (see item 1). Anyone can delete any message.
    What could happen: A visitor could delete every message in every club.
    Where: src/pages/Moderator.tsx:3, src/lib/admin.ts:7-9 · How sure: Confirmed

    Paste this into Lovable:

    Security fix: the moderator page and message deletion are only hidden in the interface. Add a role column (e.g. profiles.is_moderator, not editable by users) and a database policy so only moderators can delete other people's messages, with users still able to delete their own. Remove the hard-coded email check in Moderator.tsx and the service role client. The moderator delete must go through the normal signed-in client under that policy. Keep the moderator page working for me.
    

    How to check it's fixed: As a normal user, try deleting someone else's message from the browser console: it must fail.

    6. No privacy policy while you collect emails, payments and analytics

    What's wrong: The app collects emails and names, takes card payments through Stripe, sends book notes to OpenAI, and loads Google Analytics, but there is no privacy policy page anywhere. The footer only says "Made with love".
    What could happen: Stripe, Google and the app stores can require a privacy policy, and launching without one breaks privacy laws in the EU, UK and California.
    Where: src/components/Footer.tsx:2, index.html:6-12 · How sure: Confirmed

    Paste this into Lovable:

    Add a Privacy Policy page at /privacy, linked from the footer and the sign-up form. It must describe what this app collects (email, display name, club messages, payment status via Stripe, book notes sent to OpenAI, usage analytics via Google Analytics), the services that process it (Supabase, Stripe, OpenAI, Google Analytics), how long data is kept, how users can get a copy or delete their data, and a contact email <your email>. Plain language, marked as a draft for me to review. Don't change anything else.
    

    (This is a starting draft, not legal advice.)

    How to check it's fixed: Click "Privacy" in the footer on the live site and read it.

    🟡 Fix in your first week

    1. Google Analytics loads before anyone agrees

    What's wrong: Google Analytics loads on every visit with no consent banner.

    What could happen: EU/UK visitors have not consented, which breaks cookie rules.

    Where: index.html:6-12 ·

    How sure: Confirmed
    Paste this into Lovable:

    Add a cookie consent banner. Don't load Google Analytics (index.html) until the visitor clicks "Accept"; "Reject" must be as easy as "Accept" and keep it off. Remember the choice and add a "Cookie settings" link in the footer.
    

    2. No terms, no account deletion, no contact address

    What's wrong: No Terms page (refunds, cancellation), no way to delete an account, no contact email.

    What could happen: Refund disputes with no written rules, and no way to honour deletion requests.

    Where: src/ has none of these ·

    How sure: Confirmed (for the files I could see)
    Paste this into Lovable:

    Add a Terms page at /terms (refund policy for the one-time $9 Premium purchase, acceptable use for club chat), a "Delete my account" button in settings that deletes the user's profile, messages and clubs on the server after a confirmation step, and a support email <your email> in the footer. Mark the Terms as a draft for me to review.
    

    3. Everyone can read every club's messages

    What's wrong: Anyone, even logged out, can read all messages in all clubs.

    What could happen: If clubs are meant to be private chats, strangers could read them.

    Where: supabase/migrations/20261001000000_init.sql:28 ·

    How sure: Likely (fine if clubs are meant to be public)
    Paste this into Lovable:

    Make messages readable only by signed-in members of the club. Add a club_members table (club_id, user_id) with row level security, join/leave functions, and change the select policy on public.messages so a user can read a message only if they are a member of its club. Keep public clubs listable.
    

    🟢 What looks good

    Clubs table: row-level security is on, and only the owner can edit a club; being publicly readable looks intentional.

    Settings: the Stripe return links use https://bookclub.app/…, not localhost, and nothing in the code tells Google to stay away. (Only the bookclub.app domain assumption needs confirming.)

    ⚪ What I couldn't check

    Login code and page protection: the export doesn't contain src/main.tsx, the router or the login pages.

    Check it yourself (2 minutes): while logged out, open /upgrade, /moderator and any club page and confirm they ask you to log in.

    Supabase dashboard settings: I can't see redirect URLs, email confirmation or backups.

    Check it yourself: Supabase → Authentication → URL Configuration should list your real domain, not localhost; Settings → Database should show backups on; also open Advisors → Security Advisor and fix anything it flags.

    • Spending caps: Check yourself: set limits in OpenAI (see item 4) and keep an eye on Supabase usage.

    The live site: you gave no URL. After fixing, send it and I'll re-check the built site (and whether the old keys are still visible in the browser).

    Scorecard

    AreaResult
    Keys and passwords🔴
    Who can see or change your data🔴
    Logins and accounts🔴 (admin page only hidden)
    Payments🔴
    Privacy and legal pages🔴
    Settings and launch mistakes🟢 (⚪ dashboard settings)
    Surprise bills and abuse🔴
    Live siteNot checked: no URL given

    Words used in this report

    • Service role key: Supabase's master key that ignores all database protections. It must never be in the browser.

    Row-level security (RLS): database rules that decide which rows each user may see or change. A table without them is open to anyone with the public key.

    • VITE_ variable: a setting whose value is sent to every visitor's browser.
    • Webhook: a message Stripe sends your server to prove a payment really happened.
    • Edge Function: a small piece of server code on Supabase where secrets can stay hidden.

    This is a pre-launch check done by an AI assistant reading your project, not a professional security audit or legal advice. It catches the common, costly mistakes; it cannot promise there are none left. Re-run it after big changes.

    Success.tsx

    PDF · document

    Generated

    Example file from a real run - the skill writes it into your workspace.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    You built an app with AI. It works. But is it safe to put in front of real customers? Most AI-built apps that get into trouble fail in the same few places: a secret key shipped to every visitor, a database anyone can read, a "Premium" button that unlocks without paying, no privacy policy. Green Light finds those and tells you, in plain English, what to do.

    Ask "is my app ready to launch?" and you get one report:

    • A verdict: 🔴 not ready, 🟡 ready with a fix list, or 🟢 ready, decided by clear rules, not vibes
    • Every problem ranked, with what could actually happen ("a stranger could download every customer's email") instead of jargon
    • A copy-paste fix prompt for each problem, written for Lovable, Bolt, Replit, v0, Cursor or whatever you use
    • A two-minute test to prove each fix worked, so you don't have to take the AI's word for it
    • What it couldn't check, listed honestly, with steps to check it yourself. A database setting it can't see is never assumed safe

    Seven areas checked

    1. Keys and passwords (in code, in the browser, in old commits)
    2. Who can see or change your data (Supabase row level security, Firebase rules, storage buckets, one user reading another's data)
    3. Logins and accounts (unprotected pages and routes, admin areas only hidden in the browser, plain-text passwords)
    4. Payments (prices set by the browser, paid features without payment, unverified webhooks, test mode left on)
    5. Privacy and legal pages (privacy policy, cookie consent, terms, account deletion, children's data)
    6. Settings and launch mistakes (debug mode, localhost links, hidden from Google, error pages)
    7. Surprise bills and abuse (open AI, email and SMS endpoints, no spending caps)

    Built to be safe

    • Read-only: it writes only the report, and fixes things only if you ask
    • Never prints your secret keys
    • Knows which keys are public on purpose (Supabase anon key, Firebase web config, Stripe publishable key), so no false alarms about them
    • Optional live-site check uses ordinary page loads on your own site only

    Tested on five deliberately broken sample apps (Lovable-style Supabase, Next.js + Firebase, Next.js + Supabase, Replit Express + SQLite) with blind runs in Claude Code: it found the planted problems, gave the right verdict, and never leaked a key.

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Security.

    A pre-launch check by an AI assistant that catches the common, costly mistakes. It is not a professional penetration test, certification or legal advice, and it cannot promise nothing is left. Apps handling health, financial or children's data need a specialist review. It cannot see settings that live only in a dashboard (for example Supabase rules); it says so and tells you where to look. The live-site check is optional and limited to ordinary page loads of your own site. Tested end to end in Claude Code; other agents not yet tested end to end.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before buying a skill.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 3 days ago

    Nothing to install. It runs in your AI agent.

    Optional
    Git

    Permissions

    Network Access
    Read Files
    Write Files
    Terminal / Shell

    Terminal or shell is optional: read-only searches (grep, git log, git ls-files) and an optional package audit. Files: reads project files; writes only launch-check-report.md. Network is optional: only ordinary page loads of your own site when you give a URL. Services: none.

    Listed3 days ago

    What's inside

    Frequently Asked Questions

    Yes, that's who it's for. Open your project in your AI tool (or export it from Lovable/Bolt to GitHub and open it in Claude Code, Cursor or Codex) and ask "is my app ready to launch?". The report explains everything in plain words.

    No. It only reads your project and writes a report. It fixes things only if you ask, and gives you copy-paste prompts for your builder.

    It's a pre-launch check by an AI assistant that catches the common, costly mistakes. It is not a professional penetration test, certification or legal advice, and it cannot promise nothing is left. Apps handling health, financial or children's data need a specialist review; it will tell you so.

    If you give it your URL and your tool can load web pages, it does a few gentle checks (HTTPS, exposed files, keys in the downloaded code, cookie banner). Ordinary page loads only, your own site only.

    It says it couldn't see them, treats that as "not ready" if your browser talks to the database directly, and tells you exactly where to look in the dashboard.

    Yes, use both. Green Light also covers payments, privacy pages, settings and costs.

    New in Security