Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Vibe-Coded App Checkup: Leaked Keys, RLS, Payments

    1

    This skill finds the ones a machine can find, asks you about the rest, and writes one report.

    Free

    Security scanned
    Vibe-Coded App Checkup: Leaked Keys, RLS, Payments

    Vibe-Coded App Checkup: Leaked Keys, RLS, Payments

    Example session with this skill installed

    이 앱 이대로 공개해도 돼? 점검해줘. (Is this app OK to launch as it is? Check it.)

    • Read your context and instructions
    • Compiled the vibe-coded app checkup

    Summary

    Ran on a test repository with risks planted on purpose (2 commits, 23 files). 20 failures, 11 warnings. Every secret below is shown as its first 4 characters only.

    What I will check (shown first)

    Keys in files and the whole git history · browser-exposed variables · RLS and Firebase rules · payment amounts · admin and debug routes · what code cannot show · the risk grade. Read-only.

    scan.mjs output (excerpt, real run)

    바이브코딩 앱 점검 — dirty
    커밋 2개 · 파일 23개 · 0.1초
    
    실패 20 · 경고 11
    
    9. [실패] Supabase service_role 키 — eyJh****(219자)
       scripts/seed.mjs:2
       히스토리 — 커밋 1개에 있음
    11. [실패] Stripe 라이브 비밀 키 — sk_l****(38자)
       빌드된 번들에 들어 있다 — 사이트를 여는 사람 누구나 내려받는다.
    16. [실패] 브라우저로 나가는 변수에 비밀값 이름 — NEXT_PUBLIC_SUPABASE_SERVICE_ROLE_KEY
    18. [실패] RLS가 켜지지 않은 테이블 3개 — orders · notes · regranted
    20. [실패] Firebase 규칙이 누구나 쓰게 열려 있다
    29. [경고] 결제 금액을 요청에서 받아 그대로 쓰는 것 같다
    30. [경고] 관리자 경로에 인증 검사가 안 보인다 (2곳)
    

    (실패 = fail, 경고 = warning, 히스토리 = git history.)

    Other failures in the same run: private key block, GitHub token, AWS access + secret keys, Anthropic and OpenAI keys, Google OAuth secret, Stripe webhook secret, Toss live secret key, a committed .env, a committed key file, .gitignore missing .env*, a using (true) policy, a hard-coded password comparison.

    Grade (example answers to the four questions)

    The grade comes from your answers, so this part is illustrative. With these answers: web app · stores member emails · one-off card payments · the maker keeps running it, grade.md gives

    B. What B requires before launch includes a privacy policy, RLS with owner-only policies, server-side amount check and webhook verification, admin 2FA, and a tested backup.

    Do now

    1. Revoke and reissue the service_role, Stripe live, AWS and AI keys in each console (you, not the agent).
    2. Move the service key out of NEXT_PUBLIC_, redeploy, then enable RLS on orders · notes · regranted with real policies.
    3. Only after revoking: clean the history and add .env* to .gitignore.

    The full report goes outside the repository, by default next to it as <repo>-checkup/report.md.

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

    About this skill

    Apps built fast with AI coding tools tend to ship with the same few holes: a service key committed once and deleted later (still in git history), a secret variable named NEXT_PUBLIC_..., a table created without row level security, a payment amount the browser decides, an admin route nobody locked. This skill finds the ones a machine can find, asks you about the rest, and writes one report.

    What it does

    1. Shows what it will check (seven lines) and confirms the folder. Read-only: it changes nothing.
    2. Runs scan.mjs (Node, no dependencies) over the working tree and the whole git log -p --all:
      • Keys: Supabase service_role (decodes the JWT role claim, so the public anon key is not flagged), Supabase secret keys, Stripe, Toss Payments, OpenAI, Anthropic, AWS, Google, GitHub, Slack, private key blocks, .env and key files that were ever committed, secrets inside built bundles.
      • Browser-exposed variables (NEXT_PUBLIC_ / VITE_ / EXPO_PUBLIC_) with secret-like names.
      • SQL migrations: tables without enable row level security (including RLS enabled inside do $$ blocks), using (true) policies, views that bypass RLS. Firebase rules open to everyone.
      • Heuristics (warnings): payment amount taken from the request, admin routes with no visible auth check, open debug routes, hard-coded password comparisons.
      • Every secret is printed as its first 4 characters + ****. Never in full.
    3. Human checklist for what code cannot show: backups and a restore drill, whose name the accounts are in, privacy policy, webhook signature checks, personal data in logs, 2FA on admin.
    4. Grade (P / A / B / C) from four questions: how it is delivered (web, PWA, extension, desktop, native), whose data it holds, whether money moves, who runs it after launch. Each grade lists what must be done before going public.
    5. Report saved outside the repo (so a public repo never carries a map of its own holes): summary, what to do now, grade, findings, what it could not see.
    6. Fixes only when you ask. Key rotation happens in your provider console; the skill gives the order (revoke, redeploy, check logs, then clean history) and never logs in for you.

    Rules it follows

    • Never says "safe". Zero findings means "nothing in what this scan covers".
    • Does not call your live services to test whether a key still works.
    • Cites only the Korean privacy and e-finance law articles listed in grade.md; the report says it is not legal advice.

    Language

    Instructions, checklist and report template are in Korean (the grade includes Korean privacy and payment rules). Your agent can write the report in any language.

    Source: https://github.com/Stophyuk/solutionzip-skills

    Cannot see what lives only in dashboards (tables or policies created in the Supabase UI, live environment variable values) or payment / admin logic split across many files; those go to the report's 'could not see' section. Heuristic checks are warnings, not verdicts. Instructions are in Korean and the grade uses Korean privacy and payment rules.

    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 any payment.

    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

    Free forever

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 16 days ago

    • Free with an Agensi account

    Requires

    Node.js 18+
    git

    Permissions

    Terminal / Shell
    Read Files
    Write Files
    Listed16 days ago

    What's inside

    Frequently Asked Questions

    No. It only reads the repository and its git history. It does not test keys against live services. It fixes things only when you pick them, and key rotation stays in your own provider console.

    Only the first 4 characters plus ****, with the file and line. The report is saved outside the repository so it never gets committed next to the code it describes.

    No. It decodes the JWT and only fails on role service_role. The anon key is meant to be public; what protects your data is RLS, which it checks separately.

    They come from whose data the app holds and whether money moves through it, from a prototype with no real users (P) up to apps handling identity numbers or holding funds (C). Each grade lists what must be done before going public. Delivery type and who runs it after launch add tasks.

    New in Code Quality & Review