More screenshots
Works with the AI Agents you already use
Next.js + Supabase + Vercel Audit: Code, QA and Costs
Pre-launch security audit for Next.js + Supabase + Vercel: up to 13 review roles return verified findings with file:line evidence, fixes and cost drivers.
$14
Next.js + Supabase + Vercel Audit: Code, QA and Costs
Example session with this skill installed
Audit ~/projects/tidepool, full mode. Here's my Vercel usage for the last 30 days: Image Optimization - Transformations 41,880 / 5,000 included, Function Invocations 212K, Fast Data Transfer 38 GB.
- Read your context and instructions
- Compiled the next js
Full mode on tidepool (Next.js 15.3.2, App Router, 15 source files). 12 roles ran as parallel subagents; skipped qa-regression: no previous report.
critical 3 | high 7 | medium 17 | low 10 · verification: 37 of 37 re-checked in a clean context, 37 confirmed, 0 dropped.
Top 5 of 10
- SEC-1 | critical | OpenAI key in a NEXT_PUBLIC_ var | S | rotate the key, move the call to an authenticated server action
- SEC-2 | critical | deleteEntry without auth, via service role | XS | getUser() + author_id filter, drop the admin client
- DB-1 | critical | invoices table without RLS | S | enable RLS + owner-scoped select policy
- SEC-5 | high | cron route without CRON_SECRET | XS | Bearer CRON_SECRET check, 401 otherwise
- FV-1 | high | image optimizer open to any host | XS | restrict remotePatterns to the storage host and path
Usage mapping: 41,880 transformations come from FV-1, the open hostname (next.config.ts:6), and FV-2, two formats x 11 widths (next.config.ts:8-9). Biggest lever: lock remotePatterns, then one format. Current prices: see the Vercel pricing page.
Example finding
DB-1 | critical | invoices table has no row level security
- location: supabase/migrations/20260815120000_billing.sql:1
- evidence:
create table public.invoices ( - why it matters: anyone holding the public anon key can read every invoice (customer_email, amounts) through the REST API
- fix:
alter table public.invoices enable row level security;plus a select policy scoped to the workspace owner - effort: S · confidence: confirmed · roles: database, security
Files: audit-reports/audit-tidepool-2026-10-09-1730.md and .html, next to the repo.
Connects securely to your tools. The creator never sees your data.
About this skill
This skill audits an existing Next.js + Supabase + Vercel codebase the way a small review team would: five developer roles, five QA roles and three FinOps roles, each with a precise checklist.
It detects your stack first and skips roles that don't apply. In Claude Code the roles run as parallel subagents; on claude.ai they run in grouped passes.
Every finding has a severity, file:line, the quoted code, a fix and an effort estimate. Suspicions without evidence go to a separate "could not verify" list. A final verifier re-opens the cited lines in a clean context and drops anything the code doesn't show.
Typical catches: tables without RLS, service-role keys in client components, secrets in NEXT_PUBLIC_ variables, server actions and cron routes without auth, wide image remotePatterns, missing indexes, N+1 queries, mobile Safari pitfalls, missing alt text.
Paste your Vercel or Supabase usage page and the FinOps roles map each metric to the code behind it. The skill never quotes prices; it names the billing driver. Give it last month's report and it shows what was fixed, what is still open and what came back.
Quick mode is fast and still thorough: the 5 core roles that catch the costliest problems (security, database, backend, frontend, Vercel costs), with every finding re-checked by the verifier. In a blind test on a repo the skill had never seen, quick mode caught 9.5 of 10 planted issues and full mode 10 of 10, with no false alarms. Full mode adds the QA, accessibility, i18n and remaining FinOps roles. Token cost, measured honestly on a small test repo: quick used about 0.5M subagent tokens and full about 1.1M. Larger repos cost more, so check your plan's limits before a full run.
Output: a top-10 fix list, a markdown report and a self-contained HTML report. Read-only: it never edits code, runs migrations, calls any API or writes and runs scripts, and nothing from the skill is saved on your machine.
After handover, if the client sends a list of small changes, Change Request Triage tells you which items are warranty fixes and which are paid.
- Roles: 5 dev (frontend, backend, database, security, DevOps), 5 QA (functional, edge cases, mobile and browser, accessibility and i18n, regression), 3 FinOps (Vercel, Supabase, third-party APIs).
- Modes: quick is fast and thorough on the 5 core roles (security, database, backend, frontend, Vercel costs), every finding verified; full runs all roles that apply. On a repo over about 150 source files with no mode given, it runs quick first and offers full.
- Token cost: about 0.5M subagent tokens for quick and 1.1M for full, measured on a small test repo; larger repos cost more.
- The skill has SKILL.md, 13 role prompts plus a verifier, stack detection, the finding format, usage mapping, synthesis rules, an HTML report template and a fictional sample report; your agent loads it all from Agensi as text.
- Single-user license. Static review only: it doesn't replace a penetration test, and a short report doesn't prove there are no other issues.
- Best on App Router projects with migrations in the repo; without migrations the database checks become runtime checks.
How to use
- 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
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 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
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 1 month ago
Permissions
Read-only on your code. It reads the repo files with Read, Grep and Glob, and uses the shell only for read-only commands (ls, cat, grep, find, wc, git log/show/diff/ls-files/rev-parse) to read local git history. It never edits, formats, installs, builds, runs dev servers or migrations, runs package managers or npm audit, never writes or runs scripts, and never calls Vercel, Supabase or any other API. It writes two report files, markdown and HTML, directly with the agent's file-writing tool, in a folder you name or in audit-reports/ next to the repo, never inside it, and never overwrites an earlier report. Nothing from the skill itself is ever written to disk. Any key or token found in evidence is masked after its first 4 characters.