SAMPLE Sample report. Built from InvoicePond, an invented v0-style app written by Nordik Audit with planted problems. Not a customer and not a real company. A limited audit, not a full one, as the report says under "What we checked, and who checked it".

App Audit technical report: v0-invoicepond

Full details

This is the technical report: the full record of every finding the audit made, for a developer or for your AI tool. The plain report is the short version, with what to fix first and a fix prompt for each item. It groups problems of the same kind into one item, so its number of things to fix is smaller than the number of problems counted here; the table below shows which finding sits in which item. The plain report lists items 7 to 13 in short; their fix prompts, checks and steps are in the section "Fix prompts for the other items" below.

Summary: 16 distinct problems (19 confirmed findings, 3 of them the same root cause as another) among 21 candidate findings: 8 high, 3 medium, 5 low, each problem counted once; 2 rejected. Most severe: C001 "NEXT_PUBLIC_SUPABASE_SERVICE_ROLE_KEY is exposed to the browser and looks like a secret (Supabase s…" (high); C016 "Direct dependency geist pulls in 3 vulnerable packages (next@15.2.9, postcss@8.4.31, sharp@0.33.5);…" (high); C017 "Direct dependency next pulls in 3 vulnerable packages (next@15.2.9 itself, postcss@8.4.31, sharp@0.…" (high); 13 more problems are listed below.

Checked against the code by an AI reviewer; each critical or high code finding was checked again by a second AI reviewer. Each finding below says how its verdict was reached. "Confirmed" means the evidence is present at this revision. For secret-shaped text it means only that the text is there: no credential was tested against any live service.

How the plain report labels each finding

The plain report puts findings of the same kind in one item and labels each item by what could happen to your users: a settled finding where strangers can read or change other users' data, act as another user, or move money is "Fix today" whatever its rating here; a package update is at most "Fix this week"; anything else follows its rating. The ratings below are the audit's own and are not changed.

The plain report lists 13 things to fix (its items). Together they hold 16 problems, grouped by kind; the same problem at another place is not counted again.

FindingRatingPlain itemPlain urgencyWhyPlace
C001high1Fix todaystrangers can read or change other users' datalib/supabase/admin.ts line 5
C002high2Fix todaystrangers can read or change other users' dataapp/api/invoices/[id]/route.ts line 8
C003high3Fix todaystrangers can read or change other users' dataapp/api/invoices/route.ts line 24
C019high4Fix todaystrangers can read or change other users' dataapp/api/invoices/route.ts line 6
C020high5Fix todaystrangers can read or change other users' datalib/supabase/middleware.ts line 28
C006high6Fix todaystrangers can read or change other users' dataapp/dashboard/invoices/actions.ts line 9
C004high7Fix todaystrangers can move money with fake payment messagesapp/api/webhooks/payments/route.ts line 5
C005high7Fix todaystrangers can read or change other users' dataapp/api/webhooks/payments/route.ts line 11
C021medium8Fix todaystrangers can read or change other users' dataapp/api/invoices/route.ts line 7
C009medium9Fix todaystrangers can read or change other users' datasupabase/migrations/001_create_tables.sql line 42
C010medium9Fix todaystrangers can read or change other users' datasupabase/migrations/001_create_tables.sql line 45
C008low10Worth a lookits ratingnext.config.mjs line 17
C011low11Worth a lookits ratingapp/api/invoices/route.ts line 31
C012low12Worth a lookits ratingnext.config.mjs line 12
C016high13Fix this weekpackage update, at most Fix this weekpackage-lock.json line 1527
C017high13Fix this weekpackage update, at most Fix this weekpackage-lock.json line 1733
C018high13Fix this weekpackage update, at most Fix this weekpackage-lock.json line 2634
C014low13Worth a lookpackage update, at most Fix this weekpackage-lock.json line 1065
C015low13Worth a lookpackage update, at most Fix this weekpackage-lock.json line 1087

Fix prompts for the other items

The plain report lists items 7 to 13 in short. Each one is written out here with the same number, in the same way as the items in the plain report: what it means, the fix prompt, how to check it and any steps you do yourself.

Fix today 7. Fake payment messages could change your data
  • What this means for you: Your app accepts messages that say a payment went through, but it does not check that they really come from your payment service. Someone could send a fake message and mark an invoice as paid without paying. It shows up in 2 places in your code. The prompt names each one.
  • How urgent: Fix today
  • How long it takes: usually about 30 minutes (a code change, a secret from your payment service, then the check)

Fix prompt: Paste this into v0's chat as-is. You don't need to understand it.

A security audit of my v0 app found a problem. Please fix it. In `app/api/webhooks/payments/route.ts` (around lines 5 and 11) the route that receives payment messages acts on them without checking that they really come from the payment service. Before doing anything else, check the message's signature with the webhook signing secret from the project's secret settings, and reject any message whose signature is missing or wrong. Any error you send back must be a short general message, never the raw error text. Do not add a SECURITY DEFINER database function. A database rule you add must not let users change status, role, payment or owner fields, or link a row to data of another user. Keep everything else working the same. Then tell me in plain words what you changed.
Show me how to check / step by step
  • How to check it's fixed: Paste this into v0's chat after the fix:
Check the fix you just made for: fake payment messages could change your data. Read the route in `app/api/webhooks/payments/route.ts` (around lines 5 and 11). Tell me whether a payment message with a missing or wrong signature is turned away before anything changes, and whether the signing secret comes from the project's secret settings. Change nothing. Answer Pass or Fail for each place, with one plain sentence why.
  • Your own step: your app needs the signing secret from your payment service, so it can tell real payment messages from fake ones.
    1. In your payment service's dashboard, find the webhook signing secret for your app and copy it (in Stripe: Developers, then Webhooks, then your app's address, then Signing secret).
    2. Paste the prompt below into v0's chat. v0 tells you the name to give it.
    3. Add it under that name in v0: open the project menu (the ... button), then Settings, then Environment Variables. Never paste it into the chat.
I am adding my payment service's webhook signing secret to the project's secret settings myself, so never ask me to paste it here. Tell me the exact name to give it, and make the route that receives payment messages read it from there. When I say it is in, tell me in plain words that the route uses it.
  • Details for a developer: C004, C005 in the technical report
Fix today 8. Someone can see other people's data
  • What this means for you: Your app takes a record number from the link and does not check who owns it. Someone could change the number and see or change another person's data.
  • How urgent: Fix today
  • How long it takes: usually about 15 minutes (a code change in one file, then the check)

Fix prompt: Paste this into v0's chat as-is. You don't need to understand it.

A security audit of my v0 app found a problem. Please fix it. In `app/api/invoices/route.ts` (around line 7) a record is read or changed by the id in the request, without checking who owns it. Before reading or changing it, check that the record belongs to the signed-in user (for example, filter by the user's id), and stop if it does not. What the audit saw: in `app/api/invoices/route.ts` at line 7 the code reads: "client_id: z.string().uuid(),". Any error you send back must be a short general message, never the raw error text. Do not add a SECURITY DEFINER database function. A database rule you add must not let users change status, role, payment or owner fields, or link a row to data of another user. Keep everything else working the same. Then tell me in plain words what you changed.
Show me how to check / step by step
  • How to check it's fixed: Paste this into v0's chat after the fix:
Check the fix you just made for: someone can see other people's data. Read the code in `app/api/invoices/route.ts` (around line 7). Tell me whether a signed-in user could still open or change a record that belongs to someone else, for example by changing the number in the link. What the audit saw: in `app/api/invoices/route.ts` at line 7 the code reads: "client_id: z.string().uuid(),". Fail only if this flaw is still there at that code: someone can see other people's data. That is the code the audit looked at, not the problem itself: a correct fix may leave that line as it is. Change nothing. Answer Pass or Fail, with one plain sentence why.
  • Details for a developer: C021 in the technical report
Fix today 9. Anyone can read a whole table of your data
  • What this means for you: A rule in your database lets every visitor read every row of this table, signed in or not. Someone could copy all of it. It shows up in 2 places in your code. The prompt names each one.
  • How urgent: Fix today
  • How long it takes: usually about 15 minutes (a change to a database rule, then the check)

Fix prompt: Paste this into v0's chat as-is. You don't need to understand it.

A security audit of my v0 app found a problem. Please fix it. In `supabase/migrations/001_create_tables.sql` (around lines 42 and 45) a database rule lets everyone read every row of this table (its row level security policy is "using true"). Fix the rule itself: replace it with one that lets each signed-in user read only their own rows. Do not add a SECURITY DEFINER function, a view or any other way around the database rules. If a page that visitors open without signing in needs one row, give the table a random share_token column and a select rule that returns only the row whose token the page sends, for example `using (share_token = current_setting('request.headers', true)::json->>'x-share-token')`. Keep that page at its current address: do not add, rename or move route folders; the page can read the token from the link's query string (?token=). Any error you send back must be a short general message, never the raw error text. Do not add a SECURITY DEFINER database function. A database rule you add must not let users change status, role, payment or owner fields, or link a row to data of another user. Keep everything else working the same. Then tell me in plain words what you changed.
Show me how to check / step by step
  • How to check it's fixed: Paste this into v0's chat after the fix:
Check the fix you just made for: anyone can read a whole table of your data. Look at the database rules for the table in `supabase/migrations/001_create_tables.sql` (around lines 42 and 45). Tell me whether someone who is not signed in, or another signed-in user, could still read or change rows that are not theirs. Change nothing. Answer Pass or Fail for each place, with one plain sentence why.
  • Details for a developer: C009, C010 in the technical report
Worth a look 10. A setting lets other websites call your app
  • What this means for you: A setting says any other website may call your app with your visitor's sign-in. Most browsers refuse that exact mix today, so the harm now is small, but a small change later could make it real.
  • How urgent: Worth a look
  • How long it takes: usually about 5 minutes (one settings change, then the check)

Fix prompt: Paste this into v0's chat as-is. You don't need to understand it.

A security audit of my v0 app found a problem. Please fix it. In `next.config.mjs` (around line 17) the app lets every other website call it with the visitor's sign-in cookies (CORS allows any origin together with credentials). Allow only your own site's address, and turn off sign-in cookies for anyone else. Any error you send back must be a short general message, never the raw error text. Do not add a SECURITY DEFINER database function. A database rule you add must not let users change status, role, payment or owner fields, or link a row to data of another user. Keep everything else working the same. Then tell me in plain words what you changed.
Show me how to check / step by step
  • How to check it's fixed: Paste this into v0's chat after the fix:
Check the fix you just made for: a setting lets other websites call your app. Tell me which other websites may call my app now with a visitor's sign-in. It should be only my own site. Change nothing. Answer Pass or Fail, with one plain sentence why.
  • Details for a developer: C008 in the technical report
Worth a look 11. Error messages show visitors how your app works inside
  • What this means for you: When something goes wrong, your app sends the full error text to the visitor. That can show details about your database that help someone break in.
  • How urgent: Worth a look
  • How long it takes: usually about 15 minutes (a code change in one file, then the check)

Fix prompt: Paste this into v0's chat as-is. You don't need to understand it.

A security audit of my v0 app found a problem. Please fix it. In `app/api/invoices/route.ts` (around line 31) the app sends the full error text back to the visitor when something fails. Send a short general message instead (such as "Something went wrong"), and keep the full error only in the server logs. Any error you send back must be a short general message, never the raw error text. Do not add a SECURITY DEFINER database function. A database rule you add must not let users change status, role, payment or owner fields, or link a row to data of another user. Keep everything else working the same. Then tell me in plain words what you changed.
Show me how to check / step by step
  • How to check it's fixed: Paste this into v0's chat after the fix:
Check the fix you just made for: error messages show visitors how your app works inside. Find what a visitor sees when the request in `app/api/invoices/route.ts` (around line 31) fails. Tell me if it is a short general message or the full error text. Change nothing. Answer Pass or Fail, with one plain sentence why.
  • Details for a developer: C011 in the technical report
Worth a look 12. Your app does not turn on the browser's safety settings
  • What this means for you: Browsers have safety settings a site can turn on, for example to stop other sites from showing your pages inside theirs. Your app does not turn them on.
  • How urgent: Worth a look
  • How long it takes: usually about 5 minutes (one settings change, then the check)

Fix prompt: Paste this into v0's chat as-is. You don't need to understand it.

A security audit of my v0 app found a problem. Please fix it. In `next.config.mjs` (around line 12) the app sets no security headers. Add the usual ones for every page (Content-Security-Policy, X-Frame-Options or frame-ancestors, X-Content-Type-Options, Referrer-Policy, Strict-Transport-Security), and keep the app working. Any error you send back must be a short general message, never the raw error text. Do not add a SECURITY DEFINER database function. A database rule you add must not let users change status, role, payment or owner fields, or link a row to data of another user. Keep everything else working the same. Then tell me in plain words what you changed.
Show me how to check / step by step
  • How to check it's fixed: Paste this into v0's chat after the fix:
Check the fix you just made for: your app does not turn on the browser's safety settings. List the safety settings (security headers) my pages send now. Tell me if any of the ones the fix added are missing. Change nothing. Answer Pass or Fail, with one plain sentence why.
  • Details for a developer: C012 in the technical report
Fix this week 13. Some packages your app uses have known problems
  • What this means for you: Your app uses old versions of 5 packages (next, postcss, sharp, zod and @supabase/auth-js) that have public warnings about weak spots. We did not check whether your app uses the weak parts. One prompt updates them all.
  • How urgent: Fix this week
  • How long it takes: usually an hour or more (package updates, then a click through your whole app)

Fix prompt: Paste this into v0's chat as-is. You don't need to understand it.

A security audit of my v0 app found a problem. Please fix it. In `package-lock.json` the app uses package versions that have public security warnings. The lowest versions that clear the warnings: next 15.5.28, postcss 8.5.23, sharp 0.35.5, zod 4.6.6, @supabase/auth-js 2.70.0. Install the lowest version listed for each package and stay on the major version the app uses now: do not move to a newer major release (for example from version 15 to 16) only because it exists. Only zod 3.25.76 to 4.6.6 needs a new major version, because no fixed version exists in the major version the app uses now: follow its official migration guide and change the code that uses it. Development-only tools with warnings (such as the build tool) count too: when no fixed version exists in their current major, move them to the lowest fixed major and keep the build working. Do not swap a package for a different one. When you are done, check the packages for security warnings again and make sure none of these, and no new ones, are left. Make sure the lockfile is updated and the app still builds and works. Any error you send back must be a short general message, never the raw error text. Do not add a SECURITY DEFINER database function. A database rule you add must not let users change status, role, payment or owner fields, or link a row to data of another user. Keep everything else working the same. Then tell me in plain words what you changed.
Show me how to check / step by step
  • How to check it's fixed: Also click through your app yourself to see that it still works. Paste this into v0's chat after the fix:
Check the fix you just made for: some packages your app uses have known problems. Tell me which version of each package the app uses now. Each should be at least: next 15.5.28, postcss 8.5.23, sharp 0.35.5, zod 4.6.6, @supabase/auth-js 2.70.0. Then make sure the app still builds. Change nothing. Answer Pass or Fail, with one plain sentence why.
  • Details for a developer: C016, C017, C018, C014, C015 in the technical report

Confirmed findings

A finding that has the same root cause as another is listed directly under that root finding, with a smaller heading, and is counted once.

high C001: NEXT_PUBLIC_SUPABASE_SERVICE_ROLE_KEY is exposed to the browser and looks like a secret (Supabase service_role)
  • Severity: High (lowered from Critical by the second reviewer, who said: Right now only server files import this module, so the key is not in today's browser bundle, but the public prefix leaves one import away from leaking it.)
  • Check families: service-role, secrets (the first is the main one)
  • Location: lib/supabase/admin.ts:5
  • Status: Confirmed (deeper check): the text is present at this revision. It was not tested against any live service.
  • Finding state: Reproduced: the deterministic checks found the same evidence again in the files at this revision. No code was run.
  • Evidence: return createClient(process.env.NEXT_P…(24 chars)!, process.env.NEXT_P…(37 chars)!, {, line SHA-256 c27664c1113b2d4000ea7ea85ab3ccd4d11a355903eecc0efbc8b44642da63fe
  • Verdict reached by: deeper check
  • Deeper check: The admin key, which skips every database access rule, is read from a variable whose NEXT_PUBLIC_ prefix tells Next.js to copy its value into browser code, so anyone who gets hold of the shipped JavaScript can take the key and read or change every customer's invoices through Supabase's public API. (relied on lib/supabase/admin.ts:1-8)

Reproduction:

  1. In a clone of the repository, run git checkout --detach 2d7f14fb67c9c49af0e0dc23dc85b7ad127687ca.
  2. Open lib/supabase/admin.ts at line 5.
  3. Compare with the masked excerpt return createClient(process.env.NEXT_P…(24 chars)!, process.env.NEXT_P…(37 chars)!, { (the line's SHA-256 is c27664c1113b2d4000ea7ea85ab3ccd4d11a355903eecc0efbc8b44642da63fe).

Limitations: Confirmed means only that this text is present at revision 2d7f14fb67c9c49af0e0dc23dc85b7ad127687ca. No credential was tested against any live service, so whether it is real, still valid or already revoked is unknown.

high C016: Direct dependency geist pulls in 3 vulnerable packages (next@15.2.9, postcss@8.4.31, sharp@0.33.5); upgrade target: next 15.5.28, postcss 8.5.23, sharp 0.35.5
  • Severity: High (lowered from Critical by the report's package rule: the public advisory is rated Critical for the package in general, and this audit did not check whether the app uses the weak part, so it is listed as High and still to be updated this week)
  • Location: package-lock.json:1527
  • Status: Confirmed (OSV evidence): the evidence is present at this revision.
  • Finding state: Reproduced: the deterministic checks found the same evidence again in the files at this revision. No code was run.
  • Evidence: geist: next@15.2.9, postcss@8.4.31, sharp@0.33.5, line SHA-256 61e39c93c2a763da9252ba4099fb4a0eef20394358d2d9d6d90dec637346b6ed
  • Verdict reached by: OSV evidence

Reproduction:

  1. In a clone of the repository, run git checkout --detach 2d7f14fb67c9c49af0e0dc23dc85b7ad127687ca.
  2. Open package-lock.json at line 1527.
  3. Compare with the masked excerpt geist: next@15.2.9, postcss@8.4.31, sharp@0.33.5 (the line's SHA-256 is 61e39c93c2a763da9252ba4099fb4a0eef20394358d2d9d6d90dec637346b6ed).

Limitations: The versions locked in the lockfile (package-lock.json, yarn.lock or pnpm-lock.yaml) were matched against a downloaded OSV.dev advisory snapshot, offline, and grouped under the direct dependency whose lockfile dependency graph pulls them in; whether the app uses the vulnerable code path was not checked.

high C017: Direct dependency next pulls in 3 vulnerable packages (next@15.2.9 itself, postcss@8.4.31, sharp@0.33.5); upgrade target: next 15.5.28
  • Severity: High (lowered from Critical by the report's package rule: the public advisory is rated Critical for the package in general, and this audit did not check whether the app uses the weak part, so it is listed as High and still to be updated this week)
  • Location: package-lock.json:1733
  • Status: Confirmed (OSV evidence): the evidence is present at this revision.
  • Finding state: Reproduced: the deterministic checks found the same evidence again in the files at this revision. No code was run.
  • Evidence: next: next@15.2.9 itself, postcss@8.4.31, sharp@0.33.5, line SHA-256 0aa341d4f0523a3cf6dc9ce8ac901d7a59d3aa35589977cdb3910fe7b2e79dfb
  • Verdict reached by: OSV evidence

Reproduction:

  1. In a clone of the repository, run git checkout --detach 2d7f14fb67c9c49af0e0dc23dc85b7ad127687ca.
  2. Open package-lock.json at line 1733.
  3. Compare with the masked excerpt next: next@15.2.9 itself, postcss@8.4.31, sharp@0.33.5 (the line's SHA-256 is 0aa341d4f0523a3cf6dc9ce8ac901d7a59d3aa35589977cdb3910fe7b2e79dfb).

Limitations: The versions locked in the lockfile (package-lock.json, yarn.lock or pnpm-lock.yaml) were matched against a downloaded OSV.dev advisory snapshot, offline, and grouped under the direct dependency whose lockfile dependency graph pulls them in; whether the app uses the vulnerable code path was not checked.

high C018: Direct dependency zod pulls in 1 vulnerable package (zod@3.25.76 itself); upgrade target: zod 4.6.6
  • Severity: High
  • Location: package-lock.json:2634
  • Status: Confirmed (OSV evidence): the evidence is present at this revision.
  • Finding state: Reproduced: the deterministic checks found the same evidence again in the files at this revision. No code was run.
  • Evidence: zod: zod@3.25.76 itself, line SHA-256 56b3b27b0184747f4fd4cb220854973aea2ffa0950fdc1197c1aff12347aa799
  • Verdict reached by: OSV evidence

Reproduction:

  1. In a clone of the repository, run git checkout --detach 2d7f14fb67c9c49af0e0dc23dc85b7ad127687ca.
  2. Open package-lock.json at line 2634.
  3. Compare with the masked excerpt zod: zod@3.25.76 itself (the line's SHA-256 is 56b3b27b0184747f4fd4cb220854973aea2ffa0950fdc1197c1aff12347aa799).

Limitations: The versions locked in the lockfile (package-lock.json, yarn.lock or pnpm-lock.yaml) were matched against a downloaded OSV.dev advisory snapshot, offline, and grouped under the direct dependency whose lockfile dependency graph pulls them in; whether the app uses the vulnerable code path was not checked.

high C002: GET /api/invoices/[id]: no identity check of its own; whether middleware protects it cannot be decided
  • Severity: High
  • Check families: service-role, route-authorization, object-scoping (the first is the main one)
  • Location: app/api/invoices/[id]/route.ts:8
  • Status: Confirmed (AI reviewer and second reviewer (agreed)): the evidence is present at this revision.
  • Finding state: Source-supported: confirmed from a reading of the code at this revision (static evidence only); the deterministic checks did not settle it on their own, and no code was run.
  • Evidence: const { data: invoice, error } = await supabase, line SHA-256 8f0794c44f6289e5da5b0a07bcd2cbbdcd35daad07161c8bdebdf36296b76236
  • Verdict reached by: AI reviewer and second reviewer (agreed)
  • AI reviewer: This GET route uses the admin client with no session check and returns the whole invoice plus the client's email and billing address for any id, and the open invoices select policy lets anyone list those ids. (relied on app/api/invoices/[id]/route.ts:4-18)
  • Second reviewer: Anyone without signing in can call GET /api/invoices/<id> with any invoice id and get the full invoice, its line items and the client's name, email and billing address, because the service-role client reads the row with no identity or owner check and the middleware does not guard /api. (relied on app/api/invoices/[id]/route.ts:4-18)

Reproduction:

  1. In a clone of the repository, run git checkout --detach 2d7f14fb67c9c49af0e0dc23dc85b7ad127687ca.
  2. Open app/api/invoices/[id]/route.ts at line 8.
  3. Compare with the masked excerpt const { data: invoice, error } = await supabase (the line's SHA-256 is 8f0794c44f6289e5da5b0a07bcd2cbbdcd35daad07161c8bdebdf36296b76236).

Limitations: Read from the source: identity checks, helpers and middleware are followed through imports of tracked files only. A check inside a package, a custom check that is not recognised, or a protection added by the host or a proxy is not seen. No request was made to the app.

high C003: POST /api/invoices: no identity check of its own; whether middleware protects it cannot be decided
  • Severity: High
  • The same root cause is also reported at C019, C020.
  • Check families: service-role, route-authorization, object-scoping (the first is the main one)
  • Location: app/api/invoices/route.ts:24
  • Status: Confirmed (AI reviewer and second reviewer (agreed)): the evidence is present at this revision.
  • Finding state: Source-supported: confirmed from a reading of the code at this revision (static evidence only); the deterministic checks did not settle it on their own, and no code was run.
  • Evidence: const { data, error } = await supabase, line SHA-256 0d1f6d4798bb0945648580086435abdff5388b647d239e6ee1f63d9cffab0eea
  • Verdict reached by: AI reviewer and second reviewer (agreed)
  • AI reviewer: The middleware never blocks /api paths, and this handler writes invoices and items with the admin client without checking who is calling, so anyone can create invoices. (relied on app/api/invoices/route.ts:13-35)
  • Second reviewer: Anyone without signing in can POST to /api/invoices and the route inserts the invoice and its items through the service-role client, because neither the handler nor the middleware (which only guards /dashboard) checks who is calling. (relied on app/api/invoices/route.ts:13-35)

Reproduction:

  1. In a clone of the repository, run git checkout --detach 2d7f14fb67c9c49af0e0dc23dc85b7ad127687ca.
  2. Open app/api/invoices/route.ts at line 24.
  3. Compare with the masked excerpt const { data, error } = await supabase (the line's SHA-256 is 0d1f6d4798bb0945648580086435abdff5388b647d239e6ee1f63d9cffab0eea).

Limitations: Read from the source: identity checks, helpers and middleware are followed through imports of tracked files only. A check inside a package, a custom check that is not recognised, or a protection added by the host or a proxy is not seen. No request was made to the app.

high C019: The invoice's owner id comes from the request body and is written with the service-role client, so anyone can create invoices inside anothe…
  • Found by AI: proposed by an AI reading the code, not by a rule-based check, and then confirmed.
  • Severity: High
  • Same root cause as C003 (listed above, which is confirmed too): the AI reviewers judged this the same problem seen at another place, so it is counted once.
  • Check families: service-role, route-authorization, object-scoping, request-trust (the first is the main one)
  • Family added by Nordik from the AI's own sentence (a service-role operation on a target the request chooses is a service-role defect first): service-role
  • Location: app/api/invoices/route.ts:6
  • Status: Confirmed (AI reviewer and second reviewer (agreed)): the evidence is present at this revision.
  • Finding state: Source-supported: confirmed from a reading of the code at this revision (static evidence only); the deterministic checks did not settle it on their own, and no code was run.
  • Evidence: user_id: z.string().uuid(),, line SHA-256 8b765b27f74c57458e77a79d1bbc33f0556e258b099d48244d756c0ff0da737a
  • Verdict reached by: AI reviewer and second reviewer (agreed)
  • AI reviewer: The POST handler takes user_id straight from the request body and inserts it with the admin client, which skips row security, and never checks a session, so anyone can create invoices under any user's account. (relied on app/api/invoices/route.ts:5-28)
  • Second reviewer: Anyone, signed in or not, can POST to /api/invoices with another user's id in user_id, and the route writes that invoice into the victim's account through the service-role client, which skips row-level security and never compares the id with a session. (relied on app/api/invoices/route.ts:5-28)

Reproduction:

  1. Open app/api/invoices/route.ts at this revision and go to line 6.
  2. Read the quoted line and the code around it. The problem to look for: The invoice's owner id comes from the request body and is written with the service-role client, so anyone can create invoices inside another user's account.
  3. Follow the code that reaches this line (the route, the caller or the setting that uses it) and note every check on the way: login, validation, escaping, permission.
  4. Try an input or request that reaches this line, in a local copy of the app or by tracing it by hand. The problem is shown if it gets through unchecked and the result is the one described above; if a check on the way stops it, the finding does not apply.

Limitations: Found by AI: an AI read a bounded set of files and proposed this, and it did not follow other files, so code it did not read could change the answer. The finding's state or verdict says how far it was checked.

high C020: The middleware only requires a login for /dashboard paths, so the /api routes that use the service-role client can be called by anyone with…
  • Found by AI: proposed by an AI reading the code, not by a rule-based check, and then confirmed.
  • Severity: High
  • Same root cause as C003 (listed above, which is confirmed too): the AI reviewers judged this the same problem seen at another place, so it is counted once.
  • Check families: service-role, route-authorization, object-scoping (the first is the main one)
  • Location: lib/supabase/middleware.ts:28
  • Status: Confirmed (AI reviewer and second reviewer (agreed)): the evidence is present at this revision.
  • Finding state: Source-supported: confirmed from a reading of the code at this revision (static evidence only); the deterministic checks did not settle it on their own, and no code was run.
  • Evidence: if (!user && request.nextUrl.pathname.startsWith("/dashboard")) {, line SHA-256 bb74c4590c7c9d9c109dad8435afbc27e45d051e1c849f62b808a45b07949ecd
  • Verdict reached by: AI reviewer and second reviewer (agreed)
  • AI reviewer: The middleware runs on every path but only redirects signed-out visitors on /dashboard, and the /api routes that use the admin client have no login check of their own, so anyone can call them without signing in. (relied on lib/supabase/middleware.ts:24-34)
  • Second reviewer: The middleware runs on every path but only redirects signed-out visitors on /dashboard paths, so anyone can call the /api routes without logging in, and those routes read and write invoices through the service-role client with no check of their own. (relied on lib/supabase/middleware.ts:24-34)

Reproduction:

  1. Open lib/supabase/middleware.ts at this revision and go to line 28.
  2. Read the quoted line and the code around it. The problem to look for: The middleware only requires a login for /dashboard paths, so the /api routes that use the service-role client can be called by anyone without signing in.
  3. Follow the code that reaches this line (the route, the caller or the setting that uses it) and note every check on the way: login, validation, escaping, permission.
  4. Try an input or request that reaches this line, in a local copy of the app or by tracing it by hand. The problem is shown if it gets through unchecked and the result is the one described above; if a check on the way stops it, the finding does not apply.

Limitations: Found by AI: an AI read a bounded set of files and proposed this, and it did not follow other files, so code it did not read could change the answer. The finding's state or verdict says how far it was checked.

high C004: POST /api/webhooks/payments accepts webhook events without verifying a signature (2 places)
  • Severity: High
  • The same root cause is also reported at C005.
  • Check families: request-trust, route-authorization (the first is the main one)
  • Location: app/api/webhooks/payments/route.ts:5
  • Status: Confirmed (AI reviewer and second reviewer (agreed)): the evidence is present at this revision.
  • Finding state: Source-supported: confirmed from a reading of the code at this revision (static evidence only); the deterministic checks did not settle it on their own, and no code was run.
  • Evidence: const event = await request.json(), line SHA-256 e9be6946fd3a101b325708681a2edf4c7cb77a236f53f4ce94c98b7ec5fb9ae4
  • Verdict reached by: AI reviewer and second reviewer (agreed)
  • AI reviewer: The webhook trusts the JSON body and marks any named invoice as paid using the admin client, and the configured webhook secret is never checked anywhere in the code, so anyone can fake a payment. (relied on app/api/webhooks/payments/route.ts:4-17)
  • Second reviewer: Anyone can POST a fake payment.succeeded event with any invoice id to this webhook, and the route marks that invoice as paid through the service-role client without checking any signature or shared secret. (relied on app/api/webhooks/payments/route.ts:4-17)

Reproduction:

  1. In a clone of the repository, run git checkout --detach 2d7f14fb67c9c49af0e0dc23dc85b7ad127687ca.
  2. Open app/api/webhooks/payments/route.ts at line 5.
  3. Compare with the masked excerpt const event = await request.json() (the line's SHA-256 is e9be6946fd3a101b325708681a2edf4c7cb77a236f53f4ce94c98b7ec5fb9ae4).
  4. The same finding is also at app/api/webhooks/payments/route.ts:11.

Limitations: A heuristic reading of one handler: it cannot tell whether the value is recomputed or checked elsewhere, so this is never confirmed on its own. No request was made to the app.

high C005: POST /api/webhooks/payments: no identity check of its own; whether middleware protects it cannot be decided
  • Severity: High
  • Same root cause as C004 (listed above, which is confirmed too): the AI reviewers judged this the same problem seen at another place, so it is counted once.
  • Check families: request-trust, route-authorization, service-role, object-scoping (the first is the main one)
  • Location: app/api/webhooks/payments/route.ts:11
  • Status: Confirmed (AI reviewer and second reviewer (agreed)): the evidence is present at this revision.
  • Finding state: Source-supported: confirmed from a reading of the code at this revision (static evidence only); the deterministic checks did not settle it on their own, and no code was run.
  • Evidence: await supabase, line SHA-256 d4ea8389aaa8d3ef266cc01ffb0390ffefeb99116932406993659c93039096c8
  • Verdict reached by: AI reviewer and second reviewer (agreed)
  • AI reviewer: The webhook route has no caller check (no signature and no use of the webhook secret), and the middleware does not guard it, so anyone can post to it and mark invoices as paid. (relied on app/api/webhooks/payments/route.ts:4-17)
  • Second reviewer: This is the same missing sender check as the unsigned-webhook finding: a forged event lets any caller mark any invoice paid through the service-role client. (relied on app/api/webhooks/payments/route.ts:4-17)

Reproduction:

  1. In a clone of the repository, run git checkout --detach 2d7f14fb67c9c49af0e0dc23dc85b7ad127687ca.
  2. Open app/api/webhooks/payments/route.ts at line 11.
  3. Compare with the masked excerpt await supabase (the line's SHA-256 is d4ea8389aaa8d3ef266cc01ffb0390ffefeb99116932406993659c93039096c8).

Limitations: Read from the source: identity checks, helpers and middleware are followed through imports of tracked files only. A check inside a package, a custom check that is not recognised, or a protection added by the host or a proxy is not seen. No request was made to the app.

high C006: Server action deleteInvoice: no identity check of its own; whether middleware protects it cannot be decided
  • Severity: High
  • Check families: service-role, route-authorization, object-scoping (the first is the main one)
  • Location: app/dashboard/invoices/actions.ts:9
  • Status: Confirmed (AI reviewer and second reviewer (agreed)): the evidence is present at this revision.
  • Finding state: Source-supported: confirmed from a reading of the code at this revision (static evidence only); the deterministic checks did not settle it on their own, and no code was run.
  • Evidence: await supabase.from("invoice_items").delete().eq("invoice_id", invoiceId), line SHA-256 df3396d45c965e1c59497d6c5133eaa76888956cf7809251684b1ba310172768
  • Verdict reached by: AI reviewer and second reviewer (agreed)
  • AI reviewer: deleteInvoice uses the admin client and deletes whatever invoice id it is given without checking that the caller owns it, so any caller who can invoke the action can delete other users' invoices; compare renameClient just below, which does check. (relied on app/dashboard/invoices/actions.ts:7-23)
  • Second reviewer: Any signed-in user can call the deleteInvoice server action with another user's invoice id (ids can be read through the open invoice read policy) and it deletes that invoice and its items through the service-role client, with no login or owner check. (relied on app/dashboard/invoices/actions.ts:7-12)

Reproduction:

  1. In a clone of the repository, run git checkout --detach 2d7f14fb67c9c49af0e0dc23dc85b7ad127687ca.
  2. Open app/dashboard/invoices/actions.ts at line 9.
  3. Compare with the masked excerpt await supabase.from("invoice_items").delete().eq("invoice_id", invoiceId) (the line's SHA-256 is df3396d45c965e1c59497d6c5133eaa76888956cf7809251684b1ba310172768).

Limitations: Read from the source: identity checks, helpers and middleware are followed through imports of tracked files only. A check inside a package, a custom check that is not recognised, or a protection added by the host or a proxy is not seen. No request was made to the app.

medium C021: The client id from the request is never checked against the caller, so an invoice can be linked to another user's client record, and later …
  • Found by AI: proposed by an AI reading the code, not by a rule-based check, and then confirmed.
  • Severity: Medium
  • Check families: object-scoping, request-trust (the first is the main one)
  • Location: app/api/invoices/route.ts:7
  • Status: Confirmed (AI reviewer): the evidence is present at this revision.
  • Finding state: Source-supported: confirmed from a reading of the code at this revision (static evidence only); the deterministic checks did not settle it on their own, and no code was run.
  • Evidence: client_id: z.string().uuid(),, line SHA-256 1e4dd386ca5ab97f28014331908c3971ba50fc73ec8c28c484f6b85f19e314a6
  • Verdict reached by: AI reviewer
  • AI reviewer: The client_id from the request is inserted with the admin client and nobody checks that it belongs to the caller, so an invoice can point at another user's client, and the GET route then returns that client's name, email and billing address. (relied on app/api/invoices/route.ts:5-28)

Reproduction:

  1. Open app/api/invoices/route.ts at this revision and go to line 7.
  2. Read the quoted line and the code around it. The problem to look for: The client id from the request is never checked against the caller, so an invoice can be linked to another user's client record, and later reads of that invoice then show that client's details.
  3. Follow the code that reaches this line (the route, the caller or the setting that uses it) and note every check on the way: login, validation, escaping, permission.
  4. Try an input or request that reaches this line, in a local copy of the app or by tracing it by hand. The problem is shown if it gets through unchecked and the result is the one described above; if a check on the way stops it, the finding does not apply.

Limitations: Found by AI: an AI read a bounded set of files and proposed this, and it did not follow other files, so code it did not read could change the answer. The finding's state or verdict says how far it was checked.

medium C009: Policy invoices_public_read on invoices lets every caller read every row
  • Severity: Medium
  • Location: supabase/migrations/001_create_tables.sql:42
  • Status: Confirmed (AI reviewer): the evidence is present at this revision.
  • Finding state: Source-supported: confirmed from a reading of the code at this revision (static evidence only); the deterministic checks did not settle it on their own, and no code was run.
  • Evidence: create policy "invoices_public_read" on public.invoices, line SHA-256 4f0b3fcad4f2e6aeceb3d583f5586df80e56b0d8f14a553503ea38bff518729b
  • Verdict reached by: AI reviewer
  • AI reviewer: Row security is on, but the permissive select policy using (true) has no role restriction and the SQL never revokes the default grants, so anyone with the public key can list every user's invoices, including totals, statuses and owner ids, through the REST API. (relied on supabase/migrations/001_create_tables.sql:31-43)

Reproduction:

  1. In a clone of the repository, run git checkout --detach 2d7f14fb67c9c49af0e0dc23dc85b7ad127687ca.
  2. Open supabase/migrations/001_create_tables.sql at line 42.
  3. Compare with the masked excerpt create policy "invoices_public_read" on public.invoices (the line's SHA-256 is 4f0b3fcad4f2e6aeceb3d583f5586df80e56b0d8f14a553503ea38bff518729b).

Limitations: Read from the SQL files committed under supabase/ only; row-level security or policies changed in the Supabase dashboard, and SQL that is not committed, are not seen. No request was made to the database.

medium C010: Policy invoice_items_read on invoice_items lets every caller read every row
  • Severity: Medium
  • Location: supabase/migrations/001_create_tables.sql:45
  • Status: Confirmed (AI reviewer): the evidence is present at this revision.
  • Finding state: Source-supported: confirmed from a reading of the code at this revision (static evidence only); the deterministic checks did not settle it on their own, and no code was run.
  • Evidence: create policy "invoice_items_read" on public.invoice_items, line SHA-256 1032ba2e61b08fccc3ada18aae662437d8dae4ebfa87b8dbd7445ff7ba83040a
  • Verdict reached by: AI reviewer
  • AI reviewer: The permissive select policy using (true) on invoice_items, with no revoke in the SQL, lets anyone with the public key read the line items and prices of every user's invoices. (relied on supabase/migrations/001_create_tables.sql:31-51)

Reproduction:

  1. In a clone of the repository, run git checkout --detach 2d7f14fb67c9c49af0e0dc23dc85b7ad127687ca.
  2. Open supabase/migrations/001_create_tables.sql at line 45.
  3. Compare with the masked excerpt create policy "invoice_items_read" on public.invoice_items (the line's SHA-256 is 1032ba2e61b08fccc3ada18aae662437d8dae4ebfa87b8dbd7445ff7ba83040a).

Limitations: Read from the SQL files committed under supabase/ only; row-level security or policies changed in the Supabase dashboard, and SQL that is not committed, are not seen. No request was made to the database.

low C014: Direct dependency @supabase/ssr pulls in 1 vulnerable package (@supabase/auth-js@2.67.3); upgrade target: @supabase/auth-js 2.70.0
  • Severity: Low
  • Location: package-lock.json:1065
  • Status: Confirmed (OSV evidence): the evidence is present at this revision.
  • Finding state: Reproduced: the deterministic checks found the same evidence again in the files at this revision. No code was run.
  • Evidence: @supabase/ssr: @supabase/auth-js@2.67.3, line SHA-256 19d962b9a1efd686ca6c05c909def944515b01496b3089cb59a5cada3ce20aa9
  • Verdict reached by: OSV evidence

Reproduction:

  1. In a clone of the repository, run git checkout --detach 2d7f14fb67c9c49af0e0dc23dc85b7ad127687ca.
  2. Open package-lock.json at line 1065.
  3. Compare with the masked excerpt @supabase/ssr: @supabase/auth-js@2.67.3 (the line's SHA-256 is 19d962b9a1efd686ca6c05c909def944515b01496b3089cb59a5cada3ce20aa9).

Limitations: The versions locked in the lockfile (package-lock.json, yarn.lock or pnpm-lock.yaml) were matched against a downloaded OSV.dev advisory snapshot, offline, and grouped under the direct dependency whose lockfile dependency graph pulls them in; whether the app uses the vulnerable code path was not checked.

low C015: Direct dependency @supabase/supabase-js pulls in 1 vulnerable package (@supabase/auth-js@2.67.3); upgrade target: @supabase/auth-js 2.70.0
  • Severity: Low
  • Location: package-lock.json:1087
  • Status: Confirmed (OSV evidence): the evidence is present at this revision.
  • Finding state: Reproduced: the deterministic checks found the same evidence again in the files at this revision. No code was run.
  • Evidence: @supabase/supabase-js: @supabase/auth-js@2.67.3, line SHA-256 6600ff8b88168ee643ba6174f30a4310cbc43f504fc560cc06ac7ffdc28b06df
  • Verdict reached by: OSV evidence

Reproduction:

  1. In a clone of the repository, run git checkout --detach 2d7f14fb67c9c49af0e0dc23dc85b7ad127687ca.
  2. Open package-lock.json at line 1087.
  3. Compare with the masked excerpt @supabase/supabase-js: @supabase/auth-js@2.67.3 (the line's SHA-256 is 6600ff8b88168ee643ba6174f30a4310cbc43f504fc560cc06ac7ffdc28b06df).

Limitations: The versions locked in the lockfile (package-lock.json, yarn.lock or pnpm-lock.yaml) were matched against a downloaded OSV.dev advisory snapshot, offline, and grouped under the direct dependency whose lockfile dependency graph pulls them in; whether the app uses the vulnerable code path was not checked.

low C008: CORS allows any origin together with credentials
  • Severity: Low (lowered from Medium by the AI reviewer, who said: Browsers refuse credentialed requests when the allowed origin is a wildcard, so this cannot be used to read data across sites; it is a misconfiguration.)
  • Location: next.config.mjs:17
  • Status: Confirmed (AI reviewer): the evidence is present at this revision.
  • Finding state: Reproduced: the deterministic checks found the same evidence again in the files at this revision. No code was run.
  • Evidence: { key: 'Access…(27 chars)', value: '*' },, line SHA-256 2d4878963acf4df3ac2a7bc22a46c7d17e561b6818cae7d2b6d3b47ef1003b45
  • Verdict reached by: AI reviewer
  • AI reviewer: next.config sends a wildcard allowed origin together with allow-credentials on every /api path; this is a broken CORS setup, though browsers block the dangerous combination. (relied on next.config.mjs:12-23)

Reproduction:

  1. In a clone of the repository, run git checkout --detach 2d7f14fb67c9c49af0e0dc23dc85b7ad127687ca.
  2. Open next.config.mjs at line 17.
  3. Compare with the masked excerpt { key: 'Access…(27 chars)', value: '*' }, (the line's SHA-256 is 2d4878963acf4df3ac2a7bc22a46c7d17e561b6818cae7d2b6d3b47ef1003b45).

Limitations: Static reading of source and configuration; headers added by the host, a CDN or a proxy, and the error text the running app actually returns, were not observed.

low C011: POST /api/invoices returns caught error text to the caller
  • Severity: Low
  • Location: app/api/invoices/route.ts:31
  • Status: Confirmed (AI reviewer): the evidence is present at this revision.
  • Finding state: Reproduced: the deterministic checks found the same evidence again in the files at this revision. No code was run.
  • Evidence: return NextResponse.json({ error: error.message }, { status: 500 }), line SHA-256 1dc7221af7b2c221ebb36954422ae0683cc5cd44ab8ea39e014e5d89ecbf9a99
  • Verdict reached by: AI reviewer
  • AI reviewer: When the insert fails, the handler sends the raw database error message back to the caller, which can reveal schema details. (relied on app/api/invoices/route.ts:24-32)

Reproduction:

  1. In a clone of the repository, run git checkout --detach 2d7f14fb67c9c49af0e0dc23dc85b7ad127687ca.
  2. Open app/api/invoices/route.ts at line 31.
  3. Compare with the masked excerpt return NextResponse.json({ error: error.message }, { status: 500 }) (the line's SHA-256 is 1dc7221af7b2c221ebb36954422ae0683cc5cd44ab8ea39e014e5d89ecbf9a99).

Limitations: Static reading of source and configuration; headers added by the host, a CDN or a proxy, and the error text the running app actually returns, were not observed.

low C012: next.config headers() sets no security headers
  • Severity: Low
  • Location: next.config.mjs:12
  • Status: Confirmed (AI reviewer): the evidence is present at this revision.
  • Finding state: Reproduced: the deterministic checks found the same evidence again in the files at this revision. No code was run.
  • Evidence: async headers() {, line SHA-256 e034a61b3cc2c881b02a084bd9ab5a54c54fe2714f7db5eb93dd3c3623aa797b
  • Verdict reached by: AI reviewer
  • AI reviewer: The headers() block only sets CORS headers, and nothing in the repository sets anti-framing, content-security, HSTS or similar security headers. (relied on next.config.mjs:12-23)

Reproduction:

  1. In a clone of the repository, run git checkout --detach 2d7f14fb67c9c49af0e0dc23dc85b7ad127687ca.
  2. Open next.config.mjs at line 12.
  3. Compare with the masked excerpt async headers() { (the line's SHA-256 is e034a61b3cc2c881b02a084bd9ab5a54c54fe2714f7db5eb93dd3c3623aa797b).

Limitations: Static reading of source and configuration; headers added by the host, a CDN or a proxy, and the error text the running app actually returns, were not observed.

Needs follow-up

None.

Rejected findings

0 findings were reviewed and rejected by the owner; they are not listed.

Since the earlier audit

No earlier audit of this source is on record, so this report compares nothing and marks nothing resolved. A problem is only reported as not found again after a rescan.

Limitations and exclusions
  • Scope: static checks that ran on this repository (detected stack: Next.js 15.2.9, Supabase): committed secrets (provider key formats adapted from gitleaks, and a generic rule); secrets exposed to the browser (variables with a public prefix such as NEXT_PUBLIC_); secret fallbacks; committed environment files and a debug environment setting; secrets committed and later deleted (git history); vulnerable or URL-sourced dependencies and install scripts; injection sinks (SQL, NoSQL, shell commands, XPath, LDAP, code evaluation), unsafe deserialization, XML entity expansion, open redirects, server-side requests to user URLs (SSRF), path traversal, regular expressions that backtrack catastrophically (ReDoS) and prototype pollution; JWT and hashing mistakes, predictable tokens and hard-coded session secrets; Supabase row-level security and policies in the committed SQL (tables without row-level security, row-level security turned off, policies that let everyone in or only test that the caller is signed in, update policies on privilege columns, and SECURITY DEFINER functions that anyone can call without a caller check); Next.js route handlers, pages/api routes, server actions and server components (pages and layouts) that read or write data without an identity check (a middleware counts only if it verifies identity and its matcher covers the route); Next.js handlers that use a Supabase service-role client without an identity and role check, or select a row by an id from the request after only a login check; Next.js ids taken from the request without an owner comparison, and request bodies written as they are; prices, totals, roles and similar values taken from the request body and written to the database (a heuristic: reported as inconclusive only); Next.js CORS with credentials, error details returned in responses, production source maps and missing security headers; secret-named keys in the env block of next.config, which Next.js inlines into the JavaScript bundles. Nothing else was assessed. An empty section means nothing found by these checks, not that the app is free of that problem.
  • No credential was tested against any live service.
  • The repository's own code was never run: no install scripts, package scripts, hooks or builds. Runtime behaviour was not observed.
  • Covered in part: payments and billing flows (C019, C002, C003, C004, C005, C006, and 2 more); authentication and authorization flows (C001, C020, C002, C003, C004, C005, and 1 more); data isolation between users (C019, C002, C003, C005, C006, C021, and 2 more). The findings named in parentheses were raised there by the rules and the AI's reading of the code, so these areas were looked at, but only in part: the AI read a bounded set of files and the fixed rules look for set patterns, so an area with a finding listed is not free of other problems.
  • Not covered: input validation beyond the injection patterns above, data flow across files, infrastructure and deployment configuration outside the repository.
  • Out of scope for a static audit (OWASP Top 10:2025): A06 Insecure Design, A08 Software or Data Integrity Failures, A09 Security Logging and Alerting Failures.
  • Dependency advisories: versions locked in package-lock.json, yarn.lock (v1 and berry) or pnpm-lock.yaml (or, without any lockfile, exact versions pinned in package.json) were matched offline against the OSV.dev npm snapshot of 2026-11-17; nothing about this repository was sent anywhere. Vulnerability data: OSV.dev (CC-BY 4.0 / CC0).
  • Skipped: 1 file(s) (lockfile (read for dependency versions only; not scanned for secrets): 1). Only the history reachable from this revision was read; other branches were not.
  • Static patterns miss issues and over-report; an empty section does not mean the app is free of that class of problem.
  • Scope: this is a limited-scope audit of nextjs-supabase, not a full audit. The limited scope covers: database policy.
  • Scope: nextjs-supabase is not currently qualified for a full audit (recorded qualified; current: not qualified (unproven, stale, unknown-version, excluded-family:service-role, excluded-family:secrets, unvalidated-family:service-role, unvalidated-family:secrets)). It is permitted only for: database policy.
  • Scope: the newest fresh held-out re-check of nextjs-supabase (2026-11-17, corpus-nextjs-9) failed some families, so these are not yet measured: route authorization 7/11, request trust 7/11. Findings in them are best-effort, and a clean result in them is not a pass. Passed that run: service role 10/11, object scoping 10/12, config headers errors 10/11, database policy 11/11, secrets 11/11.
  • Best effort: check families route authorization, service role, object scoping, request trust, config headers errors, secrets are outside this scope; their checks ran as extra best-effort checks, the findings they raised are listed and were reviewed like any other, and no finding in them is not a pass.
  • Excluded by the profile: Payments and billing flows
  • Excluded by the profile: The running app, hosting and network configuration (the audit reads the repository)
  • Excluded by the profile: Databases and dashboard settings that were not supplied as files
  • Excluded by the profile: Problems the audit does not find: it does not find everything
  • Managed-Cloud hosting cannot be told from the files; if the backend is managed Cloud, say so at intake.
  • A full audit is not currently qualified: the current evidence is not validated; the evidence was not measured with the current rule version; framework version 15.2.9 is not among the validated versions; service-role is excluded; secrets is excluded; service-role is not validated; secrets is not validated. Offered now: database-policy.
  • Qualification evidence: Recorded state: qualified (historical acceptance is not current validation). Current evidence: unproven; scope nextjs-supabase/d411-v1; observed 2026-10-14; framework versions unknown; rule audit.app-audit@1; measured candidate be2e47f54668174ea4b2872ad83f37674cfd0227. Families: route-authorization, object-scoping, request-trust, config-headers-errors, database-policy. Excluded: service-role, secrets, middleware authorization, edge-function authorization, storage policies, managed Cloud, code execution. Held-out validation pending. The October 12 full-pipeline seen-case regression recovered 20/25 cases; four of seven families missed recall, owner effort was unmeasured. The October 8 deterministic-only failure is not a complete live-pipeline failure. D-428 records one later seen-case fix, not independent acceptance.
How the audit is done

The audit is AI-performed. Claude, an AI model, reads one repository at one commit, the one you submit, and ranks what it finds.

Each critical or high code finding is sent to a second independent AI reviewer before the report, and each report says how many of them that reviewer answered.

No person checks each audit before release. Built and run by Nordik, an independent application security audit service.

The audit is not a penetration test, a certification, a compliance statement or proof that the app is safe.

What is not checked: the running app, live database settings, hosting and third-party accounts. The code is read and never run.

The audit can miss problems, and a finding can be wrong.

About this report

AI-found issues are quote-checked against the code but may still be wrong. No person read this report before it was released. An empty section means nothing was found by these checks, not that there is no risk.

Not a penetration test, a certification, a compliance statement or proof that your app is safe. We do not run your code; please run your tests before merging.

"The owner" in this report means the person who runs this service, not the owner of your app.

Findings the AI called not real

2 findings the AI called not real

2 findings were rejected on the AI's answer without the owner reading each one; each is listed, with how the verdict was reached, so it can be checked:

  • C007: Page /portal/[id]: no identity check of its own; whether middleware protects it cannot be decided, at app/portal/[id]/page.tsx:8. Rejected (AI reviewer and second reviewer (agreed)): False positive. AI reviewer: The portal page is a share link meant for the freelancer's customer: it reads through the normal session client and shows only the invoice number, total, status, due date and line items, and the real exposure comes from the open table policy covered by C009. (relied on app/portal/[id]/page.tsx:4-14) Second reviewer: The portal page is a share link to a customer that works like a key: it uses the normal user-scoped client to show one invoice's number, total, status, due date and items for an unguessable id, which matches its single purpose (the separate problem is the open read policy on the invoices table, not this page). (relied on app/portal/[id]/page.tsx:4-14)
  • C013: Policy clients_owner_all lets users update their own clients row, including billing_address, at supabase/migrations/001_create_tables.sql:35. Rejected (AI reviewer): False positive. AI reviewer: billing_address is a contact detail that freelancers enter for their own clients, not a value filled from a payment provider, and no code treats it as one, so letting owners edit it on their own rows is expected. (relied on supabase/migrations/001_create_tables.sql:2-36)

Sample technical report on an invented app. Nordik Audit. Plain report.