Sample report / AIpento

Readiness report: Taskly

Synthetic example, not a real customer. 2026-09-21 - 11 files inspected, 7 of them code. Static inspection only: your code was never executed.

This is a sample. Taskly is a small application we wrote ourselves to stand in for the shape these tools produce: no real service, no real person, and no real credential anywhere in it. Your report has exactly this structure, about your code.
VERDICTREBUILD

5 critical findings and no test suite to protect a fix. Repairing this one finding at a time costs more than rebuilding the core on a sound base.

SeverityFindings
Critical5
High6
Medium6
Low2
Needs your confirmation2

Critical

Live credential committed to the codebase

Check S1, security lens.

  • src/lib/ai.js:1

OpenAI API key, sk-p...[48 chars]. The value is not reproduced anywhere in this report.

const OPENAI_KEY = 'sk-p[REDACTED]';

Why it matters. Anyone who ever gets a copy of this repository has the key. Deleting the line does not help: it stays in git history.

What to do. Rotate the credential first, then move it to an environment variable. Treat it as exposed until it is rotated.

SQL built by joining strings

Check S2, security lens.

  • api/tasks.js:7
    sql: "SELECT * FROM tasks WHERE title LIKE '%" + q + "%'",

Why it matters. A value from a visitor becomes part of the query. This is how a form field turns into "read every row in the table".

What to do. Use parameterised queries. Never concatenate or interpolate a value into SQL.

Code built and executed at runtime

Check S4, security lens.

  • src/lib/rules.js:3
    return eval(expression);

Why it matters. If any part of the string can be influenced from outside, this runs an attacker code inside your process.

What to do. Replace with a lookup table or a parser. There is almost always a way that does not execute text.

Administrative database key referenced in client-side code

Check S7, security lens.

  • src/lib/supabase.js:6
    process.env.SUPABASE_SERVICE_ROLE_KEY,

Why it matters. A service-role key bypasses every row-level rule. In browser code it is visible to every visitor, which makes your whole database readable and writable.

What to do. Use the public key in the browser. Keep the service-role key on the server only, and rotate the exposed one.

Environment file is not excluded from version control

Check D2, data lens.

  • .env:1

1 environment file(s) present and not matched by .gitignore

Why it matters. A .env holds your keys. Unignored, the next commit publishes them to anyone with repository access.

What to do. Add .env to .gitignore, rotate anything it contains, and commit a .env.example with the names only.

High

No automated tests anywhere in the codebase

Check T1, tests lens.

  • 7 code files, 0 test files

Why it matters. Nothing can tell you whether the next change broke something that used to work. Every fix becomes a gamble, and AI edits are large.

What to do. Start with one test over the path that earns you money. One real test beats a coverage target.

Untrusted value written into the page as HTML (2 places)

Check S3, security lens.

  • src/components/TaskList.jsx:2
    container.innerHTML = tasks.map((t) => `<li>${t.title}</li>`).join('');
  • src/components/TaskList.jsx:6
    return <div dangerouslySetInnerHTML={{ __html: html }} />;

Why it matters. If that value ever contains a script, it runs in your visitor browser with their session.

What to do. Use textContent, or React children. If HTML is genuinely needed, sanitise it first.

Cross-origin requests allowed from anywhere with credentials

Check S6, security lens.

  • api/share.js:2
    cors: { origin: '*', credentials: true },

Why it matters. Any website can make authenticated calls to your API using your visitor own session.

What to do. Replace the wildcard with an explicit list of origins you control.

Database in use with no migration history

Check D1, data lens.

  • database libraries are in use; no migrations directory, schema file or ORM schema was found

Why it matters. The only record of your schema is the live database. A new environment cannot be created, and a lost database cannot be rebuilt.

What to do. Adopt a migration tool and capture the current schema as the first migration.

Database query inside a loop

Check P2, speed lens.

  • src/lib/sync.js:6

loop opens at line 5

const row = await supabase.from('tasks').select('*');

Why it matters. One hundred items becomes one hundred round trips. This is the single most common reason an AI-built app slows to a crawl after launch.

What to do. Fetch the set in one query and join in memory, or use the database own batch form.

Medium

Error caught and discarded

Check F1, failure lens.

  • src/lib/sync.js:16
    } catch (err) {}

Why it matters. The failure still happened; now nobody can see it. These are the bugs that take days to find because there is no trace.

What to do. Log it with enough context to identify the request, and decide deliberately whether to continue or stop.

Network call with no error handling in the file

Check F2, failure lens.

  • src/lib/ai.js:4
    const res = await fetch('https://api.openai.com/v1/chat/completions', {

Why it matters. Other people servers time out and return errors. An unhandled rejection takes down the request, or worse, silently does nothing.

What to do. Wrap the call, decide what the user sees when it fails, and make that path visible.

Query with no limit on how many rows come back (2 places)

Check P1, speed lens.

  • api/tasks.js:7
    sql: "SELECT * FROM tasks WHERE title LIKE '%" + q + "%'",
  • src/lib/sync.js:6
    const row = await supabase.from('tasks').select('*');

Why it matters. Fine at twenty rows. At twenty thousand it loads the whole table into memory on every request.

What to do. Add a limit and paginate. Decide the maximum page size deliberately.

No dependency lockfile

Check R1, release lens.

  • a dependency manifest is present with no matching lockfile

Why it matters. Every install can resolve different versions. What you tested and what you deployed are not the same software, and the difference appears at the worst time.

What to do. Commit the lockfile your package manager produces.

Dependency pinned to a moving target

Check R2, release lens.

  • package.json:6
    "@supabase/supabase-js": "*"

Why it matters. A version of "latest" or "*" means a stranger release can change your app without you touching it.

What to do. Pin to a specific version or a narrow range.

Low

Network call with no timeout (2 places)

Check F3, failure lens.

  • src/lib/ai.js:4
    const res = await fetch('https://api.openai.com/v1/chat/completions', {
  • src/lib/sync.js:14
    const res = await fetch(url);

Why it matters. Without a timeout the default can be minutes. One slow dependency holds your requests open until the app stops answering.

What to do. Pass an AbortSignal with a timeout to every outbound call.

Needs your confirmation

These depend on intent, which no scan can read. Each is either fine or serious, and only you can say which.

Endpoint that writes data with no visible authentication

Check S5, security lens. Severity if confirmed: High.

  • api/tasks.js:12
    export async function POST(request) {

Why we raised it. If this really is unauthenticated, a stranger can write to your data. Some endpoints are public on purpose, so a human has to confirm which this is.

What to do. Confirm intent. If it should be private, check the session server-side before the write.

No deployment configuration in the repository

Check R3, release lens. Severity if confirmed: Low.

  • no Dockerfile, platform config or CI deploy workflow was found

Why we raised it. If deploying lives only in somebody memory or a dashboard, it cannot be repeated under pressure and cannot be rolled back quickly.

What to do. Put the deploy in the repository, even as a short documented script.

What this did not check

  • Nothing was executed. No test was run, no server started, no request sent to anything you own. Everything above comes from reading the code.
  • No live target. Headers, authentication behaviour, rate limits and anything that only exists when the app is running are untested here.
  • No dependency vulnerability lookup. That needs a network call to a vulnerability database and is not part of this pass.
  • Absence of a finding is not proof of absence. These are patterns that catch the failures that recur in AI-built software. A determined review finds more.

How this was produced

7 lenses, 28 checks. Every check is proven against a deliberately broken example and a correct one before it may run, so a check that cannot fail cannot ship.

Findings marked "needs your confirmation" never decide the verdict on their own.