Managed service · Sofia, Bulgaria
QA & security audit
An independent check of your website’s quality and security, done from the inside with the access you grant. For small-business websites.
EnquireI am not taking engagements yet. Enquiries are open, and I will reply when the first audit slots open.
01
Every confirmed finding comes with three things
- 01
A scenario that reproduces it
Each finding is written as a Karate scenario: a short automated test that shows the problem every time it runs.
- 02
A fix
A concrete change that removes the problem, not general advice. I prepare it with AI assistance and review it myself.
- 03
A re-run
The same scenario runs again after the fix, and its result proves the fix holds.
A finding I cannot reproduce is reported as “unconfirmed”.
02
A worked example
A demo on an invented API with invented data, running locally. No real company’s system is involved. The figures come from an actual run.
- 23 / 23
- Karate scenarios pass on the correct API
- 3 / 3
- deliberately planted bugs are caught
- 1
- crash found by an AI review, fixed and re-run
Break it yourself
The demo API runs inside this browser tab: the requests and responses are real, the data is invented, and nothing is sent to a server. Pick a request, plant a bug and see which checks catch it.
Loading…
1 · Pick a request
Each customer is one HTTP request · 3D is illustration, the panels show the actual data
HTTP request
HTTP response
2 · Plant a bug
3 · The Karate checks
QA: each bug fails exactly its own scenarios
The bugs are switched on one at a time. Each run fails only the scenarios that cover that bug; the rest stay green.
| Planted bug | Failed | Scenarios that catch it |
|---|---|---|
| A required field is dropped | 1 / 23 | Business has all fields required by the published schema |
| The check between tenants is skipped | 2 / 23 | A foreign tenant user is hidden even when its ID is knownOrders of a foreign-business user are hidden |
| An order can be cancelled twice | 1 / 23 | Cancelling the same order twice is rejected, so no second refund is issued |
Security: one tenant cannot read another’s data
The check is written as a scenario that anyone on the team can read and run again.
Scenario: A foreign tenant user is hidden even when its ID is known
Given path 'businesses', businessId, 'users', otherUserId
When method get
Then status 404
And match response.error.code == 'NOT_FOUND'AI: a crash found, a fix and a re-run
An AI code review found that one malformed request stopped the whole server, before login. A script reproduces the crash, the fix removes it, and the same script proves it.
- Before
GET //[ → process exits with code 1 (ERR_INVALID_URL) - Fix
let url; try { url = new URL(req.url, 'http://localhost'); } catch { return error(400, 'BAD_REQUEST', 'Invalid request target'); } - After
GET //[ → 400 Bad Request next valid request → 200, server still running
03
What it is and what it is not
What it is
An inside-access audit: code, CMS admin, hosting panel, staging environments. I check what you have given me access to.
- Security headers
- Authentication
- Access control
- Forms
- Input validation
What it is not
It is not penetration testing and not external scanning. Nothing is tested without a signed scope and written authorisation from the owner.
04
How it works
- 01
Scope
We sign a scope that lists every asset: sites, environments, panels.
- 02
Authorisation
The owner of each asset gives written authorisation. Without it I do not start.
- 03
Access
You grant the access you choose: code, CMS admin, hosting panel, staging environment.
- 04
Audit
I check only what the scope lists. Each finding is reproduced before it enters the report.
- 05
Results
You receive a finished report: findings with their scenario, fix and re-run.
05
How your data is handled
- Each client has a separate folder, separate credentials and separate notes. Client data is never mixed.
- Data stays local wherever possible.
- I use AI models as a tool for triage, analysis and preparing fixes. Only I operate them; where possible the models run locally.
- Every model session is logged: date, client, assets touched.
- You receive only the finished results and never operate the tooling.
06
Enquiry
Describe the site and what worries you. Do not send passwords or access details through this form. I do the audit personally — Lyubomir Lyubenov.