The sample looks for the most severe critical problems and stops at the first ones it finds. In a few minutes you see the method, the shape of a finding and the trail β before hiring anything.
Step 01
Claude Code installed β terminal, VS Code or the app β and a paid Claude account: a subscription or API credits. Claude Code does not work on a free account, and without it nothing runs β neither this test nor the paid audit. It is the same tool that runs the paid audit: what you hire from SVO is the method, not the intelligence. How to install and pay with credits →
β οΈ The AI credit is yours, billed by Anthropic on your own account. Free is our part. So start with a small project: you see the whole method spending little.
Step 02
It can sit wherever you like β the example below is just an example. Only one thing is required, and it is the next step: the copy of your project in a folder named proj, with the kit next to it.
Linux / macOS: same thing β any folder of yours, for example ~/audit. Do not use the root /: on macOS the system volume is read-only and creating a folder there fails even with sudo.
Step 03
The folder is always named proj β the fixed name the Free Kit already expects. You can leave out node_modules, vendor and dist β they are not analysed anyway.
Bring the .git along if you can. History reveals secrets that were committed and later deleted β they are still there. It is optional: without it the audit runs the same and states in the report that this check was left out. The .git gives no access to your GitHub account.
Step 04
Give us your name and e-mail and we send you the kit link and your token. The link works once and expires in 7 days; the token is good for one run, on one project.
We ask for your e-mail so you can receive the ZIP and start the audit.
Unzip it and move the svoaudit folder into audit, beside proj β never inside it. Your token comes inside the kit, ready to use.
It must look exactly like this:
β οΈ The two side by side, never one inside the other. The session opens inside svoaudit; your project stays in proj, next to it.
Step 05
It is inside the svoaudit folder you have just unzipped. Double-click it and your Claude opens in a terminal: it asks whether you trust this folder and, once you confirm, the sample audit starts on its own.
β οΈ One question will appear, once: Claude Code asks whether you trust this folder. Say yes β it is the folder you have just unzipped. That answer is what switches the locks on: until you answer, Claude Code ignores the locks that ship with the kit, including the one that stops it writing to your code. Once accepted they apply and it does not ask again.
After that, just wait. The sample hunts on its own, confirms each finding against the counter-check and stops at the first two critical ones β it usually takes minutes. You can watch the screen, and it runs on its own to the end.
The shortcut opens the session from inside svoaudit, and that is what keeps your project separate: opening one level above and navigating to the right folder separates nothing β to Claude Code it is still a single workspace. The shortcut exists so the place is never wrong. Before starting it checks that Claude Code is installed and that the copy of your project is in proj; if anything is missing it stops and says which. Linux or macOS: bash amostra.sh, from the same folder.
Step 06
It is written inside svoaudit β outside your project folder, which is never touched:
Every finding brings file, line, the trail from input to effect, the impact and how to verify the fix. It is the same format as the paid audit β what differs is the scope, not the rigour.
Step 07
The full audit is quoted by the size of the system, and you are the one measuring, on your machine, with an open utility. It takes a few minutes.
Do not improvise or work around it. Stop and write to us β if a step did not work, the defect is in our text and we will fix it. An instruction that only works for those who already know is no instruction at all.
The kit is plain text: every file comes in the ZIP and you can read each one before running β including the guards, in .claude/. The measurer opens right here, in your browser: medir.py