Claude Code makes it work. Who checks it's safe?
Claude Code is very good at making a feature work. It is not the one answering for who can reach it. Access rules, secrets and the data your users trust you with get decided one prompt at a time, and nobody reads them as a whole.
We do. Start with the Bunker Test: we read your codebase and send back a written Top Risks summary within 5 business days. Free.
Where AI-written code leaks
Not everywhere. In a handful of places, and always the same ones:
- Access rules. Who can read and change what, decided feature by feature and sometimes only hidden in the interface instead of enforced on the server.
- Secrets. API keys and tokens that end up in the repo, in client code or in logs.
- Data exposure.Endpoints that send back more than the screen needs, including other users' data.
- Untrusted input. What a user types, reaching your database or another service without being checked first.
The code still works. That is why nobody notices until someone else does.
How it stays safe after the audit
| Every PR, two passes | An agent flags secrets and sensitive surfaces, then a senior engineer reviews |
| 24h business SLA | No volume cap. Miss it and you get credit back |
| Auth in the Core Registry | Any change to access rules gets senior review, every time |
| Security is ours | Architecture, security, CI/CD and incident response sit with us |
You keep shipping features. The parts that decide who sees what are listed in your repo as load-bearing, and they get read properly. More on how the review works on the Claude Code review page.
Start with the audit
The Bunker Test is free. Read-only access to your repo, then a written Top Risks summary within 5 business days. Yours to keep, whether or not you work with us.
We run 5 a week.
Pricing
Konstrukt Co-Build, €4,500/month. Review on every PR, security and incident response included. 3-month minimum, then month-to-month, 30 days notice.
Questions founders ask
Is code written with Claude Code secure?
Not by default, and not insecure by default either. It does what the prompt asks. Security depends on decisions nobody prompted for: who can access what, where secrets live, what an endpoint sends back. Those need someone who reads the whole picture.
What does a security review of a Claude Code project look at?
Access rules enforced on the server, secrets in the repo and client code, endpoints that expose more data than they should, and untrusted input reaching the database. After that, every PR that touches auth or access gets senior review.
What does the Bunker Test include?
Read-only access to your repo, then a written Top Risks summary within 5 business days. It is free and yours to keep, whether or not you work with us afterwards.
How much does Konstrukt cost?
€4,500 a month, everything included, with no cap on PRs. Three months minimum, then month to month with 30 days notice.