11 free tools to put between your vibe-coded app and production
A few weeks ago I wrote about the minimum security checks for vibe-coded apps, and why testing doesn’t have to cost five times what the build did. The response was good. The behaviour hasn’t changed much.
I still see apps going from “it works on my laptop” to “customers are logging in” in a single afternoon. The step in between, where someone actually checks whether the app is safe, keeps getting skipped. Sometimes it’s the cost. More often, I think, people simply don’t know where to start.
So this post is the practical follow-up. Below are 11 open-source tools, all free to use, that cover the second rung of the ladder from that first post: automated checks that run every time the AI changes your code. None of them replaces a human review. But if you are shipping without any of them, the excuse that “proper testing is too expensive” no longer holds.
First, a rule for all 11
Only point these tools at code and systems you own, or that you have written permission to test. Several of them send real attack traffic. Run the attack-style tools against a test copy of your app, not the live one holding real customer data.
The list is grouped by where each tool sits in the journey from your coding agent to production.
AI checking the AI’s work (1 to 4)
If an AI wrote most of your code, it makes sense to start with tools built for that world. These four work alongside the coding agent you already use.
- 1) Cloudflare Security Audit – A free “skill” you install into your coding agent (Claude Code, Codex and similar). Ask for a security audit and the agent works through it in six phases, and a separate agent checks each finding before it is reported. That second check matters, because AI reviewers love to raise false alarms. It reads your code. It does not test your server.
- 2) NVIDIA SkillSpector – Agent skills and plugins run with the same access as your coding agent, and most people install them without reading them. SkillSpector scans a skill before you install it, looking for prompt injection, data exfiltration and hidden code, and gives you a verdict. NVIDIA cites research finding that about a quarter of published skills contained a vulnerability and around 5% showed signs of malicious intent. Use it on any skill you didn’t write, including the ones on this list.
- 3) Strix – A team of AI agents that attack a running copy of your app the way a pentester would, inside a sandbox on your own machine. It only reports a finding once it has a working proof that the attack succeeds, and it suggests fixes. You bring your own AI model key, so budget for the usage. This is the closest thing on the list to someone actually trying to break in.
- 4) Claude Code Security Review – A GitHub Action from Anthropic that reviews the changed code in every pull request and leaves its findings as comments, before a human even opens it. If you don’t use GitHub, Claude Code has the same check built in as the
/security-reviewcommand. One caution from its own documentation: it is not hardened against prompt injection, so don’t let it run automatically on pull requests from people you don’t know.
The proven scanners (5 to 9)
These are older, rule-based tools. They are less clever than an AI reviewer, but they give you the same answer every time you run them, and they are fast enough to run on every change.
- 5) Semgrep – Searches your code for known dangerous patterns, such as user input going straight into a database query. Write a rule once and it catches the same bug in every file it appears in. One thing to know: since December 2024, the rules Semgrep maintains have carried a licence limiting them to internal, non-commercial-service use. Checking your own app is fine. If that licence bothers you, Opengrep is a community fork.
- 6) Gitleaks – Catches API keys, passwords and tokens before they are saved into your code history. Set it up to run automatically before every commit. Secret keys ending up somewhere public was one of the four minimum checks in my first post, and this is the cheapest way to stop it at the source.
- 7) TruffleHog – Finds credentials that have already leaked, anywhere in your code history, and then checks whether they still work. That tells you which leaks are urgent. Remember that deleting a key from your code does not remove it from the history. If TruffleHog says a key is live, cancel it and issue a new one.
- 8) OSV-Scanner by Google – Checks every package your app depends on against a public database of known vulnerabilities. AI coding tools add packages freely and often suggest versions that are years old. Most of the code in your app was written by someone else, and this is how you find out which of it is known to be broken.
- 9) zizmor – Finds security holes in GitHub Actions, the automation that builds and deploys your code. Badly written workflows can let an outsider run their own code inside your pipeline or leak its credentials, which is where many supply-chain attacks begin. If you deploy by copying files to a server and don’t use GitHub Actions, you can skip this one.
Once it’s live (10 and 11)
Your code can be clean and the app still exposed, because of how the server is set up. These two look at what is actually running.
- 10) Nuclei – Scans your live address for thousands of known vulnerabilities and misconfigurations, using templates maintained by the security community. It will flag exposed admin panels, leftover configuration files and software with published flaws. This covers much of the server checklist from my first post. Point it only at your own address.
- 11) garak by NVIDIA – If your app has a chatbot or any AI feature, garak throws jailbreaks, prompt injections and data-leak attempts at it, before real users do. It tests how the AI part behaves. It does not test the rest of your app, so it sits alongside the others rather than replacing them.
Where to start if you’re not a developer
Eleven tools is a lot if you built your app by describing it to an AI. You don’t need all of them on day one. If you do nothing else this week, do these four:
- Run TruffleHog over your project – and replace any key it reports as still working.
- Ask your coding agent to run Cloudflare’s audit – or
/security-reviewif you use Claude Code. Then ask it to explain each finding in plain language before it fixes anything. - Run OSV-Scanner on your project folder – and update anything it flags as critical or high.
- Run Nuclei against your own live address – and close anything it finds exposed.
What these tools won’t do
I’d be doing you a disservice if I let this list sound like a finished job.
None of these tools knows your business. A scanner can’t tell that a store manager should never see another store’s figures, or that a customer on the basic plan shouldn’t be able to reach the premium export. Those are business rules, and they are where some of the most damaging leaks come from. Strix comes closest, but it doesn’t know your rules unless you tell it.
They also produce reports, not fixes. A finding nobody reads protects no one, and some findings will be false alarms that need a person to judge. Someone has to own the results.
That is why the ladder has more rungs. These tools are the second rung, and they should run every time your code changes. The third is a focused, fixed-scope review by a person who looks at your app the way an attacker would. The full penetration test comes when your clients, your data or your revenue demand it.
Vibe coding isn’t going anywhere, and it shouldn’t. But “it works” has never meant “it’s safe”. The tools to check the difference are free. If you’ve run them and want someone to look at what they can’t catch, get in touch and we can scope a review around your app.
References
- Cloudflare, security-audit-skill (GitHub)
- NVIDIA, SkillSpector (GitHub)
- Help Net Security, “SkillSpector: NVIDIA’s open-source security scanner for AI agent skills” (2026)
- Strix, usestrix/strix (GitHub)
- Anthropic, claude-code-security-review (GitHub)
- Anthropic, “Automate security reviews with Claude Code”
- Semgrep, semgrep/semgrep (GitHub)
- Semgrep, “Compare Semgrep to Opengrep” (licensing FAQ)
- Gitleaks, gitleaks/gitleaks (GitHub)
- Truffle Security, trufflehog (GitHub)
- Google, osv-scanner (GitHub)
- zizmor, zizmorcore/zizmor (GitHub)
- ProjectDiscovery, nuclei (GitHub)
- NVIDIA, garak (GitHub)
