Skip to content
Post Mortem

For people working in tech

Vibe-coding security kit

Eleven tools, in the order you would use them: set up safely, check what you add, review what the AI builds, scan before you publish, then watch, learn and practise. We may receive compensation from some partners, but recommendations are based on how useful they are.

Use them in this order

  1. Before you start

    Snyk Agent Scan

    Agent security · Account required

    Checks the skills and connectors you attached to your AI assistant for suspicious instructions, risky permissions and unsafe behaviour.

    Best for: People installing AI skills, plug-ins and MCP connectors.

    Official Snyk project that looks at AI-agent parts, MCP servers and skills, not at your app. It covers problems like hidden instructions and mishandled keys.

    Good to know

    • Needs a Snyk account and token, and sends information about the parts it finds to Snyk, so it is not fully offline.
    • Checking a connector setup can start programs or make network requests. The maintainers advise using a sandbox for anything you do not trust.
    • Its output format is marked experimental and may change.
  2. Before you add a package

    OpenSSF Scorecard

    Project security checks · Open source

    Look beyond download counts. Check whether an open-source project reviews changes, pins its own packages and signs its releases.

    Best for: Judging tools and packages before you add them.

    A project of the Open Source Security Foundation. Useful when an AI assistant suggests a package and you want to know how carefully it is looked after.

    Good to know

    • Its checks are rules of thumb, and the project itself says they give false alarms and miss things. A good score does not mean a package is safe.
  3. While you build

    Trail of Bits Skills

    Agent skills · Open source

    Ready-made review steps you give your AI coding assistant, so it checks risky changes, unsafe settings and borrowed code instead of just asking whether your app works.

    Best for: People building with Claude Code or Codex.

    Published by the security firm Trail of Bits. Includes reviews of code changes, checks for settings that start unsafe, a review of the packages you depend on, a second opinion and a check after a fix.

    Good to know

    • Free to use and share, but if you copy or adapt the material you must credit it and share your version under the same licence (CC BY-SA 4.0).
  4. When a feature is done

    Cloudflare security-audit-skill

    Open source

    A set of instructions you give your AI coding assistant so it reviews your project for security problems in stages.

    Use: Security audits with coding agents

    Provides a structured, multi-phase audit workflow with independently verified, machine-readable findings.

  5. Before you publish

    Semgrep

    Code scanning · Open source plus paid platform

    Scans your app's code for common security mistakes before you publish it. It can also be connected to AI coding assistants so they can run scans themselves.

    Best for: Builders who want security checks while they work.

    A well-established code scanner with JavaScript and TypeScript support and an official connector for AI assistants.

    Good to know

    • The free edition only looks inside one function or file at a time and misses things the paid platform finds. A clean scan is not a full security review.
  6. Before you publish

    Google OSV-Scanner

    Package checks · Open source

    Finds known security problems in the packages your app uses, including ones your AI assistant added without you noticing.

    Best for: Checking packages before you publish and after a security alert.

    Maintained by Google and built on the OSV flaw database. It reads your project's package files, works with npm and many other ecosystems, and can run offline once its database is downloaded.

    Good to know

    • It finds known flaws only. It cannot tell you a package is genuine or free of hidden malicious code.
    • Its upgrade helper runs package-manager commands, so do not use that feature on a project you do not trust.
  7. Before you publish

    Gitleaks

    Secret scanning · Open source · Maintenance mode

    Finds passwords, secret keys and other secrets accidentally saved in your project or in its older versions.

    Best for: Checking a project before you put it on GitHub.

    A long-established secret scanner. Runs on your own computer, before you save changes, or as a GitHub check.

    Good to know

    • The maintainer says it is finished: future releases will be security fixes only, and his focus has moved to a successor called Betterleaks. It still works, but it is not gaining new features.
  8. After launch

    Aikido Security

    Commercial platform with open-source projects

    A service that scans your code, packages and cloud setup and shows what needs fixing.

    Use: Code, dependency, cloud, and runtime security

    Useful for teams that need continuous visibility into vulnerabilities, dependencies, and attack-surface risk.

  9. Learn from a real team

    Cohere security toolkit

    Open source

    Open-source security tools published by a company's own security team, useful as a worked example.

    Use: Open-source security tooling from the Cohere Security team

    Shows how a security team can build practical, open tooling around AI workflows, including an MCP server for Cohere North.

  10. Practise a bad day

    IRGame

    Multiplayer incident-response exercise platform

    Online practice sessions where a team works through a made-up incident together.

    Use: Online tabletop exercises for incident-response practice; no technical experience required

    A useful complement for teams that want live, multiplayer incident drills.

  11. When you grow

    Eye Security

    Commercial security provider

    A company that watches for attacks and responds for you, so a small team does not have to staff that itself.

    Use: Managed detection and response, SOC, incident response, and cyber insurance

    Relevant for mid-sized teams that want 24/7 managed monitoring and response rather than building it in-house.

Listing here is not an official endorsement or partnership, and none is implied. Descriptions reflect the tools' own public pages and may be out of date. Star counts come live from GitHub. Logos are trademarks of their respective owners. They are shown to identify each tool or platform and do not imply endorsement or partnership.