Contents

A published application for Security Camp 2026’s Product Security Class connects hands-on infrastructure experience with specific security questions. The most useful lessons for applicants are to explain their decisions, investigate one problem in depth, and show what remains unresolved.

Key Takeaways#

  • Connect the lectures you want to attend to problems encountered in your own projects.
  • Explain technical experience through decisions and constraints, rather than lists of tools.
  • Support your answers with focused experiments, distinguishing measurements from speculation.
  • Treat vulnerability counts as leads for investigation, rather than proof of system security.

Make Your Motivation Specific#

The applicant named two lectures: B4, covering secure coding and the use of AI, and B5, covering attack surface management through attackers’ reconnaissance techniques. Both choices addressed questions arising from their own infrastructure work.

For B4, the question was how to evaluate AI-generated code, vulnerability reports, and proposed fixes. For B5, it was how to check what outsiders could discover about services exposed through infrastructure built with Kubernetes, WireGuard, and Cloudflare Access.

This connection makes the motivation specific. Instead of writing “I want to learn security,” identify something you cannot yet confidently assess. Explain your current work, the gaps in your understanding, and how the lecture would help you make that assessment.

Show the Decisions Behind the Tools#

The application described experience with an Astro portfolio, a Cloudflare Workers API, a Discord bot, team development, and a Kubernetes environment running on Proxmox. The most persuasive passages explained why particular approaches were chosen.

The portfolio reduced writing overhead by accepting a single Markdown file as an article and publishing it automatically after a push. In team development, command handling was split into small components to keep pull requests manageable for review. These examples tell readers more than the names of frameworks or programming languages.

Review your account of your experience using these four questions.

  • What did you build, and who used it?
  • What constraints influenced the design?
  • What decisions did you make?
  • What did you verify, and what remained uncertain?

In their retrospective, the applicant identified a weakness here: they wished they had explained environment-specific constraints more clearly. This is a useful review criterion for anyone preparing similar answers.

Investigate One Problem in Depth#

For the open-ended assignment, the applicant focused on software supply chain attacks and CI/CD security. The answer examined how compromised packages or workflows could leak credentials, allowing attackers to publish further malicious packages or modify repositories.

The proposed measures addressed both prevention and containment: reviewing workflow changes, limiting token permissions, pinning dependency versions, and reducing long-lived credentials by using OIDC authentication where supported. Detection and response included monitoring publishing activity, isolating affected runners, revoking credentials, and rebuilding in a clean environment.

Keep incident descriptions accurate. The submitted answer covers Shai-Hulud and the Nx “s1ngularity” attack, but initial access methods and persistence behavior need to be checked against reports for each incident and variant. Do not combine multiple campaigns into a single sequence presented as common to every case. Similarly, lockfiles improve reproducibility, but cannot prove that pinned packages are trustworthy.

Explain the Reduction from 50 to 14#

The application’s most concrete experiment involved scanning a container image for a small Go application with Trivy. According to the applicant, vulnerability findings fell from 50 before the Dockerfile changes to 14 afterward—a reduction of 36, or 72%.

The original image retained the Go build environment. The revised multistage build compiled the application in one stage, then copied the executable into a separate Alpine runtime image configured to run as a non-root user. Removing build components reduced both the contents of the runtime image and the number of findings requiring review.

These figures were reported by the applicant and are not an independently reproduced benchmark. A reproducible record would require the image digest, scanner version, vulnerability database timestamp, and scan options. The Go and Alpine versions shown in the source were used in a past experiment. If reusing the configuration in November 2026 or later, check which versions are supported.

Prioritize Beyond Severity Labels#

Fewer findings do not prove that the revised image is secure. Each remaining vulnerability needs to be assessed based on its external reachability and impact. The applicant separated application dependencies from OS packages and considered whether affected components were used at runtime and whether fixes were available.

This is a useful starting point, but application dependencies should not always take priority over OS packages. Either can contain the most urgent issue. Prioritize based on the following evidence.

  • Is the affected component present in the deployed image and used at runtime?
  • Can attacker-controlled input reach the vulnerable operation?
  • Has exploitation been observed, and what access or data is at risk?
  • Is a fix or mitigation available? If remediation is deferred, are an owner and review date assigned?

The same approach helps when writing an application. Explain why you investigated a finding, what evidence changed your assessment, and why you deferred unresolved issues. A measured reduction is valuable, but the judgment behind it makes for a more persuasive answer.

Next Step#

Choose one project you already maintain and write a 200-word application paragraph covering one security question, the checks you performed, your measurements, and an uncertainty you still need help resolving.

Source: https://zenn.dev/uvu/articles/758ee227a614ea

Contents