roadmap
- 01
- 02
- 03
- 04
- 05
- 06
- 07
- 08
Interview for DevSecOps engineer roles ready to defend how security is built into delivery. Covers secure coding, scanning, container and pipeline hardening, cloud security, and compliance.
9664 of 9664 questions in your plan: 0 known, 0 in progress, 0 due, 0 skipped, 9664 new. 0 cut, outside your plan.
The round that defines a DevSecOps loop walks one change from commit to running container and asks, at each step, who could have tampered with it and what would have noticed. Around it sit a screen on scripting and Linux, often with a short Python or Bash exercise, a design round from middle level up on a pipeline or a cloud account, in many loops a hands-on review of a Dockerfile, a workflow file or a Terraform plan, and a behavioral conversation to close. The scenarios are specific. A CI job keeps a long-lived cloud access key in its secrets: what does it gain by exchanging its own OIDC token for a role whose trust policy checks the repository and branch in the `sub` claim? A workflow triggered by `pull_request_target` checks out the pull request's head commit: what has just happened to the secrets? A Kubernetes Secret is "encrypted" in base64: what must the API server be configured with before it is encrypted at rest, and what stops a pod from mounting it anyway? A scanner reports a long list of CVEs on a base image: which are reachable, and who may sign the statement that one is not exploitable here? Interviewers listen for whether a control is described by its tool or by its placement: which stage runs it, what it blocks, what it only reports, who reads the report. The strongest answers treat a gate as something with users. A new check runs in audit mode until its false-positive rate is known, an exception has an owner and an expiry, and what a build leaves behind, a signed image, its provenance, the policy decision that admitted it, is the evidence an auditor will ask for later.
Nothing in the first stretch is about security yet, and that is deliberate. [Python](/topics/lang-python) and Go are what the tooling is written in, [Bash](/topics/lang-bash) is what most pipeline steps run as, PowerShell covers the Windows estate, and [Linux](/topics/os-linux) supplies the users, permissions, processes and signals a container is made of. Algorithms serve the coding screen, and Git with [GitHub](/topics/dev-github) close the stretch, because the pull request is the chokepoint most later gates hang on. A long networking stretch follows, from the OSI model through addressing to ICMP, ARP and VLANs, then transport, NAT and DNS. It is placed here because Network Defense comes next and is written in those terms: a firewall rule, a WAF verdict, an IDS standing in the path and a segmentation boundary are all decisions about packets, and [zero trust](/topics/sec-network-defense-zero-trust) ends the stretch as the argument that the perimeter those rules build is not enough. From the network the plan moves to the two things a call carries, who is making it and whether the channel is private. [Federated identity](/topics/proto-auth-identity) is the protocol side of the first, OAuth2 through SAML; web transport is the second, HTTP over [TLS](/topics/proto-tls), the PKI behind a certificate, and the browser's own rules. Secure coding builds on both: application security principles, [threat modeling](/topics/sec-threat-modeling) to decide which of them a given design needs, how a service implements authentication and authorization, and [database access control](/topics/db-relational-security-access) down to least-privilege accounts and row-level security. Knowing what secure looks like, the plan turns to finding where it is not. Nmap and the scanner-concepts module cover how a finding is produced, Burp Suite and [ZAP](/topics/sec-owasp-zap) are the intercepting proxies, one for a human and one usually wired into a pipeline, and Wireshark checks what the wire actually carried. [Secrets management](/topics/sec-secrets-mgmt-concepts) closes the stretch because scanning finds leaked credentials and everything after it runs on credentials. Containers come next, [the isolation model](/topics/cloud-containers-concepts) before Docker and [Kubernetes](/topics/cloud-kubernetes), so that a hardening question gets a kernel-level answer rather than a flag name. Pipeline Hardening & GitOps is the stretch the role is named for: infrastructure as code, [CI/CD concepts](/topics/cloud-cicd-concepts) and GitOps describe how change moves, [supply chain security](/topics/sec-supply-chain) asks how you prove that what moved is what you built, and [policy as code](/topics/sec-policy-as-code) comes last because admission control needs a cluster and IaC gates need IaC. The closing stretches are where it runs and how you would know. They open on [the provider-neutral cloud model](/topics/cloud-providers-concepts), so that the shared-responsibility line and the platform's identity model are settled before one vendor's spelling of them; AWS follows as that spelling, its identity, key and audit layer taken from [policy evaluation](/topics/cloud-aws-iam-policy-evaluation) through KMS to CloudTrail, then observability signals, incident management and [detection engineering](/topics/sec-security-operations-siem-detection) for the SIEM. [Compliance frameworks](/topics/proto-infosec-standards) sit near the end on purpose: by then the controls they name exist, and the remaining question is evidence. The [behavioral round](/topics/found-behavioral) and the mechanics of the loop itself finish the plan.
Two kinds of candidate arrive here. A platform or DevOps engineer adding security already has the languages, Linux, Git and containers, and should spend the hours from Network Defense through Policy as Code, where questions about the current job's pipeline turn into questions about the next one's. A security analyst or pentester moving toward delivery has the opposite gap: recon and defense are familiar, and the time goes to Bash, Git, CI/CD concepts and Kubernetes, because the role is judged on changes that ship, not on findings that are filed. Days away from the interview, take the shortest list the page will give you: importance raised to must know, level narrowed to what the posting asks for, and study order left where it opens, at by frequency, so the questions asked most come first. Given three months instead, read the plan top to bottom once with study order set to by topics, and build one pipeline alongside the reading: a repository with branch protection, a workflow that assumes a cloud role over OIDC, an image that is scanned, signed and admitted by policy. The design round is largely a conversation about that pipeline and what you would change in it. Two neighbouring roles share this plan's edges without being in it. Service-level objectives, error budgets and capacity planning have no module here and belong to the [DevOps / SRE Engineer roadmap](/roadmaps/devops). The scanning tools here go as far as confirming a weakness and gating on it; exploitation, lateral movement and the forensics after a breach are the [Cyber Security Expert roadmap](/roadmaps/cyber-security).
The round that defines a DevSecOps loop walks one change from commit to running container and asks, at each step, who could have tampered with it and what would have noticed. Around it sit a screen on scripting and Linux, often with a short Python or Bash exercise, a design round from middle level up on a pipeline or a cloud account, in many loops a hands-on review of a Dockerfile, a workflow file or a Terraform plan, and a behavioral conversation to close.
The scenarios are specific. A CI job keeps a long-lived cloud access key in its secrets: what does it gain by exchanging its own OIDC token for a role whose trust policy checks the repository and branch in the sub claim? A workflow triggered by pull_request_target checks out the pull request's head commit: what has just happened to the secrets? A Kubernetes Secret is "encrypted" in base64: what must the API server be configured with before it is encrypted at rest, and what stops a pod from mounting it anyway? A scanner reports a long list of CVEs on a base image: which are reachable, and who may sign the statement that one is not exploitable here?
Interviewers listen for whether a control is described by its tool or by its placement: which stage runs it, what it blocks, what it only reports, who reads the report. The strongest answers treat a gate as something with users. A new check runs in audit mode until its false-positive rate is known, an exception has an owner and an expiry, and what a build leaves behind, a signed image, its provenance, the policy decision that admitted it, is the evidence an auditor will ask for later.
Nothing in the first stretch is about security yet, and that is deliberate. Python and Go are what the tooling is written in, Bash is what most pipeline steps run as, PowerShell covers the Windows estate, and Linux supplies the users, permissions, processes and signals a container is made of. Algorithms serve the coding screen, and Git with GitHub close the stretch, because the pull request is the chokepoint most later gates hang on.
A long networking stretch follows, from the OSI model through addressing to ICMP, ARP and VLANs, then transport, NAT and DNS. It is placed here because Network Defense comes next and is written in those terms: a firewall rule, a WAF verdict, an IDS standing in the path and a segmentation boundary are all decisions about packets, and zero trust ends the stretch as the argument that the perimeter those rules build is not enough.
From the network the plan moves to the two things a call carries, who is making it and whether the channel is private. Federated identity is the protocol side of the first, OAuth2 through SAML; web transport is the second, HTTP over TLS, the PKI behind a certificate, and the browser's own rules. Secure coding builds on both: application security principles, threat modeling to decide which of them a given design needs, how a service implements authentication and authorization, and database access control down to least-privilege accounts and row-level security.
Knowing what secure looks like, the plan turns to finding where it is not. Nmap and the scanner-concepts module cover how a finding is produced, Burp Suite and ZAP are the intercepting proxies, one for a human and one usually wired into a pipeline, and Wireshark checks what the wire actually carried. Secrets management closes the stretch because scanning finds leaked credentials and everything after it runs on credentials.
Containers come next, the isolation model before Docker and Kubernetes, so that a hardening question gets a kernel-level answer rather than a flag name. Pipeline Hardening & GitOps is the stretch the role is named for: infrastructure as code, CI/CD concepts and GitOps describe how change moves, supply chain security asks how you prove that what moved is what you built, and policy as code comes last because admission control needs a cluster and IaC gates need IaC.
The closing stretches are where it runs and how you would know. They open on the provider-neutral cloud model, so that the shared-responsibility line and the platform's identity model are settled before one vendor's spelling of them; AWS follows as that spelling, its identity, key and audit layer taken from policy evaluation through KMS to CloudTrail, then observability signals, incident management and detection engineering for the SIEM. Compliance frameworks sit near the end on purpose: by then the controls they name exist, and the remaining question is evidence. The behavioral round and the mechanics of the loop itself finish the plan.
Two kinds of candidate arrive here. A platform or DevOps engineer adding security already has the languages, Linux, Git and containers, and should spend the hours from Network Defense through Policy as Code, where questions about the current job's pipeline turn into questions about the next one's. A security analyst or pentester moving toward delivery has the opposite gap: recon and defense are familiar, and the time goes to Bash, Git, CI/CD concepts and Kubernetes, because the role is judged on changes that ship, not on findings that are filed.
Days away from the interview, take the shortest list the page will give you: importance raised to must know, level narrowed to what the posting asks for, and study order left where it opens, at by frequency, so the questions asked most come first. Given three months instead, read the plan top to bottom once with study order set to by topics, and build one pipeline alongside the reading: a repository with branch protection, a workflow that assumes a cloud role over OIDC, an image that is scanned, signed and admitted by policy. The design round is largely a conversation about that pipeline and what you would change in it.
Two neighbouring roles share this plan's edges without being in it. Service-level objectives, error budgets and capacity planning have no module here and belong to the DevOps / SRE Engineer roadmap. The scanning tools here go as far as confirming a weakness and gating on it; exploitation, lateral movement and the forensics after a breach are the Cyber Security Expert roadmap.