An online directory of advocates holds data that visitors and listed professionals both care about: enquiry messages, contact submissions, profile edits and sign-in details. When such a platform runs on Kubernetes, the safety of that data depends on dozens of small configuration choices that no visitor ever sees. This checklist turns KubernetesSecOps, the routine practice of securing the cluster, into twenty-one verifiable items grouped into seven sections. What follows is guidance on platform technology for site operators. It makes no comment on legal questions or on how any listed professional works.

The items are adapted from the Ops Atlas page devoted to KubernetesSecOps. For each item, record three things: the evidence you saw, the date, and the person who owns the next review.

Section A: identity and permissions

  1. Sign-in goes through a central identity provider. Staff access ends automatically when employment ends.
  2. Nobody belongs to the built-in group that bypasses every permission check. The Atlas is categorical that no user should be added to it.
  3. Automation holds narrow roles. A deployment pipeline with cluster-wide administrator rights is on the Atlas’s list of anti-patterns, since control of the pipeline then means control of everything.

For item 3, also hunt for grants that reach further than they look: list or watch on secrets, write access to workloads (which indirectly reaches whichever service accounts they run under), exec and attach, the escalate, bind and impersonate verbs, token minting for service accounts, certificate approval, edits to admission webhook configuration, and patching namespaces.

Section B: what may start

  1. Pod Security labels are set on every namespace holding live workloads. Note the mode per namespace: audit, warn or enforce.
  2. Enforcement of the baseline level covers at least nine in ten namespaces. That is the Atlas’s bar for its guardrails maturity stage.
  3. The exception register exists and every entry has an expiry. Exempting a controller’s service account quietly exempts anyone able to create that workload, so check for that pattern.

Section C: what may talk

  1. A default-deny NetworkPolicy exists for the namespaces that hold enquiry data or sign-in services.
  2. The network plugin enforces policies. Test with a deliberate blocked connection rather than relying on documentation.
  3. Allow rules are specific. Public form handlers reach only the services they need, and the database is reachable only from the application tier.

Section D: credentials

  1. Encryption at rest is configured for Secrets through a key-management provider. The Atlas recommends the second version, as the first was deprecated and later disabled by default.
  2. Existing Secrets were rewritten after encryption was turned on.
  3. The most sensitive credentials come from an external store. Tools in this space include External Secrets Operator and the Secrets Store CSI Driver.

Section E: evidence

  1. Audit logging is enabled and a sample from the last week is on file. On self-managed API servers it does nothing without a policy file.
  2. The audit policy has been read in order. The first matching rule wins, so a broad rule placed too early can hide the detail you wanted.
  3. Secrets and config maps are logged at metadata level only.

Section F: watching

  1. Runtime detection is running and has an owner.
  2. The team knows its actionable-alert ratio. The Atlas’s example improved from 5.0 to 41.4 percent by excluding audited on-call sessions.
  3. Automated response never simply deletes pods. That destroys evidence and leaves tokens valid. Isolation by relabelling is the gentler option, because network rules are additive and cannot quarantine a pod.

Section G: rhythm and lifecycle

  1. The calendar has recurring entries. Weekly refusals and alerts, monthly exceptions, quarterly permissions and benchmark, yearly tabletop exercise.
  2. A list of components in front of the site is kept with maintenance status. The Atlas recounts a 2025 episode involving a popular ingress component, rated 9.8 for severity, in which a foothold on the pod network could lead to cluster takeover and exposure of every secret; the project was subsequently archived.
  3. A recent drill has been run and written up. The Atlas’s top maturity level requires two quarters of passed drills.

How to collect evidence without a fuss

Most items can be checked in minutes by someone with read access and a terminal. For sections A and B, export the role bindings and the namespace labels; for C, list the network policies per namespace and try one deliberate connection that should fail. For D, ask the person responsible for the encryption configuration to show the file and the date it was last changed. For E and F, open the log store and search for a harmless action you performed an hour earlier, for instance adding and then removing a scratch namespace. For G, open the shared calendar and count the recurring entries. Store each export in the same folder, named with the date, and the next reviewer will thank you.

What to do with a weak result

A low score is information, not an accusation. The Atlas proposes a 30-60-90 shape that works well here. The opening month is for observation only: catalogue components, put warn-mode labels on each namespace, switch on audit logs, and run one benchmark. The second month is for guarding, meaning baseline enforcement, a default-deny pilot in a single namespace, encryption of credentials and a named recipient for alerts. The third month is for verifying, which covers the strict profile for newly created namespaces, digest-pinned images, a signature trial and a rehearsal. Write the outcome of each month on a single page and keep the pages together, so that progress is visible to anyone who joins later. Resist skipping ahead. Teams that enable every sample policy in blocking mode on the first day usually end up disabling them all by the end of the week.

How to score yourself

Count the items where you saw evidence. Fewer than seven puts you at the default end of the Atlas’s ladder, where the sensible plan is to look before you lock down. Between seven and fourteen suggests hygiene and early guardrails. Fifteen or more means you are ready to verify, with signature checks and rehearsals.

Where this sits among your other duties

Directory operators juggle listing quality, content moderation and user support, and cluster security can feel like one more demand. Keep it proportionate. Pick the three lowest-scoring sections, schedule them across a quarter, and revisit. For context on what a city-level listing page involves, see our intellectual property advocate guide for Guwahati or the equivalent listing covering Madurai, each of which is the kind of page whose enquiry flows this checklist is meant to protect.

Three notes on wording with your host or developer

  • Ask for “evidence” and “date”, not “are we secure?”.
  • Expect some answers to be “not applicable”. Managed hosting shifts duties, so ask which logs and controls the provider turns on by default, and which you must enable.
  • Keep the checklist in one shared document, so results from each review are comparable.

A cluster is never finished, and neither is this list. What matters is that the next review finds fewer undated assumptions than the last one did.

0 0 votes
Article Rating
Subscribe
Notify of
guest
0 Comments
Oldest
Newest Most Voted
0
Would love your thoughts, please comment.x
()
x