The EndpointOps Audit Checklist for Engineers Who Carry Prod Keys

The EndpointOps Audit Checklist for Engineers Who Carry Prod Keys

Pipelines get audited. Laptops that hold the pipeline credentials usually do not. If you carry the on-call pager or the deploy keys, the machine in your bag is part of your production surface, and an audit of it takes an afternoon. This checklist turns the habits behind EndpointOps into yes-or-no questions you can tick through. The Ops Atlas describes the term as an informal label for running user devices across their whole life, so there is no certificate to chase; the value is in the checks themselves.

How to use this list

Run it against a sample of ten devices first, then the whole estate. Mark each item pass, fail or unknown. Treat unknown as fail until proven otherwise; an unanswered question is a finding. Record the date, the person who checked and the evidence, and note any exception someone has agreed to, with an expiry date so that temporary waivers do not quietly become permanent. Keep a screenshot or export, not a recollection.

Section A: Do you know your fleet?

  • One list of devices exists, with owner, operating system and purchase date.
  • The management console, the security agent’s inventory and the asset list were compared in the last seven days.
  • Every device absent from one of the three sources has somebody’s name against it, plus a deadline.
  • Management coverage is measured as devices that checked in within thirty days.
  • Unsupported operating systems are counted separately and have a retirement date.

Section B: Are the tenant defaults safe?

  • The option that labels devices with no compliance policy is set to non-compliant for any tenant that gates access on device state. The default, as the Atlas write-up of EndpointOps stresses, is compliant.
  • Every device group has at least one compliance policy assigned.
  • Each setting has exactly one owning policy; duplicates were searched for after the last change.
  • Grace periods are explicit rather than left at the zero-day default.

Section C: Encryption and keys

  • Keys are stored centrally before encryption is enforced on new devices.
  • Escrow coverage is calculated as keys stored divided by devices encrypted. The Atlas worked example is 1,131 of 1,164, or 94.3 percent. Aim above ninety-five.
  • The strict boot-time check is used where it matters, and staff know that a restart clears a fresh flag.
  • The list of administrators who can read keys has been reviewed; seven built-in roles can, and few people need that power.
  • A key retrieval drill has been run and timed in the last quarter.

Section D: Protection health

  • The security agent reports on each device within the last seven days.
  • Antivirus status and sensor health are tracked separately.
  • Application control rules ran in audit mode before enforcement. Newer guidance prefers App Control for Business over its older sibling.
  • Remote management tools that attackers repurpose are reviewed against an approved list, following the 2023 joint advisory from US agencies cited by the Atlas.

Section E: The admin plane

  • Nobody uses a global administrator account for daily work.
  • Wipe, retire and delete actions, as well as scripts and role changes, need a second approver.
  • The approver group is a security group, assigned to the role with direct members; nested groups do not work.
  • Phishing-resistant sign-in is enforced for administrators.
  • Daily caps on bulk wipes, retires and deletes are known and documented.

Section F: Lifecycle and endings

  • Offboarding distinguishes retire (company data only) from wipe (everything).
  • Records for stale devices are never deleted until key custody is confirmed; deleting some Windows device objects suspends disk protection.
  • Hardware that leaves the organisation is sanitised according to current NIST guidance, SP 800-88 revision 2.
  • Personal phones are offered app-level protection instead of forced full enrolment.

Evidence worth keeping

Auditors, whether internal or external, will ask how you know. Save exports of the three device lists with dates in the file name. Keep a screenshot of the tenant default setting and the policy assignment page. Store the stopwatch time from your last key retrieval drill. File the names of everyone with permission to read keys and the date you last reviewed them. When you reconcile, note the reconciled percentage next to the console percentage, because the size of that difference tells you more than either figure alone. In the Atlas lab example the console claimed about 89 percent while only 40 percent of devices survived a three-way comparison; seeing that gap in your own data is uncomfortable but instructive.

Scoring and follow-up

Count the passes per section and track them quarter by quarter. Do not average them into one friendly number; a perfect score on section F does not offset a failure in section E. Convert each fail into a ticket with an owner. Fix the defaults and key custody first, since they influence everything else.

Where this fits in your engineering habits

If auditing feels foreign, borrow the mindset from the problem-solving techniques in our piece on building sharper critical thinking: state what you expect to find, look, and write down what you actually saw. For the wider organisational change that a device audit can trigger, the roadmap article on DevOps-led IT transformation and the skills it needs gives useful context.

Print the list, run it this week, ideally with a colleague who did not set the estate up, since fresh eyes catch assumptions the original builder no longer notices, and bring the failures to your next planning meeting. The goal is not a perfect score; it is a list of dated, owned, fixable gaps.