
An audit is only useful if you can answer each question with yes, no or “show me the evidence”. This checklist is built for engineers who want to test whether their organisation can really handle a security incident, a capability that the Ops Atlas calls ResponseOps. Work through it with a colleague, mark each line honestly and keep the gaps as your backlog. It also doubles as a career asset: anyone who can run this audit and fix what it finds has a skill set employers keep asking about.
One point of vocabulary first. In its reference entry covering ResponseOps, the Atlas admits the label is informal and that standards use “incident response”. Write the standard phrase in any formal document you produce from this audit.
How to score
Give each item 0 (absent), 1 (exists but untested) or 2 (exercised in the last twelve months, with evidence). The Atlas proposes five maturity levels: ad hoc, defined, practised, measured, adaptive. A rough mapping is that a total under a third of the maximum means ad hoc, up to two thirds means defined, and anything above needs proof of rehearsal before you call it practised.
Section A: ownership and declaration
- A named incident lead and a deputy exist, and both know it.
- An executive sponsor can approve disruptive actions at short notice.
- The criteria for declaring an incident fit on half a page.
- One person, with a backup, owns the decision that the organisation has become “aware” for regulatory purposes.
- Responders can reach each other on a channel that does not depend on corporate mail or chat.
Evidence for item 4 matters most. The Atlas observes that missed notification deadlines usually trace back to an unowned awareness call, not to ignorance of the rules.
Section B: evidence handling
- The team knows to capture memory before disk and has the tooling ready.
- A hash is taken at collection time and written down.
- Custody entries use UTC and name who handled the item.
- Evidence lands on write-once storage, and analysts work on copies.
- Log retention for authentication and audit sources is long enough to look back a month or more.
These habits trace back to a 2002 RFC on evidence collection, and a 2006 NIST guide on forensics supports them. They cost little to set up and are nearly impossible to improvise at 3 a.m.
Section C: scoping and containment
- There is a current map of accounts, hosts and data stores that can be queried within an hour.
- Containment options are listed per asset type, with the side effects written next to each.
- Anyone proposing a delay in containment must set a time limit and name an approver.
- The team has checked how its cloud network controls treat sessions that are already open.
- Credential revocation steps account for lifetimes: identity-service disablement lag and temporary cloud credentials that stay valid until expiry.
Items 14 and 15 are where audits often reveal a gap. The Atlas notes that changing a security group in one major cloud does not interrupt tracked connections, that a subnet-level deny will cut them but also catch unrelated traffic, and that disabling a user in a large identity service takes about 15 minutes for clients that check continuously and as much as an hour for the remainder. If your runbook assumes instant effect, it is wrong.
Section D: eviction and recovery
- The playbook says eviction waits until scoping is complete, with a documented list of all persistent access found.
- After a directory compromise, the plan includes a double reset of the ticket-signing secret, about ten hours between resets.
- Backups are validated as clean before any restore starts.
- The team has considered dead-man switches before isolating or powering off a host.
- Recovery can proceed alongside eradication where safe, rather than being forced into strict sequence.
For context, the Atlas cites the US cyber agency’s position that all persistent access must be accounted for before eradication, and that eradication and recovery may run together. Keep both ideas on the wall.
Section E: notification and external parties
- A clocks map exists, prepared with counsel, listing the deadlines that might apply to your data.
- It is dated, since the Atlas’s own table is stamped October 2026 and rules move.
- Contacts for law enforcement, insurers and key customers are stored somewhere reachable during an outage.
The Atlas’s October 2026 snapshot runs from a six-hour report to India’s national response team, through a 72-hour window under the EU privacy rule and a 24-hour first alert under NIS2, to the finance sector’s resilience rule, which counts four hours from the moment an incident is classified, and a US-listed firm’s four business days after it judges an incident material. Health-sector rules in the US allow 60 days, and a critical-infrastructure reporting law is not yet in force. Do not rely on any single table, mine or theirs, without checking with counsel.
Section F: learning and measurement
- An executive tabletop took place within the last year, and its gaps are tracked to closure.
- The team reports time-based and ratio measures, not incident counts.
- Post-incident reviews feed back into playbooks within a fixed period.
For item 25, the Atlas’s worked example is useful as a template. Dwell time is the alert time minus earliest attacker activity, so activity at 21:52:40 UTC and an alert at 09:10 UTC the next day yields 11 hours 17 minutes 20 seconds. Layer on the declaration delay, the containment delay, how far the affected set grew (15 hosts against an initial 6 is a factor of 2.5) and quarterly ratios for repeat entry and on-time notices. The numbers are illustrative, but each has an auditable definition, which is what makes the table useful.
Collecting the proof
An audit without artefacts is a conversation. For each item scored 2, store a pointer to the proof: a calendar invite for the tabletop, a screenshot of the retention setting, the custody template with a filled-in example, the signed-off clocks map. When a reviewer asks to see it, you should be able to open it within a minute. Items you cannot evidence drop to a 1, however confident the team feels. That rule feels harsh at first, but it keeps the score honest and it trains the habit that response work depends on: write it down, timestamp it and keep a copy somewhere safe.
Reading your results
Most first-time audits show strength in section A (people like having owners) and weakness in sections B, C and F, where the work is unglamorous. Rank gaps by likelihood and consequence, then follow the Atlas’s 30/60/90 sequence: ownership, bridge and emergency access first; playbooks, evidence store and clocks map second; tabletop, fixes, handoff to wider incident management and first measures third.
Skills from this audit travel well. If you are planning your own growth, our overview of the skills, projects and certifications on a career roadmap shows where response knowledge slots in, and the piece on moving from support and sysadmin work into engineering is a natural next read for operators who already live near production.
A caution about what the audit cannot tell you
A checklist measures preparation, not competence under stress. According to the Atlas, naming the culprit is frequently a dead end, and a tally of incidents shows how busy people were rather than how well they worked, so resist the urge to turn a score into a trophy. Rerun the audit after your next exercise, compare answers and let the evidence, not the total, drive the conversation.