Home/One-page refreshers
One-page refreshers
Every topic on one page: what it is, what to remember, what to run, what catches people, and how to check it worked.
Each one prints on a single A4 or letter page, for the desk or the incident bridge. There are 53 in all.
The tools
Writing down the infrastructure you want in files, and letting a tool make reality match them.
A tool that creates and changes cloud resources to match .tf files. OpenTofu is its open-source fork, and takes the same files with tofu in place of terraform.
A tool that logs in to machines you already have, over SSH, and runs a list of steps on each one. Nothing is installed on the machines first.
A controller inside a Kubernetes cluster that keeps the cluster matching a Git repository, and puts back changes made by hand.
The clouds
AWS, Azure, Google Cloud, and Oracle Cloud share the same building blocks under different names.
Responding
The same phases as any incident, with evidence that disappears faster, more ways for an attacker to come back, and most actions taken through APIs.
Five roles adapted from the Incident Command System, the rhythm they work to, and how they pass from one person to a team.
The fastest safe way to cut an attacker off, by what they have, and what each move leaves open.
Capture what disappears first, keep it where nobody can change it, and record who collected what.
The common paths through a cluster, where each one leaves traces, and what stops it.
What Google Cloud records by default, how to read an identity out of it, and the first moves for each kind of compromise.
Where a container keeps its evidence, when each part disappears, and the order to capture it in.
Which nodes let you checkpoint a container or capture their memory, and the commands for each on EKS and GKE.
What each cloud records about a cluster, how to keep a node, and what to rely on when you cannot reach it.
What CloudTrail records by default, how to follow a key or a role session, and the first moves for each kind of compromise.
What Azure records by default, how to read a caller or a sign-in, and the first moves for each kind of compromise.
What the Audit service records, how to read a principal, the credentials to revoke, and how to keep an instance's disks.
What each service records about a cluster, what a pod can reach on its node, and how to keep the node's disks.
What outlives a deleted pod, where attackers persist in Kubernetes, and how to rotate what cannot be revoked.
What to capture first on a suspect Linux machine, where persistence hides, and which logs matter.
What to capture first on a suspect Mac, where persistence hides, and which artifacts answer which question.
What to capture first on a suspect Windows machine, where persistence hides, and the events worth knowing by number.
Triage, authority, evidence, and hand-over for a case about someone who is allowed in.
What to look for in a pull request or a plan, so the next incident is caught in review.
Templates
The state file contains secrets in plain text. This stores it encrypted, versioned, and locked while a run is in progress.
One tag or label keeps a machine alive, captures its disks and, on AWS, its memory, and cuts it off, in that order, on four clouds and Kubernetes.
One script keeps a pod's node alive, records the pod, snapshots the node's disks, and cuts the pod off, on GKE, EKS, AKS, or OKE.
Send cloud audit logs to storage that refuses deletes for a set number of days.
Turn off password logins and root logins, and check the config before restarting so you cannot lock yourself out.
Forward system logs to a collector so they outlive the machine that wrote them.
auditd rules on Linux for account changes, sudo rights, SSH keys, persistence, kernel modules, and commands people ran as root.
osquery on Linux, macOS, and Windows, logging changes to accounts, keys, persistence, listening ports, and drivers.
Falco on plain Linux hosts, alerting on web shells, credential theft, and new programs in system folders.
Policies that make dangerous actions fail outright, so there is nothing to alert on afterwards.
Continuous evaluation that flags a resource the moment it stops matching the standard.
A delivery pipe from cloud logging to the place your detections actually run.
A separate account that does not depend on your identity provider, and shouts every time it is used.
Reach a server for troubleshooting with nothing listening on the internet, and a recording of the session.
A managed store with scheduled rotation, a record of every read, and a way to take access away.
Your own keys, rotated on a schedule, with a usage log and an off switch.
Traffic records stored cheaply for months, in a format you can query.
The first Application, which then manages every other application from Git.
Runtime alerts installed by the same pipeline that ships the workloads, so they cannot fall behind.
A clean-room virtual machine for looking at suspicious files, with no route to anything live, on four clouds.
Pipelines and pods prove who they are with short-lived tokens, so there is no cloud key to leak.
Admission policies refuse privileged pods, host mounts, and unsigned images, and report every refusal.
Cluster audit logs sent where responders can search them, so exec into a pod or a new role binding is never invisible.
Secret scanning, push protection, and dependency alerts switched on and sent to the SIEM, so a leaked key is an incident.
Perimeter policies so stolen credentials cannot copy data to an attacker's own account or project.
Alerts on the moves someone makes on the way out with data: sharing it outside, reading far more than usual, changing who can decrypt it, and quieting the logs.
Model call records on Bedrock, Vertex AI, Azure OpenAI, and OCI Generative AI, with alerts when logging or guardrails are turned off, or stolen keys go looking for models.
Alerts in Okta, Entra ID, Google Workspace, and Microsoft 365 on the moves that turn one stolen account into many. They cover admin rights granted, sign-in rules changed, MFA removed, mail forwarded out, and a log feed going quiet.
Mutual TLS between every service, and a deny-by-default rule for which service may call which.