Home/Using the templates/How templates are checked
How templates are checked
Every template runs through the same checks on every change, and again once a week.
The weekly run matters more than it sounds. Providers retire arguments and change defaults, so a template that passed in March can fail in June with nobody touching it. The weekly run is how that gets noticed.
Terraform
| Check | What it catches | Command |
|---|---|---|
| Formatting | The code is formatted the standard way, so diffs stay readable. | terraform fmt -check -diff |
| Valid | Terraform can read it and every reference resolves. | terraform init -backend=false && terraform validate |
| Lint | Unused declarations, untyped variables, missing version constraints, and deprecated syntax. | tflint --minimum-failure-severity=warning |
| Security scan | No public storage, no missing encryption, no firewall rules open to everyone. | checkov -d . --framework terraform |
Ansible and Argo CD
Ansible
| Check | What it catches | Command |
|---|---|---|
| Ansible lint | Steps that are not safe to run twice, retired module names, missing names on tasks. | ansible-lint |
| Syntax | Ansible can read the playbook and every module referenced exists. | ansible-playbook --syntax-check |
Argo CD
| Check | What it catches | Command |
|---|---|---|
| Kubernetes schema | Every manifest matches the schema for the kind it claims to be. | kubeconform -strict -summary |
| Builds | The folder produces valid output before Argo CD ever sees it. | kustomize build |
Three that need nothing installed
These run in seconds without Terraform, Ansible, or a cluster, so they work as a pre-commit hook or in a locked-down environment.
| Check | What it catches | Command |
|---|---|---|
| parse | Every file parses. Runs without Terraform installed. | python tools/parsecheck.py outcomes |
| refcheck | Every setting used is declared, every setting declared is used, and quick-start templates stay one file. | python tools/refcheck.py outcomes |
| registrycheck | The site and the repo agree, and every template with code carries its explanation: what you need, what it creates, what catches people, and how to undo it. | python tools/registrycheck.py registry/templates.yml |
How a template earns its label
Ready means every check for its tool passed on the main branch. The label is changed by hand after someone reads the green run, so it cannot flip on its own by accident.
Draft means the code exists and at least one check has not run or has not passed. Planned means there is no code yet.
The registry check also refuses a template that has code and no explanation. A page here always tells you what you need first, what gets created, what catches people, and how to undo it.
When they last ran
264 of 273 folders passed every check for their tool when
last run, between 2026-10-02 and
2026-11-12. The results are recorded in
registry/checks.json by tools/localgates.py, and each
template page shows its own.
| Tool | Version |
|---|---|
| engine | OpenTofu v1.12.6 |
| tflint | TFLint version 0.64.0 |
| checkov | 3.3.19 |
Providers the Terraform templates were validated against:
| Provider | Version |
|---|---|
| hashicorp/archive | 2.8.1 |
| hashicorp/aws | 6.66.0 |
| hashicorp/azurerm | 5.6.0 |
| hashicorp/google | 8.4.0 |
| hashicorp/random | 3.9.1 |
| integrations/github | 6.13.0 |
| oracle/oci | 9.8.0 |
Security scan decisions
The scanner ran 4,166 checks across the Terraform templates: 3,785 passed and 381 were flagged, from 69 distinct rules. Each rule it flags is either fixed or has a written reason below. A new finding with no reason fails the build.
| Rule | Why it stands | Reason |
|---|---|---|
| Administrator users have no API keysCKV2_OCI_1 | Scanner misreads the code one folder | The break-glass quick start finds the Administrators group with a data lookup, which the scanner cannot follow. The account has no API key, and its capabilities are set so it cannot be given one. |
| AdministratorAccess attachedCKV_AWS_274 | By design two folders | The break-glass user must be able to repair anything, including the sign-in setup and the policies that broke. Its use is alerted within minutes, and the guardrail template's exempt list names it. |
| App Service plan minimum instancesCKV_AZURE_212 | Scanner misreads the code two folders | The check is written for dedicated plans. The responder runs on a Flex Consumption plan, which has no standing instances and scales from zero when an event arrives. |
| App Service plan zone redundantCKV_AZURE_225 | Scanner misreads the code two folders | The check is written for dedicated plans. A Flex Consumption plan has no instances of its own to spread across zones. |
| BigQuery dataset encrypted with your own keyCKV_GCP_81 | By design one folder | The findings history dataset uses Google-managed encryption. Findings describe resources and contain none of their data, and Security Command Center writes the export as its own service agent, which a key of your own would need to be granted. |
| Classic blob read loggingCKV2_AZURE_21 | By design eight folders | Classic storage analytics logging is superseded by diagnostic settings. The Standard state backend, and the Standard container evidence account when given a workspace, send every blob read and write to Log Analytics that way. |
| Classic queue loggingCKV_AZURE_33 | Scanner misreads the code 11 folders | These storage accounts contain no queues. |
| Cloud Storage access logsCKV_GCP_62 | By design eight folders | Cloud Storage access logs are a legacy feature. Data Access audit logs record reads and writes, and the audit log template explains how to switch them on. |
| Cloud Storage versioningCKV_GCP_78 | By design six folders | The audit and container evidence buckets' retention policies already refuse overwrites and deletes, so versioning adds nothing there. The evidence function's source bucket contains a zip Terraform rebuilds from code in Git, named by its hash, so an older copy is never needed. |
| CloudTrail SNS topicCKV_AWS_252 | By design three folders | A notification for every log file delivered adds noise that nobody acts on. |
| CloudTrail encrypted with KMSCKV_AWS_35 | Quick start keeps it simple one folder | The quick start uses S3-managed encryption. Standard encrypts the trail with your own KMS key. |
| CloudTrail sent to CloudWatch LogsCKV2_AWS_10 | Scanner misreads the code two folders | The quick start keeps to one bucket. Standard sends the trail to CloudWatch Logs whenever cloudwatch_logs is true, which the scanner cannot see through the count. |
| CloudWatch log retention of a yearCKV_AWS_338 | By design three folders | CloudWatch Logs is the alerting window. The long-term record is the trail's copy in S3: under Object Lock for the audit log, and kept for the retention you choose for object reads. |
| Compute instance boot volume in-transit encryptionCKV_OCI_4 | Scanner misreads the code two folders | The check reads is_pv_encryption_in_transit_enabled inside launch_options, which Oracle's provider documents as deprecated at create and for updates only. The template sets the field the create call uses, at the top level, where the check does not look. |
| Compute instance monitoring enabledCKV_OCI_6 | By design two folders | The forensic workstation runs no Oracle Cloud Agent plugins, which is what monitoring needs. An agent is a way to run commands on the machine containing the evidence, and a port-forwarding session needs none. The network is watched instead, by flow logs and an alarm on refused traffic. |
| Default security group left openCKV2_AWS_12 | Quick start keeps it simple one folder | Nothing in the forensic workstation quick start uses the VPC's default security group. Standard empties it, so nothing can be placed in an open group by accident. |
| Event Grid topic managed identityCKV_AZURE_191 | By design one folder | An identity on the status topic is only used to deliver to destinations that require one. The function sends to it as its own identity, and what you subscribe is your choice. |
| Event Grid topic public network access offCKV_AZURE_193 | Needs a virtual network one folder | The status topic accepts events over its public endpoint from the function's identity only, with key access off. Closing the endpoint needs a private endpoint and a virtual network the function runs in. |
| Firehose encrypted with a customer-managed keyCKV_AWS_241 | Quick start keeps it simple one folder | The quick start encrypts the stream with the AWS-owned key to stay one file. Standard encrypts it with its own key. |
| GitHub Actions OIDC trust conditionCKV_GCP_125 | Scanner misreads the code two folders | Both conditions are exact matches on GitHub's subject claim, which names the repository and the branch or environment. The scanner reads the repository name only when it is typed into the condition itself, so a name held in a local or a variable fails it. The Standard variable also refuses wildcards in the organisation or repository. |
| IAM permissions management without constraintsCKV_AWS_109 | Scanner misreads the code 16 folders | In the key policies, a resource of * means this key and nothing else. The evidence responder's role uses * because EC2 describe actions accept no narrower resource, and the responder must act on whichever instance was tagged. It has no delete, stop, or terminate actions. |
| IAM resource of *CKV_AWS_356 | Scanner misreads the code 18 folders | In the key policies, a resource of * means this key and nothing else. The evidence responder's role uses * because EC2 describe actions accept no narrower resource, and the responder must act on whichever instance was tagged. It has no delete, stop, or terminate actions. |
| IAM user outside single sign-onCKV_AWS_273 | By design two folders | A break-glass account exists for when single sign-on is down, so it has to be an IAM user outside it. It is refused everything without MFA, and every use raises an alert. |
| IAM write access without constraintsCKV_AWS_111 | Scanner misreads the code 18 folders | In the key policies, a resource of * means this key and nothing else. The evidence responder's role uses * because EC2 describe actions accept no narrower resource, and the responder must act on whichever instance was tagged. It has no delete, stop, or terminate actions. |
| KMS key protected from deletionCKV_GCP_82 | Quick start keeps it simple one folder | Standard keys set prevent_destroy. The key quick start leaves it off so it can be removed after trying it. |
| Key Vault firewallCKV_AZURE_109 | Quick start keeps it simple four folders | The key quick start leaves the vault reachable so it works from anywhere. Standard adds a default-deny firewall. The container evidence vault gets one once you name its addresses, so the Terraform runner is not locked out by default. |
| Key Vault private endpointCKV2_AZURE_32 | Needs a virtual network eight folders | A private endpoint needs a virtual network and private DNS these templates cannot assume. Standard closes the vault with a default-deny firewall and named addresses. |
| Key Vault public network access offCKV_AZURE_189 | Needs a virtual network eight folders | Switching it off entirely locks out whoever runs Terraform unless a private endpoint exists. Standard uses a default-deny firewall with named addresses. |
| Key Vault purge protectionCKV_AZURE_110 | Quick start keeps it simple one folder | The key quick start leaves purge protection off so it can be removed completely. Standard turns it on. |
| Key Vault recoverableCKV_AZURE_42 | Quick start keeps it simple one folder | Tied to purge protection: off in the key quick start so it can be removed, on in Standard. |
| Key backed by an HSMCKV_AZURE_112 | By design five folders | The key template's Standard level uses an HSM-backed key. Quick starts, and the storage keys inside other templates, use software keys that Key Vault still protects. |
| Key expiration dateCKV_AZURE_40 | By design seven folders | Each key carries a rotation policy that expires every version after a year and rotates thirty days before. A fixed expiration date on the key itself would fight that rotation. |
| Lambda code signingCKV_AWS_272 | Needs a virtual network four folders | Code signing needs a signing profile and a pipeline that signs each release. It belongs with the Hardened level's pipeline. |
| Lambda concurrency limitCKV_AWS_115 | By design four folders | Reserving concurrency fails in accounts with a low overall limit, which is common in new accounts. Leaving it unreserved means containment competes with other functions, which is worth knowing in a large account. |
| Lambda dead-letter queueCKV_AWS_116 | Quick start keeps it simple three folders | The quick start stays one file. Standard keeps every event the responder failed on in an encrypted queue for 14 days. |
| Lambda inside a VPCCKV_AWS_117 | By design four folders | The responder only calls AWS APIs. Placing it in a VPC would need endpoints or a NAT for each of them and adds a way for containment to fail. |
| Lambda tracingCKV_AWS_50 | By design four folders | Tracing adds cost and permissions. The responder logs every step, and its status messages record what happened in order. |
| Log group encrypted with KMSCKV_AWS_158 | Quick start keeps it simple eight folders | Quick starts keep the default log group encryption to stay one file. Standard encrypts these log groups with your own key: the evidence responder's logs, the Okta event archive, and the Bedrock invocation log. |
| Network security group ingress rules are statelessCKV_OCI_21 | By design three folders | A stateless rule needs a matching return rule and blocks nothing a stateful one allows. These groups accept SSH from one address, statefully. |
| Object Storage encrypted with your own keyCKV_OCI_9 | Quick start keeps it simple six folders | Quick starts use Oracle-managed keys to stay one file. Standard uses your own key from Vault. |
| Object Storage eventsCKV_OCI_7 | By design 10 folders | Nothing needs to react to new objects in these buckets. |
| Object Storage versioningCKV_OCI_8 | By design eight folders | Retention rules and versioning cannot both be on for one bucket in Oracle Cloud, and the retention rule is the control this template exists for. |
| Policy attached to a userCKV_AWS_40 | By design two folders | The break-glass user's policies are attached to it directly. A group would add a second way to grant administrator access, by adding someone to that group. |
| Pub/Sub topics encrypted with a customer-managed keyCKV_GCP_83 | By design six folders | Pub/Sub encrypts every message at rest with a Google-managed key. A key of your own adds the power to revoke access: SIEM ingest Standard takes one through kms_key_name, which the scanner cannot see when given as a variable. The other topics carry findings, alerts, and reminders, which describe resources and contain none of their data. |
| Retention locked with Bucket LockCKV2_GCP_4 | Quick start keeps it simple one folder | The quick start leaves the retention policy unlocked so it can be removed tomorrow. Standard has a lock switch. |
| S3 cross-region replicationCKV_AWS_144 | By design 18 folders | Replicating to a second region doubles storage cost and needs a second bucket and key. Some organisations need it, so it is left for them to add. |
| S3 encrypted with KMSCKV_AWS_145 | Quick start keeps it simple eight folders | Quick starts use S3-managed encryption to stay one file. Standard encrypts with your own KMS key. |
| S3 event notificationsCKV2_AWS_62 | By design 18 folders | Nothing needs to react to new objects in these buckets. |
| S3 lifecycle configurationCKV2_AWS_61 | By design six folders | The state quick start keeps every version, and Standard expires them. The evidence bucket has no expiry on purpose: how long evidence is kept is a legal hold decision, and Object Lock enforces the minimum. |
| S3 server access loggingCKV_AWS_18 | By design 18 folders | Server access logs need a second bucket and are delivered on a best-effort basis. CloudTrail data events record object access reliably, and the audit log template is where to switch them on. |
| S3 versioning on the SIEM failed-events bucketCKV_AWS_21 | By design six folders | Firehose writes each refused batch once, under its own unique key, and the batches expire after 30 days. There is never an older copy worth keeping. |
| SNS topic encryptedCKV_AWS_26 | By design six folders | Two cases. The key baseline's topic is deliberately not encrypted with the key it watches, because switching that key off would also silence the alert about it. Quick start topics stay unencrypted to keep them to one file. CloudWatch alarms cannot publish to a topic under the AWS managed key, so encryption needs a key with its own policy, which Standard adds. |
| Secrets Manager secret encrypted with your own keyCKV_AWS_149 | Quick start keeps it simple one folder | The quick start uses the AWS-managed Secrets Manager key to stay one file. Standard gives the secret its own key, usable only through Secrets Manager. |
| Security group not attachedCKV2_AWS_5 | By design two folders | The admin access template creates a security group with no inbound rules for the instances you launch afterwards. Nothing is attached to it at apply time. |
| Security list restricts SSH from 0.0.0.0/0CKV_OCI_19 | Scanner misreads the code three folders | The check counts a security list with no inbound rules as open to SSH. These lists have no inbound rules at all. |
| Session Manager data encrypted with your own keyCKV_AWS_112 | Quick start keeps it simple one folder | Sessions always run over TLS. The quick start leaves out the extra session key to stay one file. Standard encrypts each session with its own key. |
| Storage SAS expiration policyCKV2_AZURE_41 | By design two folders | Nothing here issues shared access signatures. The evidence accounts have shared keys off, so none can be made, and the function reaches its own account as its identity. |
| Storage account namingCKV_AZURE_43 | Scanner misreads the code one folder | The name is built from a variable, lowercased and trimmed to 24 characters, which the scanner cannot evaluate. |
| Storage account shared key access offCKV2_AZURE_40 | By design five folders | The flow log accounts receive flow logs from Network Watcher. Microsoft's flow log pages do not say flow logs are written with shared key access off, so it stays on until that is tested. The evidence capture function's own account keeps keys on for the quick start, which connects with one, and for the Functions runtime at Standard; the evidence account itself has them off. |
| Storage encrypted with your own keyCKV2_AZURE_1 | Quick start keeps it simple eight folders | Quick starts use Microsoft-managed keys to stay one file. Standard uses your own key from Key Vault for anything that contains evidence or logs. The evidence capture function's own account contains its code package and runtime state, and stays under Microsoft-managed keys. |
| Storage firewall denies by defaultCKV_AZURE_35 | By design one folder | The container evidence account denies every other network once allowed_ip_ranges is set. It stays reachable until then, because a responder outside a fixed list, such as in Cloud Shell, could otherwise not store evidence during an incident. Sign-in is Entra ID only, and responders can add blobs but not read or delete them. |
| Storage private endpointCKV2_AZURE_33 | Needs a virtual network 11 folders | A private endpoint needs a virtual network and private DNS these templates cannot assume. Standard closes the account with a default-deny firewall and named addresses. |
| Storage public network access offCKV_AZURE_59 | Needs a virtual network 11 folders | Blob public access is already off, and Standard accounts deny the network by default, except the container evidence account, which does once you name its addresses. Switching public network access off entirely needs a private endpoint. |
| Storage replicationCKV_AZURE_206 | Quick start keeps it simple seven folders | The state and container evidence quick starts use locally redundant storage to keep a trial cheap. Standard uses geo-redundant storage. |
| Storage soft deleteCKV2_AZURE_38 | Quick start keeps it simple five folders | The audit and container evidence quick starts lack it. There the immutability policy is what protects the logs and the evidence, and no blob can be deleted while it applies. The evidence capture function's own account has a code package Terraform rewrites on each change, and nothing worth restoring. |
| Subnet has a network security groupCKV2_AZURE_31 | Quick start keeps it simple one folder | The brokered shell quick start leaves AzureBastionSubnet on Azure's defaults to stay short. Standard gives it the rules Microsoft documents for Bastion. |
| VCN has an inbound security listCKV_OCI_16 | By design three folders | The brokered shell and forensic workstation subnets have no inbound rules on purpose. SSH arrives through a network security group that accepts the bastion's own address only. |
| VM disk encrypted with a customer-supplied keyCKV_GCP_38 | By design one folder | A customer-supplied key is raw key material sent with every disk operation, and losing it loses the disk. The forensic workstation's disk uses Google-managed encryption; a Cloud KMS key gives the same control over access without handling raw keys. |
| VPC flow logsCKV2_AWS_11 | Quick start keeps it simple one folder | The forensic workstation quick start leaves flow logs out to stay one file. Standard records every flow and alerts on each refused connection. |
What the checks cannot tell you
They show the code is well formed, uses settings that exist, and creates nothing obviously unsafe. They say nothing about your account, your permissions, or your policies. The check at the end of each template page is there for that.