prowler-sdk-check
Guides the creation and update of Prowler security checks using standardized SDK patterns.
Install
mkdir -p .claude/skills/prowler-sdk-check && curl -L -o skill.zip "https://agentskills.codes/api/skills/download/6540" && unzip -o skill.zip -d .claude/skills/prowler-sdk-check && rm skill.zipInstalls to .claude/skills/prowler-sdk-check
Activation
This is the description your AI agent reads to decide when to run this skill — the better it matches your request, the more reliably it fires.
Creates Prowler security checks following SDK architecture patterns. Trigger: When creating or updating a Prowler SDK security check (implementation + metadata) for any provider (AWS, Azure, GCP, K8s, GitHub, etc.).Key capabilities
- →Generate check boilerplate code
- →Create mandatory metadata JSON files
- →Integrate service-level clients
- →Implement report object models
- →Structure provider-specific security checks
How it works
It follows a strict directory pattern to generate implementation and metadata files that satisfy the Prowler SDK check architecture.
Inputs & outputs
When to use prowler-sdk-check
- →Implement a new security check for AWS or Azure
- →Generate metadata files for security scanning
- →Integrate service-level methods for data collection
- →Standardize check logic using SDK models
About this skill
Check Structure
prowler/providers/{provider}/services/{service}/{check_name}/
├── __init__.py
├── {check_name}.py # class {check_name}(Check)
├── {check_name}.metadata.json
└── {check_name}_fixer.py # optional, see Fixers
Naming: {service}_{resource}_{security_control} (e.g. s3_bucket_object_versioning). Tests go in tests/providers/{provider}/services/{service}/{check_name}/{check_name}_test.py (see prowler-test-sdk).
Hard Rules (enforced in code)
Check.__init__ and CheckMetadata (prowler/lib/check/models.py) reject the check at load time if any of these fail:
| Rule | Enforced by |
|---|---|
CheckID == class name == file name (without .py) | Check.__init__ |
CheckID has no - | valid_check_id |
ServiceName == CheckID.split("_")[0] and lowercase (so the AWS Lambda service is awslambda) | validate_service_name |
CheckTitle <= 150 chars and does not start with Ensure | validate_check_title |
Description and Risk <= 400 chars each | validate_description, validate_risk |
CheckType is [] for every provider except AWS; AWS values must be paths in prowler/providers/aws/config/check_types.json | validate_check_type |
RelatedUrl is "" (deprecated, use AdditionalURLs) | validate_related_url |
Remediation.Recommendation.Url is empty or starts with https://hub.prowler.com/ (use https://hub.prowler.com/check/{check_id}) | validate_recommendation_url |
Remediation.Code.CLI is not a URL | valid_cli_command |
ResourceType non-empty | valid_resource_type |
ResourceGroup in VALID_RESOURCE_GROUPS (or "") | validate_resource_group |
Every Categories item in VALID_CATEGORIES | valid_category |
AdditionalURLs has no empty or duplicate items | validate_additional_urls |
Severity in critical, high, medium, low, informational | Severity enum |
Tool-wrapper providers (iac, llm, image, or plug-ins with is_external_tool_provider = True) skip most of these.
Validate a metadata file without running a scan:
uv run python -c "import sys; from prowler.lib.check.models import CheckMetadata; CheckMetadata.parse_file(sys.argv[1]); print('OK')" prowler/providers/{provider}/services/{service}/{check_name}/{check_name}.metadata.json
Check Implementation
from prowler.lib.check.models import Check, Check_Report_AWS
from prowler.providers.aws.services.s3.s3_client import s3_client
class s3_bucket_object_versioning(Check):
"""S3 buckets have object versioning enabled."""
def execute(self) -> list[Check_Report_AWS]:
findings = []
for bucket in s3_client.buckets.values():
report = Check_Report_AWS(metadata=self.metadata(), resource=bucket)
if not bucket.versioning_retrieved:
report.status = "MANUAL"
report.status_extended = f"Cannot evaluate versioning for S3 Bucket {bucket.name}: the versioning configuration could not be retrieved. Verify that the scanning credentials are allowed to call s3:GetBucketVersioning."
elif bucket.versioning:
report.status = "PASS"
report.status_extended = f"S3 Bucket {bucket.name} has versioning enabled."
else:
report.status = "FAIL"
report.status_extended = f"S3 Bucket {bucket.name} has versioning disabled."
findings.append(report)
return findings
- Only use the service client (
{service}_client); never call the provider SDK from a check. If data is missing, extend the service (seeprowler-provider). execute()returns a list (empty if there are no resources), one report per audited resource.status_extendedmust end with a period and identify the resource.resource=accepts a dict, a Pydantic model, a dataclass or an object withto_dict(); anything else becomes{}and logs an error.- Severity can be adjusted per finding with
report.check_metadata.Severity = "..."when the risk depends on the scenario. - Secret-scanning checks use a dedicated batched structure: read
docs/developer-guide/secret-scanning-checks.mdxfirst.
Report Class per Provider (prowler/lib/check/models.py)
| Provider | Class | Fields set from resource |
|---|---|---|
| aws | Check_Report_AWS | resource_id (id or name), resource_arn, region |
| azure | Check_Report_Azure | resource_name, resource_id, location; set report.subscription yourself |
| gcp | Check_Report_GCP | resource_id, resource_name, project_id, location; also accepts location=, resource_name=, resource_id=, project_id= kwargs |
| kubernetes | Check_Report_Kubernetes | resource_id (uid), resource_name, namespace |
| oraclecloud | Check_Report_OCI | resource_id, resource_name, compartment_id, region |
| all others | CheckReport<Name> (e.g. CheckReportGithub, CheckReportM365, CheckReportStackIT, CheckReportMongoDBAtlas, CheckReportAlibabaCloud) | grep ^class CheckReport in models.py for the exact name and constructor kwargs |
Statuses
| Status | When |
|---|---|
PASS | Resource is compliant |
FAIL | An insecure condition was actually detected |
MANUAL | Human verification needed, or the data could not be retrieved |
Permission / availability errors are NOT findings
Never FAIL because an API call failed (missing permission or scope, API not enabled, feature not licensed). The service must expose the failure distinctly from an empty result (None instead of [], a *_retrieved / *_unavailable flag, an *_error attribute or a *_lookup_failed set). A 404/not-found usually means "not configured" and IS a legitimate FAIL; a definitively disabled API is a legitimate FAIL only when enabling that API is the audited control (e.g. GCP Access Approval).
When a whole dataset is missing, emit ONE account/tenant/project-level MANUAL finding naming the required permission, and leave Severity untouched:
if {service}_client.{data} is None:
report = CheckReport{Provider}(metadata=self.metadata(), resource={})
report.resource_name = "<Tenant-level resource>"
report.resource_id = "<stable-id>"
report.status = "MANUAL"
report.status_extended = "Cannot evaluate <requirement>: <data> could not be retrieved. Verify that <permission> is granted to the scanning identity."
return [report]
Metadata
Complete, validated examples live in assets/. Writing style: docs/developer-guide/check-metadata-guidelines.mdx.
| Field | Rule |
|---|---|
CheckTitle | Declarative compliant state, starts with the resource, no "Check/Verify/Ensure". Singular when one report per resource ("S3 bucket has object versioning enabled") |
CheckType | AWS only, e.g. "Software and Configuration Checks/AWS Security Best Practices", "Effects/Data Exposure", "TTPs/Initial Access"; partial paths allowed. [] elsewhere |
SubServiceName, ResourceIdTemplate | Being deprecated: "" |
ResourceType | AWS: Security Hub type (AwsS3Bucket) or Other. Azure: Resource Graph type. GCP: Cloud Asset type (storage.googleapis.com/Bucket). Kubernetes: KIND. Providers without a standard: NotDefined |
ResourceGroup | compute, container, serverless, database, storage, network, IAM, messaging, security, monitoring, api_gateway, ai_ml, governance, collaboration, devops, analytics. Chosen by the audited resource, not the service (EC2 security group -> network) |
Description | What the finding means, Markdown allowed, no risk talk, no "pass/fail" words |
Risk | Concrete CIA impact and attack paths, not generic "security issues" |
Remediation.Code.CLI | One raw command that flips FAIL to PASS, or "" |
Remediation.Code.NativeIaC | AWS: CloudFormation YAML code block. Azure: Bicep. Minimal, comment the fixing line |
Remediation.Code.Terraform | Minimal HCL code block, comment the fixing line |
Remediation.Code.Other | Numbered console steps |
Remediation.Recommendation.Text | Principle-based guidance (least privilege, defense in depth), <= 400 chars |
Categories | Only values from VALID_CATEGORIES (e.g. encryption, internet-exposed, logging, secrets, resilience, threat-detection, trust-boundaries, vulnerabilities, identity-access, forensics-ready, privilege-escalation, gen-ai, ci-cd). A new category needs a validator change plus a docs update |
DependsOn, RelatedTo | Lists of CheckIDs, usually [] |
Configurable Checks
max_rules = ec2_client.audit_config.get("max_security_group_rules", 50) # always pass a default
Adding a new variable requires ALL of:
- Default value under the provider key in
prowler/config/config.yaml. - Typed
Optional[...] = Nonefield inprowler/config/schema/{provider}.py(providers are registered inprowler/config/schema/registry.py). Values that fail validation are dropped with a warning; misspelled keys pass through silently. tests/config/fixtures/config.yamlif tests depend on it.- Docs:
docs/user-guide/cli/tutorials/configuration_file.mdx.
In tests pass the value with set_mocked_aws_provider(audit_config={...}) (or the provider fixture equivalent).
Fixers
- File
{check_name}_fixer.pynext to the check with a module-leveldef fixer(...) -> bool(True on success; log and return False on error). Docstring lists the permissions needed. run_fixer(prowler/lib/check/check.py) picks the arguments by parameter name:resource_id+region,regiononly,resource_arn, orresource_id.- Fixer settings come from
prowler/config/fixer_config.yaml(override with--fixer-config) through{service}_client.fixer_config. - The
--fixerflag is only registered for AWS (`prowler/provider
Content truncated.
When not to use it
- →Non-security feature development
- →UI component creation
Limitations
- →Restricted to Prowler SDK pattern
- →Requires manual implementation of service-level data collection
- →Boilerplate code must be extended with business logic
How it compares
It standardizes the check implementation process, ensuring every new security check is automatically compatible with the existing Prowler framework.
Compared to similar skills
prowler-sdk-check side by side with the closest alternatives in the catalog.
| Skill | Installs | Updated | Safety | Difficulty |
|---|---|---|---|---|
| prowler-sdk-check (this skill) | 1 | 5mo | Review | Intermediate |
| cloud-penetration-testing | 3 | 9mo | Review | Advanced |
| building-cloud-siem-with-sentinel | 0 | 5mo | Review | Advanced |
| azure-cloud-security-review | 0 | 7mo | Review | Advanced |
Try saying
Example prompts that trigger this skill in your AI assistant.
More by prowler-cloud
View all by prowler-cloud →You might also like
cloud-penetration-testing
davila7
This skill should be used when the user asks to "perform cloud penetration testing", "assess Azure or AWS or GCP security", "enumerate cloud resources", "exploit cloud misconfigurations", "test O365 security", "extract secrets from cloud environments", or "audit cloud infrastructure". It provides comprehensive techniques for security assessment across major cloud platforms.
building-cloud-siem-with-sentinel
26zl
This skill covers deploying Microsoft Sentinel as a cloud-native SIEM
azure-cloud-security-review
shyamagu-ms
>
performing-cloud-forensics-investigation
yanacuti1121
Conduct forensic investigations in cloud environments by collecting and
cloud-iam-deep
elementalsouls
Cloud IAM red-team attack chain across AWS, Azure, GCP — focused on EXTERNAL exploitation paths and post-credential-discovery privilege analysis. Covers IAM enumeration (aws iam, az role, gcloud iam), STS/AssumeRole chaining, Azure Managed Identity abuse (via SSRF/leak), GCP service account JSON abu
cloud-defense
transilienceai
Detect and break the cloud post-compromise attack chain (AWS / Azure / GCP) — per-stage CloudTrail / Activity-Log / Audit-Log detection signals and the preventive controls that close each step. Use for cloud detection engineering, hardening, remediation write-ups, or blue-team posture review of the