Skip to content

How it works

Awnexa has one job: to gather the facts about your AWS configuration and keep track of how they change. Judging those facts — reviews, suggestions, IaC, finding the cause of an incident — is left to Claude Code.

A scan reads your regions in parallel and saves the result as a snapshot. Global services such as IAM, CloudFront and Organizations are read only once.

For every API and region, Awnexa records whether collection was complete, partial, access_denied, failed or unsupported — or not_applicable (the service is not offered or not enabled in that region), skipped (excluded with --services) or unknown (not run, for example because the scan was stopped). If a count is unknown it is null, not 0. An AI agent that reads the coverage will say “not checked” instead of “does not exist”.

Awnexa keeps the API response behind each resource, with secret values masked. Every fact has a confidence level:

Confidence Meaning
direct Returned by an AWS API
derived Computed mechanically; the rule that produced it is recorded
inferred An estimate, never stored as a fact

Security group rules are recorded as what is allowed, not as evidence of actual traffic.

Awnexa builds a graph from ID and ARN references and records whether each target was found in the scan (resolved), referenced but not collected (unresolved) or outside the account (external). An environment variable of a Lambda function or ECS task definition whose value names a resource (an SQS queue URL, an RDS endpoint, an ARN) becomes an inferred relationship, and so does a bare name such as TABLE_NAME=orders when exactly one resource of that type in the scan has the name; only the variable name is kept, never the value. awnexa impact lists what depends on a resource, with the paths and the limits of what was scanned.

awnexa snapshot diff shows what was created, modified or deleted between two scans. A deletion is never confirmed from a partial scan; it is reported as unknown. Relationships added or removed between resources present in both scans are listed too, and are left unconfirmed when the scan without the relationship did not collect its source completely.

awnexa scan --metrics also reads a usage summary: 14 days of daily values for a few CloudWatch metrics per resource (requests, invocations, connections, traffic, utilization, storage) of EC2 instances, NAT gateways, RDS, Lambda, SQS, SNS, load balancers, target groups, API Gateway, DynamoDB, ECS services, S3 buckets and CloudFront. They help Claude Code tell which dependants of a resource were active. They are kept apart from configuration and are not part of snapshot diffs. CloudWatch bills GetMetricData per metric requested, so this is off by default; awnexa status shows an estimate for the latest scan and the scan output reports what was requested.

Custom metrics and fine-grained metric time series, log contents and CloudTrail events, S3 objects, database data, and secret or SecureString values. Secrets Manager, Systems Manager parameters and KMS are read as metadata only.