Produce read-only cloud security and FinOps findings with StackQL
Run StackQL’s pinned Docker audit against approved cloud scopes and return security findings, waste estimates and coverage gaps from its JSON artifact.
npx skills add agentskillexchange/skills --skill produce-read-only-cloud-security-finops-findings-stackql
Use this skill for a repeatable, read-only security and FinOps evidence pass across an explicitly approved AWS, GCP or Azure/Entra scope. The agent runs StackQL’s packaged Docker audit and turns its findings JSON into a triage handoff. It is useful when the operator needs the same checks and a retained artifact across clouds instead of assembling console screenshots or inventing SQL queries for each run.
What this skill actually does
Use the cited StackQL checkout, Docker and Compose. Copy cicd/audit/.env.audit.example to cicd/audit/.env.audit, then configure only the intended clouds with read-only credentials. Specify the relevant region, organization or management-group scope; understand that omitted Azure management-group scope can expand to the tenant. Unconfigured clouds are skipped and must be reported as unassessed.
Preserve the audit engine commit and provider version pins in the example configuration. Select and record a verified compatible StackQL image tag or digest. Run docker compose -f docker-compose.audit.yaml pull, then docker compose -f docker-compose.audit.yaml run --rm audit. The query service runs in the Compose network and the audit service writes cicd/audit/output/findings.json. The engine is downloaded at runtime from stackql/stackql-audit-action at the configured revision; it is not bundled into the main repository.
Read the findings artifact and return the assessed scope, failed or skipped coverage, affected resources, supporting checks and estimated waste where present. The pinned engine includes checks for unattached EBS volumes and Azure Advisor cost recommendations. Distinguish estimates and provider recommendations from measured savings. Keep original finding identifiers and evidence so a person can verify the report before taking action. A failed run or missing artifact is an execution gap, not a clean audit.
The boundary is one packaged findings run and its reviewable handoff. Do not execute remediation snippets, delete unused resources, resize services, provision infrastructure or enable arbitrary SQL writes. This extends the existing generic SQL inventory lane with a preconfigured audit engine and combined security/FinOps artifact; it does not replace provider consoles or become a general StackQL administration listing.
Multi-Framework is justified by the Docker/Compose command and JSON artifact interface, which any command-capable agent can invoke. MCP is optional and not used by this route. The main repository’s adoption supports the StackQL identity; no separate popularity claim is made for the audit engine. Metadata, pinned configuration, bootstrap and sample check sources were inspected, but cloud access, image compatibility, installation and runtime behavior were not tested. Metadata verification is not a security review.