The Agent Skills Directory

[SAFE]: The skill serves as a technical guide for the Atmos policy engine, providing procedural steps for creating policy manifests and test cases.
[COMMAND_EXECUTION]: The skill includes shell commands for local development testing, such as go test ./pkg/policy -run TestPolicyCatalog -v. These are standard unit testing patterns used to verify policy logic against provided fixtures.
[INDIRECT_PROMPT_INJECTION]: The skill defines a surface for processing external data such as Terraform plans and GitHub Actions events.
Ingestion points: Data is ingested from terraform.plan JSON files, GitHub Actions event metadata, and atmos.affected inputs.
Boundary markers: The skill mandates the use of strict kind: Policy documents and specifies that evaluation occurs over detached, read-only JSON data.
Capability inventory: Capabilities include running local Go tests and triggering defined lifecycle hooks (action: hooks).
Sanitization: The skill explicitly states that the CEL and Starlark execution environments are restricted and do not have access to the filesystem, network, or automation host APIs, which effectively mitigates the risk of malicious data influencing the host system.

atmos-policy