The Agent Skills Directory

[CREDENTIALS_UNSAFE]: Hardcoded plaintext credentials detected in local files.
Evidence: password = "SuperSecret123!" in examples/vulnerable_terraform.tf.
Context: This credential resides inside an intentionally vulnerable mockup file designed to test IaC scanning engines and verify rule sets, representing low real-world exposure.
[DATA_EXFILTRATION]: Use of remote web APIs to parse infrastructure manifests outside the whitelisted environments.
Evidence: curl -sSX POST --data-binary @deployment.yaml https://v2.kubesec.io/scan in references/kubernetes_manifests.md.
Context: Transmitting infrastructure configuration blueprints to external third-party hosts poses metadata exposure risks if the scanned resource structures contain sensitive corporate architectural information.
[INDIRECT_PROMPT_INJECTION]: Vulnerability surface arising from ingestion of untrusted infrastructure definitions.
Ingestion points: The tool ingests workspace files including .tf, .yaml, and .json files via multiple workflow pathways.
Boundary markers: No parsing boundaries or text encasements are defined to isolate embedded payload directives from standard execution strings.
Capability inventory: Possesses extensive execution access to call external command line binaries (checkov, tfsec, terrascan, kube-linter, cfn-lint).
Sanitization: No filter controls are enforced over ingested file comment headers or manifest fields before forwarding them to scanners.

iac-security