The Agent Skills Directory

Infrastructure Best Practices (SAFE): The skill provides high-quality guidance on security, including non-root container users, read-only filesystems, dropped Kubernetes capabilities, and pinning GitHub Actions by SHA.
Indirect Prompt Injection (LOW): The skill enables an attack surface for indirect injection as it processes local directory contents using powerful infrastructure tools like kubectl, terraform, and helm.
Ingestion points: Local file system (templates/ and user-provided directories processed by the bash tool).
Boundary markers: Absent; no explicit instructions for the agent to verify the contents of the files before applying them to infrastructure.
Capability inventory: Use of Bash to execute kubectl apply, helm upgrade, and terraform apply (SKILL.md, MAKEFILE.md, TERRAFORM.md).
Sanitization: Absent; the skill assumes the provided infrastructure manifests are trustworthy.
Credentials (SAFE): Examples in KUBERNETES.md use generic placeholders like postgres://user:pass@host/db, which is standard illustrative documentation and does not constitute a secret exposure.

managing-infra