The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill retrieves Kubernetes Custom Resource Definition (CRD) schemas from the official GitHub repository of Datree, a well-known service in the Kubernetes ecosystem.
Evidence: The kubeconform command in SKILL.md (Stage 4) uses -schema-location 'https://raw.githubusercontent.com/datreeio/CRDs-catalog/...'.
[EXTERNAL_DOWNLOADS]: The skill automatically handles missing dependencies by installing the widely-used pyyaml library from the Python Package Index (PyPI) into a temporary virtual environment.
Evidence: pip install --quiet pyyaml is executed within scripts/detect_crd_wrapper.sh.
[COMMAND_EXECUTION]: The skill executes local validation binaries (yamllint, kubeconform, kubectl) and bundled Python scripts to analyze manifests. These operations are limited to the primary purpose of the skill.
Evidence: Commands defined in SKILL.md under Stages 0, 1, 2, 4, and 5.
[INDIRECT_PROMPT_INJECTION]: As the skill ingests and processes untrusted Kubernetes manifests, it includes specific guardrails to prevent instructions within those files from influencing the agent's behavior.
Evidence: SKILL.md mandates a "Read-Only Boundary," instructing the agent to never modify files or offer to apply fixes automatically, limiting output to reports only.
[SAFE]: The use of subprocess.run in scripts/test_count_yaml_documents.py is identified as benign, as it is used exclusively for internal regression testing of the skill's own components using local test fixtures.

k8s-yaml-validator