The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes several shell commands (git, kubectl, argocd, pre-commit) using variables such as and derived from user input. These strings are interpolated into commands without documented sanitization, which could lead to command injection if malicious input is provided. Evidence: kubectl get applicationset -n argocd <service-name> and git add argo-cd-helm-values/kube-addons/<service-name>/ in Workflows/CreateApplicationSet.md.- [COMMAND_EXECUTION]: The workflow runs pre-commit run --all-files, which executes hooks defined in the repository configuration. This is a potential vector for local code execution if the configuration file is untrusted. Evidence: Workflows/CreateApplicationSet.md and Workflows/ValidateApplicationSet.md.- [PROMPT_INJECTION]: The skill presents an indirect prompt injection surface as it ingests user-provided metadata to generate Kubernetes manifests and execute system commands.
Ingestion points: Service names, chart URLs, and version strings provided by users in Workflows/CreateApplicationSet.md.
Boundary markers: None; user input is used directly in file paths and command arguments.
Capability inventory: File system writes (mkdir, touch), Git operations (add, commit), and Kubernetes CLI execution (kubectl, argocd).
Sanitization: The skill does not specify any validation or escaping of user-provided strings.- [EXTERNAL_DOWNLOADS]: The skill communicates with a local notification service via curl at http://localhost:8888/notify. Evidence: SKILL.md and workflow files.- [EXTERNAL_DOWNLOADS]: The skill references an external Git repository on Azure DevOps for configuration synchronization. Evidence: https://hyperadevops@dev.azure.com/hyperadevops/devops-team/_git/argocd in ClusterInventory.md.

ArgocdAppInstall