The Agent Skills Directory

[CREDENTIALS_UNSAFE]: Multiple scripts within the skill (deploy-service.sh, diagnose.sh, install.sh, and traffic-split.sh) are designed to access sensitive Kubernetes configuration files at specific paths.
Evidence: The scripts construct paths to kubeconfig files using the pattern $HOME/.kube/aks-rg-hypera-${CLUSTER}-config and export them to the KUBECONFIG environment variable, exposing cluster credentials.
[COMMAND_EXECUTION]: The main skill instructions include the use of sudo for administrative operations, which is a high-privilege action.
Evidence: SKILL.md instructs the user/agent to execute sudo mv func /usr/local/bin/ to install a downloaded binary.
[REMOTE_CODE_EXECUTION]: The skill facilitates the download and execution of external binary files, which bypasses package manager security checks.
Evidence: SKILL.md provides commands to fetch the func binary from GitHub releases using curl, apply executable permissions with chmod +x, and move it to a system path.
[EXTERNAL_DOWNLOADS]: The skill relies heavily on fetching configuration and installation manifests from external remote sources during runtime.
Evidence: The install.sh script and SKILL.md perform numerous kubectl apply -f operations targeting various GitHub repository URLs (e.g., https://github.com/knative/serving/releases/download/...).

knative