The Agent Skills Directory

[COMMAND_EXECUTION]: The script references/autoscaling-script.md and the ConfigMap script in references/aws-auto-scaling.md use shell variables to construct and execute infrastructure commands. If these variables are supplied from untrusted sources without validation, they could be exploited for command injection.
Evidence: kubectl get hpa ${DEPLOYMENT}-hpa -n $ENVIRONMENT -w in references/autoscaling-script.md uses unvalidated positional arguments.
Evidence: aws autoscaling create-auto-scaling-group --auto-scaling-group-name "$ASG_NAME" in references/aws-auto-scaling.md interpolates variables directly into shell commands.
[INDIRECT_PROMPT_INJECTION]: The skill provides an attack surface where untrusted input (via script arguments or configuration files) could influence the behavior of privileged tools like kubectl and aws CLI.
Ingestion points: Positional arguments $1 and $2 in references/autoscaling-script.md and $1 in scripts/validate-config.sh.
Boundary markers: Absent. No explicit validation or "ignore embedded instructions" logic is present for the input variables.
Capability inventory: The skill utilizes kubectl apply, kubectl get, and multiple aws autoscaling commands which have high impact on the infrastructure.
Sanitization: While the scripts use set -euo pipefail for execution safety, they lack validation or sanitization of the content within the variables before interpolation into YAML manifests or CLI flags.

autoscaling-configuration