chaos-engineering

CRITICAL E004: Prompt injection detected in skill instructions.

• Potential prompt injection detected (high risk: 0.90). The embedded "Companion check" directs the agent to run local filesystem commands and offer to install companion skills—actions unrelated to chaos-engineering and effectively a hidden/unauthorized behavior outside the skill's stated purpose, so it constitutes a prompt injection.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The skill includes an explicit runtime command that fetches and applies remote Kubernetes YAML — "kubectl apply -f https://hub.litmuschaos.io/api/chaos/3.0.0?file=charts/generic/pod-delete/experiment.yaml" — which would retrieve external content and execute it in-cluster, meeting the criteria for a risky runtime dependency.

MEDIUM W013: Attempt to modify system services in skill instructions.

• Attempt to modify system services in skill instructions detected (high risk: 1.00). The skill explicitly instructs running privileged and state-changing commands (e.g., "sudo tc qdisc add ..." to modify network qdiscs, kubectl/pg_ctl/aws failover operations, and a companion check that runs filesystem commands and suggests installing skills via npx), which directs the agent to alter the host or infrastructure state and potentially require sudo/privileged access.