The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides scripts and instructions that perform destructive infrastructure operations, such as aws ec2 terminate-instances for host termination and kubectl delete for removing Kubernetes components.
[PRIVILEGE_ESCALATION]: Multiple functions in references/infrastructure-chaos.md require administrative access via sudo, specifically for modifying /etc/hosts to simulate DNS failures and for installing software using system package managers.
[EXTERNAL_DOWNLOADS]: The skill downloads and applies remote Kubernetes manifests from the Litmus Chaos GitHub repository and encourages the installation of third-party binary tools, introducing potential supply chain risks.
[DATA_EXFILTRATION]: Automated reputation scanners flagged the primary documentation URL (https://jeffallan.github.io/claude-skills/skills/devops/chaos-engineer/) as blacklisted, posing a risk of redirection to malicious sites or data exfiltration.
[INDIRECT_PROMPT_INJECTION]: The execution flow relies on external telemetry data from systems like Prometheus and Kubernetes status APIs to validate experiment safety, which creates an attack surface.
Ingestion points: Telemetry values from Prometheus API queries and resource state descriptions from Kubernetes.
Boundary markers: Absent; the agent is instructed to process raw output from monitoring tools without protective delimiters.
Capability inventory: Highly privileged tools including AWS CLI and kubectl with delete permissions, and sudo for system configurations.
Sanitization: No validation or sanitization is performed on external metrics before they are used to control experiment triggers and safety rollbacks.

chaos-engineer