The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill instructions include usage of sudo to perform high-privilege system modifications, such as installing the stress-ng utility and modifying the /etc/hosts system file to simulate network failures.
[EXTERNAL_DOWNLOADS]: Automated workflows in the skill fetch and apply remote Kubernetes manifests directly from litmuschaos.github.io, an external source not explicitly defined in the trusted vendor environment.
[COMMAND_EXECUTION]: The skill relies on powerful CLI tools like kubectl, aws-cli, and toxiproxy-cli to execute destructive actions, including instance termination, pod eviction, and network traffic manipulation as part of its core resilience testing functionality.
[INDIRECT_PROMPT_INJECTION]: Potential vulnerability surface detected where the agent processes external, untrusted data to make decisions.
Ingestion points: The skill retrieves runtime data from Prometheus API queries and Kubernetes ChaosResult statuses.
Boundary markers: The instructions lack explicit delimiters or warnings to ignore embedded instructions within the processed metric data.
Capability inventory: The skill possesses capabilities to modify infrastructure state via aws ec2 terminate-instances and kubectl apply/delete operations.
Sanitization: There is no evidence of sanitization or strict schema validation for external data before it is used to determine if an experiment should be rolled back or considered successful.

chaos-engineer