The Agent Skills Directory

[COMMAND_EXECUTION]: The skill contains various scripts and instructions for executing powerful administrative commands necessary for chaos testing. These include terminating cloud instances (aws ec2 terminate-instances), deleting Kubernetes resources (kubectl delete), and using network traffic control tools (toxiproxy-cli, pumba) to simulate failures.
[PRIVILEGE_ESCALATION]: Certain utility scripts use sudo to perform operations that require elevated permissions, such as installing stress testing packages (apt-get install -y stress-ng) and modifying the system's local resolution file (/etc/hosts) to simulate DNS outages.
[EXTERNAL_DOWNLOADS]: The skill references and downloads configuration manifests from external sources, specifically the Litmus Chaos operator deployment file from its official project documentation site (litmuschaos.github.io).
[INDIRECT_PROMPT_INJECTION]: The skill includes an attack surface where it ingests external metric data from monitoring systems (like Prometheus) and possesses the capability to execute infrastructure changes. If these external metrics were manipulated by an adversary, they could potentially trigger automated chaos experiments unexpectedly.
Ingestion points: Metric values are retrieved from Prometheus and Datadog API endpoints within the monitoring and dashboard scripts (references/chaos-tools.md).
Boundary markers: None identified in the provided monitoring script examples.
Capability inventory: The skill has access to the AWS SDK (boto3) for instance management, the Kubernetes client for resource orchestration, and shell access for resource stress testing.
Sanitization: The scripts do not explicitly demonstrate sanitization of the metric data retrieved before using it in decision-making logic.

chaos-engineer