The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of powerful infrastructure management and failure injection tools via subprocess commands.
Evidence: Uses aws-cli for critical operations like aws ec2 terminate-instances and aws rds reboot-db-instance --force-failover in references/chaos-tools.md and references/game-days.md.
Evidence: Uses kubectl for resource manipulation, including kubectl apply for remote manifests and kubectl patch to modify resource states in SKILL.md and references/kubernetes-chaos.md.
Evidence: Employs resource exhaustion tools such as stress-ng, iperf3, and pumba to simulate system failures.
[PRIVILEGE_ESCALATION]: Several scripts require administrative privileges to perform system-level modifications.
Evidence: Instructions in references/infrastructure-chaos.md use sudo to install software (sudo apt-get install -y stress-ng) and modify core system files (sudo sh -c 'echo "127.0.0.1 {domain}" >> /etc/hosts') for DNS hijacking simulation.
[EXTERNAL_DOWNLOADS]: Fetches configuration and deployment manifests from remote sources.
Evidence: Downloads the Litmus Chaos operator directly from https://litmuschaos.github.io/litmus/litmus-operator-v2.14.0.yaml for installation via kubectl apply.
[DATA_EXFILTRATION]: The skill uses network libraries to interact with external APIs and monitoring services, which involves sending system and experiment data.
Evidence: Python scripts in references/chaos-tools.md and references/infrastructure-chaos.md use the requests library to communicate with api.gremlin.com and a Prometheus metrics server.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted data from external sources during its workflow.
Ingestion points: Reads application logs via kubectl logs and fetches metrics from Prometheus APIs to determine experiment verdicts.
Boundary markers: No explicit delimiters or instructions are used to separate untrusted data from the agent's logic.
Capability inventory: Possesses significant capabilities including cloud provider CLI access (aws), cluster administration (kubectl), and system-level file modification (sudo).
Sanitization: Lacks evidence of validation or sanitization for data retrieved from external logs or metric queries before processing.

chaos-engineer