The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill provides scripts and instructions that utilize sudo to perform restricted system actions, such as modifying the /etc/hosts file for DNS failure simulation and installing performance-testing packages like stress-ng.
[COMMAND_EXECUTION]: The skill generates and executes high-impact infrastructure commands to simulate failures. This includes terminating EC2 instances via the AWS CLI (aws ec2 terminate-instances), deleting Kubernetes pods and resources (kubectl delete), and triggering database failovers.
[EXTERNAL_DOWNLOADS]: The skill fetches external configuration files, specifically Kubernetes manifests from litmuschaos.github.io (a well-known chaos engineering project), and encourages the installation of third-party tools via brew and apt-get.
[INDIRECT_PROMPT_INJECTION]: The skill ingests external data such as system architecture descriptions and performance metrics to design experiments, creating a surface where malicious instructions embedded in system documentation could influence the generated failure injection logic.
Ingestion points: Reads experiment design templates and game day scenarios in references/experiment-design.md and references/game-days.md.
Boundary markers: No explicit boundary markers or warnings are used to instruct the agent to ignore potentially malicious content within input data.
Capability inventory: The skill has access to the filesystem, network (via requests), and powerful CLI tools (kubectl, aws, toxiproxy-cli).
Sanitization: There is no evidence of input validation or escaping for external parameters used in the generated code and manifests.
[DYNAMIC_EXECUTION]: The skill dynamically generates and executes failure injection scripts (Bash and Python) and Kubernetes Custom Resource manifests based on runtime user input and environment analysis.

chaos-engineer