Chaos Engineer helps Claude plan and write failure injection experiments for distributed systems. Each experiment starts with a steady state and a hypothesis, limits its blast radius, and has a tested abort path before it begins.
The skill includes worked examples for pod deletion with Litmus Chaos, network latency with toxiproxy and instance kills with Chaos Monkey, and covers game day planning.
What You Get
- An experiment design with hypothesis, metrics and blast radius
- Failure injection manifests and scripts, such as a Litmus ChaosEngine
- Monitoring and alert setup for the experiment
- Rollback procedures and safety controls
- A learning summary with tracked improvements
Common Use Cases
- Testing how a service behaves when a third of its pods are deleted
- Adding latency to a database dependency
- Planning and running a game day
- Putting continuous chaos tests into a CI/CD pipeline
- Checking zone or region failure handling
What Makes It Unique
Safety comes first: steady state before injection, the smallest possible blast radius, one variable at a time, and no production run without circuit breakers, feature flags or canary isolation. Reference files cover experiment design, infrastructure failures, Kubernetes chaos, tooling and game days.
Under The Hood
The skill is a SKILL.md with five reference files and no scripts of its own. Its examples tell the agent to run kubectl commands to read deployments and logs, apply a ChaosEngine manifest and patch it to stop, and to install and run toxiproxy and Chaos Monkey. That means it runs shell commands against your cluster and can inject real failures, so the blast radius limits in the skill matter. It writes manifests and config files in your project and does not fetch web pages.