chaos-experiment-designer

Design DSDS chaos experiments with deterministic setup, explicit hypotheses, and measurable outcomes. Use when planning failure drills, resilience validation, pre-production risk exercises, and repeat

SKILL.md

GitHub

Chaos Experiment Designer

Overview

Create focused chaos experiments that test one resilience claim at a time. Ensure each experiment is reproducible and has clear pass/fail criteria.

Workflow

  1. Define the hypothesis
  • State one reliability claim in measurable form.
  • Identify target metrics and acceptable bounds.
  1. Define steady state
  • Specify baseline workload and expected healthy metric band.
  • Establish pre-fault observation window.
  1. Design fault injection
  • Pick minimal fault set to test the hypothesis.
  • Set trigger timing and fault duration explicitly.
  1. Define safeguards
  • Add blast radius limits and abort conditions.
  • Identify critical user journeys that must stay protected.
  1. Define evaluation plan
  • Add pass/fail rules tied to SLOs and invariants.
  • Include recovery expectations (MTTR, error decay, queue normalization).
  1. Package experiment
  • Return experiment spec, run steps, success criteria, and follow-up checks.

Output format

  1. Hypothesis
  2. Steady-state definition
  3. Fault plan
  4. Safety and abort rules
  5. Pass/fail criteria
  6. Follow-up actions

References

  • Read references/chaos-experiment-template.md for reusable experiment skeletons.
  • Use these repo sources:
  • docs/05-devs-chaos-and-analysis.md (Chapter 27)
  • canonical-catalogue/DSDS Canonical Catalogue - scenarios to ship.csv
  • canonical-catalogue/DSDS Canonical Catalogue - Failure modes & propagation semantics.csv