研究与检索SKILL.md
故障实验设计
v1在受控环境中规划故障假设、实验边界和观测指标。
技能概述
这项技能做什么
在受控环境中规划故障假设、实验边界和观测指标。
对应原始技能标识:chaos-engineer。本站整理中文说明,技能原文由 Jeffallan/claude-skills 维护。
适用场景与输入
分类:研究与检索。建议准备:研究问题、已知资料、比较对象、证据范围和判断标准。
获取与使用
- 打开固定版本的完整技能目录,按仓库说明获取技能。
- 原目录包含
references等配套资源。请获取完整技能目录,单独下载本站的 SKILL.md 不包含这些文件。 - 在 Claude Code 或具备相应工具能力的兼容 Agent 中加载;兼容 SKILL.md 格式不代表各平台工具完全一致。
- 参照页面的使用示例提供材料,按照原文步骤执行,并核对实际输出。
输出与验证
区分事实、假设与推断;按证据组织结论,列出局限和下一步验证建议。本站未执行来源仓库中的脚本,也不把示例当作已经完成的运行结果。
来源与许可
- 原始 SKILL.md
- 完整技能目录及配套资源
- 固定提交:
078a25cc5bfd8011459be7b91d004e900c1d02d3 - 许可证:MIT;技能原文和下载内容中保留以下版权与许可说明。
MIT License
Copyright (c) 2025
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.
如何使用这个 Skill
下载 SKILL.md,查看文件中的说明与依赖,再按你的 Agent 支持的方式导入。
SKILL.mdMarkdown · 1 个文件
SKILL.md
---
name: chaos-engineer
description: Designs chaos experiments, creates failure injection frameworks, and facilitates game day exercises for distributed systems — producing runbooks, experiment manifests, rollback procedures, and post-mortem templates. Use when designing chaos experiments, implementing failure injection frameworks, or conducting game day exercises. Invoke for chaos experiments, resilience testing, blast radius control, game days, antifragile systems, fault injection, Chaos Monkey, Litmus Chaos.
license: MIT
metadata:
author: https://github.com/Jeffallan
company: https://synergetic.solutions
version: "1.1.0"
domain: devops
triggers: chaos engineering, resilience testing, failure injection, game day, blast radius, chaos experiment, fault injection, Chaos Monkey, Litmus Chaos, antifragile
role: specialist
scope: implementation
output-format: code
related-skills: sre-engineer, devops-engineer, kubernetes-specialist
---
# Chaos Engineer
## When to Use This Skill
- Designing and executing chaos experiments
- Implementing failure injection frameworks (Chaos Monkey, Litmus, etc.)
- Planning and conducting game day exercises
- Building blast radius controls and safety mechanisms
- Setting up continuous chaos testing in CI/CD
- Improving system resilience based on experiment findings
## Core Workflow
1. **System Analysis** - Map architecture, dependencies, critical paths, and failure modes
2. **Experiment Design** - Define hypothesis, steady state, blast radius, and safety controls
3. **Execute Chaos** - Run controlled experiments with monitoring and quick rollback
4. **Learn & Improve** - Document findings, implement fixes, enhance monitoring
5. **Automate** - Integrate chaos testing into CI/CD for continuous resilience
## Reference Guide
Load detailed guidance based on context:
| Topic | Reference | Load When |
|-------|-----------|-----------|
| Experiments | `references/experiment-design.md` | Designing hypothesis, blast radius, rollback |
| Infrastructure | `references/infrastructure-chaos.md` | Server, network, zone, region failures |
| Kubernetes | `references/kubernetes-chaos.md` | Pod, node, Litmus, chaos mesh experiments |
| Tools & Automation | `references/chaos-tools.md` | Chaos Monkey, Gremlin, Pumba, CI/CD integration |
| Game Days | `references/game-days.md` | Planning, executing, learning from game days |
## Safety Checklist
Non-obvious constraints that must be enforced on every experiment:
- **Steady state first** — define and verify baseline metrics before injecting any failure
- **Blast radius cap** — start with the smallest possible impact scope; expand only after validation
- **Automated rollback ≤ 30 seconds** — abort path must be scripted and tested before the experiment begins
- **Single variable** — change only one failure condition at a time until behaviour is well understood
- **No production without safety nets** — customer-facing environments require circuit breakers, feature flags, or canary isolation
- **Close the loop** — every experiment must produce a written learning summary and at least one tracked improvement
## Output Templates
When implementing chaos engineering, provide:
1. Experiment design document (hypothesis, metrics, blast radius)
2. Implementation code (failure injection scripts/manifests)
3. Monitoring setup and alert configuration
4. Rollback procedures and safety controls
5. Learning summary and improvement recommendations
## Concrete Example: Pod Failure Experiment (Litmus Chaos)
The following shows a complete experiment — from hypothesis to rollback — using Litmus Chaos on Kubernetes.
### Step 1 — Define steady state and apply the experiment
```bash
# Verify baseline: p99 latency < 200ms, error rate < 0.1%
kubectl get deploy my-service -n production
kubectl top pods -n production -l app=my-service
```
### Step 2 — Create and apply a Litmus ChaosEngine manifest
```yaml
# chaos-pod-delete.yaml
apiVersion: litmuschaos.io/v1alpha1
kind: ChaosEngine
metadata:
name: my-service-pod-delete
namespace: production
spec:
appinfo:
appns: production
applabel: "app=my-service"
appkind: deployment
# Limit blast radius: only 1 replica at a time
engineState: active
chaosServiceAccount: litmus-admin
experiments:
- name: pod-delete
spec:
components:
env:
- name: TOTAL_CHAOS_DURATION
value: "60" # seconds
- name: CHAOS_INTERVAL
value: "20" # delete one pod every 20s
- name: FORCE
value: "false"
- name: PODS_AFFECTED_PERC
value: "33" # max 33% of replicas affected
```
```bash
# Apply the experiment
kubectl apply -f chaos-pod-delete.yaml
# Watch experiment status
kubectl describe chaosengine my-service-pod-delete -n production
kubectl get chaosresult my-service-pod-delete-pod-delete -n production -w
```
### Step 3 — Monitor during the experiment
```bash
# Tail application logs for errors
kubectl logs -l app=my-service -n production --since=2m -f
# Check ChaosResult verdict when complete
kubectl get chaosresult my-service-pod-delete-pod-delete \
-n production -o jsonpath='{.status.experimentStatus.verdict}'
```
### Step 4 — Rollback / abort if steady state is violated
```bash
# Immediately stop the experiment
kubectl patch chaosengine my-service-pod-delete \
-n production --type merge -p '{"spec":{"engineState":"stop"}}'
# Confirm all pods are healthy
kubectl rollout status deployment/my-service -n production
```
## Concrete Example: Network Latency with toxiproxy
```bash
# Install toxiproxy CLI
brew install toxiproxy # macOS; use the binary release on Linux
# Start toxiproxy server (runs alongside your service)
toxiproxy-server &
# Create a proxy for your downstream dependency
toxiproxy-cli create -l 0.0.0.0:22222 -u downstream-db:5432 db-proxy
# Inject 300ms latency with 10% jitter — blast radius: this proxy only
toxiproxy-cli toxic add db-proxy -t latency -a latency=300 -a jitter=30
# Run your load test / observe metrics here ...
# Remove the toxic to restore normal behaviour
toxiproxy-cli toxic remove db-proxy -n latency_downstream
```
## Concrete Example: Chaos Monkey (Spinnaker / standalone)
```bash
# chaos-monkey-config.yml — restrict to a single ASG
deployment:
enabled: true
regionIndependence: false
chaos:
enabled: true
meanTimeBetweenKillsInWorkDays: 2
minTimeBetweenKillsInWorkDays: 1
grouping: APP # kill one instance per app, not per cluster
exceptions:
- account: production
region: us-east-1
detail: "*-canary" # never kill canary instances
# Apply and trigger a manual kill for testing
chaos-monkey --app my-service --account staging --dry-run false
```
Maintained by [@jeffallan](https://github.com/jeffallan), Principal Consultant at [Synergetic Solutions](https://synergetic.solutions)
[Documentation](https://jeffallan.github.io/claude-skills/skills/devops/chaos-engineer/)
---
## Source and license
Original skill: https://github.com/Jeffallan/claude-skills/blob/078a25cc5bfd8011459be7b91d004e900c1d02d3/skills/chaos-engineer/SKILL.md
Complete source directory: https://github.com/Jeffallan/claude-skills/tree/078a25cc5bfd8011459be7b91d004e900c1d02d3/skills/chaos-engineer
This catalog copy preserves the original instructions. Supporting scripts and resources are available in the original directory; this single-file export does not bundle them.
MIT License
Copyright (c) 2025
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.
使用示例
请使用「故障实验设计」技能处理下面的任务。 目标:在受控环境中规划故障假设、实验边界和观测指标。 输入材料:研究问题、已知资料、比较对象、证据范围和判断标准。 输出要求:区分事实、假设与推断;按证据组织结论,列出局限和下一步验证建议。 请先核对所需资料和工具;缺少关键信息时提出具体问题。涉及对外发布、删除或生产环境变更时,先提供可审查的计划,获得授权后再执行。

关注官方公众号
抢先获取热门 Skill 实操干货、产品更新与官方活动资讯。

加入官方交流群
对接创作者与同好,交流技能开发,共建 AI 技能生态。