agentleFS
kryptophonikkryptophonik/SRE-Skills/skills/sre-chaos-engineering/SKILL.md

sre-chaos-engineering

Skill4 starsChanged 23 days agoMITView on GitHub

Use to design or run a chaos / fault-injection experiment or game day, with hard no-go rules, approval and abort criteria. Manual invocation only. Not for live incident response.

The file

We split this skill into the rules it gives an agent

11 rules. About 0 are common, 11 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈11

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Wait for explicit human approval before any prod-changing command, MCP write tool, or external post.

    line 13
  2. Distinctive:

    No tested rollback / abort mechanism

    line 24
  3. Distinctive:

    No named owner present for the full run

    line 25
  4. Distinctive:

    No timebox (start/end window and max duration)

    line 26
  5. Distinctive:

    No explicit human approval for this experiment, environment, and blast radius

    line 27
  6. Distinctive:

    Prefer production-like environments.

    line 35
  7. Distinctive:

    Check the No-go rules section.

    line 43
  8. Distinctive:

    Run Experiment within timebox.

    line 49
  9. Distinctive:

    Always have a 'Stop' button (Abort condition) and test it first.

    line 85
  10. Distinctive:

    Start with a small blast radius (1% of traffic or 1 instance).

    line 86
  11. Distinctive:

    Ensure observability is working *before* starting the experiment.

    line 87

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Chaos Engineering Skill

Safety gates (REQUIRED)

  1. Read-only diagnosis first.
  2. Propose each change with risk (low/med/high) and reversibility.
  3. Wait for explicit human approval before any prod-changing command, MCP write tool, or external post.
  4. Act only as approved, verify, and stop and escalate if anything gets worse.

Full rules: sre-safety-gates. Read-only MCP setup: sre-safety-gates/references/mcp.md.

No-go rules

if ANY is true, do not run — report why and stop

  • Error budget for the target service is exhausted or burning above normal
  • An active incident or ongoing change freeze exists
  • No tested rollback / abort mechanism
  • No named owner present for the full run
  • No timebox (start/end window and max duration)
  • No explicit human approval for this experiment, environment, and blast radius
  • Observability for steady-state metrics is not working
  • On-call and stakeholders not notified

Chaos principles

  • Build a hypothesis around steady-state behavior.
  • Vary real-world events (server crashes, network latency).
  • Prefer production-like environments; run in production only with explicit approval and smallest blast radius.
  • Automate experiments to run continuously once proven safe.
  • Minimize blast radius.

Experiment workflow

Steps

  1. Check the No-go rules section; stop if any apply.
  2. Define 'Steady State': Metrics that indicate the system is healthy.
  3. Formulate Hypothesis: 'If we do X, the system will still be Y'.
  4. Design Experiment: Choose tool and parameters (see tools catalog, Chaos).
  5. Define Abort Criteria and rollback (see Abort criteria).
  6. Propose plan: target, blast radius, owner, timebox, risk, rollback — get explicit approval.
  7. Run Experiment within timebox; monitor abort criteria continuously.
  8. Analyze: Compare results to hypothesis.
  9. Improve Resilience: File fixes for any failures found.

Abort criteria

any trigger -> halt injection immediately, roll back, notify

  • SLO burn rate exceeds agreed threshold (e.g., >2x normal)
  • Error rate or p99 latency exceeds pre-agreed limit
  • User-facing impact outside the planned blast radius
  • Unexpected alerts in unrelated services
  • Timebox expires
  • Owner or on-call calls stop

Common experiments

Network

  • Latency: Inject 200ms delay into service-to-service calls.
  • Packet Loss: Drop 5% of incoming packets.
  • Partition: Block traffic between two availability zones. (high risk)

Infrastructure

  • Pod Kill: Terminate one pod in a deployment.
  • Node Drain: Simulate losing an entire cluster node. (high risk)
  • Resource Stress: Force 90% CPU or Memory usage on a pod.

Application

  • API Errors: Return 500s for a percentage of requests.
  • Dependency Failure: Simulate database or cache being unavailable. (high risk)

Safety checks

  • Always have a 'Stop' button (Abort condition) and test it first.
  • Start with a small blast radius (1% of traffic or 1 instance).
  • Ensure observability is working before starting the experiment.
  • Notify stakeholders and on-call engineers.

Output

  • Experiment record: hypothesis, approval (who/when), blast radius, timebox, results, abort events, follow-up tickets

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.