Advanced

Presets, global mode, scenarios, and fetch interception

Presets

cruel.enable(cruel.presets.development)
cruel.enable(cruel.presets.staging)
cruel.enable(cruel.presets.production)
cruel.enable(cruel.presets.harsh)
cruel.enable(cruel.presets.nightmare)
cruel.enable(cruel.presets.apocalypse)

Profiles

cruel.profile("testing", { fail: 0.2, delay: 100 })
cruel.useProfile("testing")

Global Mode

cruel.enable({ fail: 0.1, delay: [100, 500] })
cruel.disable()
cruel.toggle()
cruel.isEnabled()

Scoped Chaos

await cruel.scope(async () => {
  await api("...")
}, { fail: 0.2 })

Fluent API

cruel.wrap(fn).fail(0.1)
cruel.wrap(fn).slow(500)
cruel.wrap(fn).timeout(0.05)
cruel.wrap(fn).flaky()
cruel.wrap(fn).nightmare()

Factory

import { createCruel } from "cruel"

const myCruel = createCruel({ delay: 100 })

Scenarios

cruel.scenario("outage", {
  chaos: { fail: 1 },
  duration: 5000,
})

await cruel.play("outage")
cruel.stop()

await cruel.play("networkPartition")
await cruel.play("highLatency")
await cruel.play("degraded")
await cruel.play("recovery")

Fetch Interception

cruel.patchFetch()

cruel.intercept("api.openai.com", {
  rateLimit: { rate: 0.1, retryAfter: 60 },
  delay: [100, 500],
})

cruel.intercept(/api\.anthropic\.com/, {
  fail: 0.1,
  status: [529],
})

cruel.unpatchFetch()

Production Rollout Checklist

For production-like chaos drills, keep blast radius controlled:

Start with low rates (0.01 - 0.05) and increase gradually.
Run chaos in a scoped block or dedicated test path first.
Attach onChaos and record event types, model IDs, and latency.
Keep retry, fallback, and circuit-breaker logic enabled during tests.
Run deterministic passes with cruel.seed(...) before random passes.
Stop chaos immediately if error budget or latency thresholds are exceeded.

Advanced

On this page