a3fe15b: Use the canonical AlarmInvocationInfo type from @cloudflare/workers-types for the alarm() parameter instead of an inline type. This is a no-op for users (the shape is identical), but keeps the override aligned with the Durable Object base class signature.

ca72a22: Refreshed the examples/ projects:

• All examples now use the latest TypeScript, Vitest, and Wrangler, and target a current Workers compatibility_date.
• Worker types are generated by wrangler types, matching current Cloudflare guidance.
• The examples/timeout/ snippet is now a fully runnable example.
• Integration tests now run on arm64 hosts (e.g. Apple Silicon) and reliably clean up after themselves.

No changes to the published library API.

fc7e7f4: Add return types to exported functions and public methods to satisfy ESLint and improve type checking.

eabe7ac: Clarify the license for this library matches that of @cloudflare/workers-sdk, which is dual licensed under either MIT OR Apache-2.0.

df8699a: Tighten the return type of Container#onError from any to unknown. Subclasses that override onError can still return any value. This should be a no-op for most users.

45274ea: Preserve original errors as cause when wrapping abort/timeout errors during container startup, making it easier to debug the underlying failure.

19c1709: Reset container state after failed startup or terminal monitor errors, avoid stale monitor callbacks updating newer instances, and apply configured constructor startup options.

8442f40: Fix a race in Container where concurrent container.fetch calls to a cold container could throw "start() cannot be called on a container that is already running.".

cf01432: Ensure pending stop events are processed when the persisted container lifecycle state is still running but the underlying container has already exited.

Migrate the root unit test suite from Jest to Vitest and add a test:unit script for running src/tests directly.

cf41295: Fix subclass sleepAfter overrides being ignored during the initial activity timeout setup. Previously, the base Container constructor called renewActivityTimeout() inside blockConcurrencyWhile() before subclass class-field initializers ran, so the first sleepAfterMs was always computed from the base default ('10m') regardless of whether the subclass declared sleepAfter = '2h'. A container could then be killed by the activity timeout before the subclass's longer window took effect on the next renewActivityTimeout call. Declarations like:

class BigContainer extends Container {
  sleepAfter = '2h';
}

are now honored from the very first alarm check.

07fedbb: Preserve Cloudchamber startup rate-limit errors in the Containers helper and return HTTP 429 from containerFetch() when startup is rate limited.

7901d0b: Attach custom labels to your containers to help with observability and attribution. Set a labels property on your Container subclass, or pass labels in the startOptions argument to start() / startAndWaitForPorts(), and tag containers by tenant, environment, feature flag, canary cohort, or any other dimension you want to track. In local development, labels are visible on the underlying Docker container via docker inspect.

class MyContainer extends Container {
  labels = { tenant: 'acme', env: 'prod' };
}