eve version
eve@0.13.3
Node.js version
Not recorded.
Where does the bug occur?
Self-hosted production path: eve build && eve start with a configured Workflow world.
Describe the bug
The self-host guide presents eve build followed by eve start as the normal non-Vercel deployment path, and in the same section says a self-deployed agent can select another world through experimental.workflow.world (docs).
In 0.13.3, the dev build registers the workflow entrypoint directly on the active world queue. The test checks for registerHandler("__eve_wkf_workflow_", POST) in dev output (source). The production-build test explicitly asserts that the generated handler contains no registerHandler call and no getWorld import (source).
Observed behavior: the configured world works under eve dev, but eve build && eve start does not register its queue handler. Queue-backed custom worlds then cannot dispatch the workflow entrypoint. This can present as Unhandled queue and is separate from the queue-namespace configuration problem.
Steps to reproduce
- Configure a queue-backed custom world in
agent.ts.
- Confirm that a turn works under
eve dev.
- Run
eve build and then eve start with the same configuration.
- Send the same turn and observe that the world cannot dispatch the workflow queue.
Expected behavior
The documented production self-host path registers the configured world queue handler before it accepts work, just as the development path does.
Actual behavior
The code intentionally emits direct queue registration only for dev output, leaving the documented production path unable to run a custom queue-backed world.
Suggested fix
- Register the direct workflow handler for a configured self-hosted world in production output as well.
- Add an integration test that runs
eve build && eve start with a queue-backed world and completes one turn.
- Until fixed, change the docs so they do not present this path as a working custom-world deployment.
eve version
eve@0.13.3Node.js version
Not recorded.
Where does the bug occur?
Self-hosted production path:
eve build && eve startwith a configured Workflow world.Describe the bug
The self-host guide presents
eve buildfollowed byeve startas the normal non-Vercel deployment path, and in the same section says a self-deployed agent can select another world throughexperimental.workflow.world(docs).In 0.13.3, the dev build registers the workflow entrypoint directly on the active world queue. The test checks for
registerHandler("__eve_wkf_workflow_", POST)in dev output (source). The production-build test explicitly asserts that the generated handler contains noregisterHandlercall and nogetWorldimport (source).Observed behavior: the configured world works under
eve dev, buteve build && eve startdoes not register its queue handler. Queue-backed custom worlds then cannot dispatch the workflow entrypoint. This can present asUnhandled queueand is separate from the queue-namespace configuration problem.Steps to reproduce
agent.ts.eve dev.eve buildand theneve startwith the same configuration.Expected behavior
The documented production self-host path registers the configured world queue handler before it accepts work, just as the development path does.
Actual behavior
The code intentionally emits direct queue registration only for dev output, leaving the documented production path unable to run a custom queue-backed world.
Suggested fix
eve build && eve startwith a queue-backed world and completes one turn.