Jobs

This page is for whoever runs the Cadence instance itself. Cadence does some of its work in background jobs that run without you. The Jobs screen lists those jobs by state, so you can see what the instance is working through and, above all, what has failed.

This screen is reached only in the browser. There is no Model Context Protocol tool, no HTTP API route, and no command-line command for it — a program driving Cadence does not see the job backend, and this screen is where you look at it.

Who can open it

Open it at Settings → Jobs (/settings/jobs). It is gated exactly like the rest of Settings: you must be signed in and hold instance admin — the instance-wide setting for whoever runs the Cadence instance, which is separate from any role you hold in a space (see Spaces, teams, and who can do what and the instance-admin note in Authentication). Holding admin in a space does not grant it.

• A signed-in operator who is not an instance admin is refused with 403 Forbidden.
• A signed-out visitor is refused and sent to sign in.

What it shows

The screen groups the instance's background jobs by state:

• queued — waiting to run.
• in flight — running now.
• scheduled — set to run at a later time.
• blocked — waiting on another job before they can run.
• failed — jobs that ran and errored.

The reason this screen exists is the last of those. Before it, a failed background job left no trace anywhere you could look, so "nothing is wrong" and "I cannot see what is wrong" looked the same. The screen makes a failure visible; it does not retry a job for you, so a failed job stays failed until you act on it.

Each job's row lists the arguments it was called with. One argument is held back: the agent instruction — the prompt handed to a worker — is filtered out of the listing, so it does not appear in every row of a queue. You can still read that prompt on the task it belongs to.

A back link returns you to Settings.