Schedule workflows and inspect each occurrence's run #29

Parent

#3

What to build

Let existing scheduling controls select a workflow target and task inputs, then link each occurrence to its workflow run and distinguish dispatch from the eventual outcome. Reuse the manual start/run experience and the host's existing scheduler.

Acceptance criteria

• Own a finite generic schedule-target contribution that supplies workflow selection and bounded free-form task input to existing controls. Core scheduling imports no workflow domain code; normal prompt schedules remain usable.
• Show target environment/project, runnable saved workflow, inputs, and actual revision-at-dispatch semantics. Do not claim future occurrences are pinned to the currently displayed revision or invent an unavailable required-input contract.
• Each occurrence invokes the existing start service with durable occurrence identity and a saved snapshot. Redelivery/lost acknowledgement produces one run; distinct occurrences produce distinct runs. Existing pending occurrence identities and permission ceilings remain intact.
• Own bounded persisted occurrence-to-run history/read access. Link the exact run and display dispatch pending/succeeded/failed separately from workflow running/completed/failed/awaiting review. Run source/revision evidence remains truthful after a workflow edit, reload, or restart.
• Enable/pause/edit/delete/run-now follow existing schedule semantics and destination authorization. Plugin removal retains state with an actionable unavailable target. The client creates no clock/timing owner and never substitutes another environment.
• Controls/history are keyboard/touch-accessible with semantic labels/textual state. Missing capabilities, permission failures, disconnection, and stale cached actions behave like the other workflow surfaces; update concise user guidance for how to schedule a run.
• Focused real-backend tests cover create/edit/pause/run-now, occurrence redelivery, changed saved revisions between occurrences, restart, authenticated remote target isolation, and prompt-schedule compatibility. Verify scheduled runs use the same run/thread/evidence/gate paths as manual starts; when parallel authoring is available, include its graph in the integrated walkthrough without making scheduling depend on that authoring ticket.

Blocked by

• Run workflows with native threads, evidence, and recovery #26 — Run workflows with native threads, evidence, and recovery.

Ticket specification reviewed and agreed by Codex and Claude Opus 5.5.