fix(workflow): the stage polls workflow/state twice every 2s, and a failed refresh claims there are no workflows #1564

Implementation: FlyBot (ours) — PR follows within 24h. Do not pick up without a pickup-ok comment on this issue.

Written-by: FlyBot Lead
Claimed-by: FlyBot Lead
Claim-Id: ec726c6fec184d469e5f162ca84fcd35
Branch: dev4/fix/p98-stage-poll
Base: 176d33d
At: 2026-10-28T06:55:07Z

Implementation: FlyBot (ours) — PR follows within 24h. Do not pick up without a pickup-ok comment.

Written-by: FlyBot Lead

Baseline: live main 176d33d0 (2026-10-28).

What happens

Two useEffects in plugins/workflow/client.js run the same 2s poll, and both fire:

• :1201-1207 — dependency [pack], setInterval at :1205
• :1210-1219 — dependency pack?.workflows?.some(w => w.activeCount > 0), setInterval at :1214

Both call refresh(), so while a run is active the whole workflow/state (every graph plus history) is fetched twice per tick. Measured with a copy of client.test.js plus one timing case (delivery() with activeCount = 1):

PROBE read timestamps (ms): [0,2001,2001,4003] — two workflow/state calls in the same millisecond at 2001 ms.

Second defect in the same place: refresh() (:1155-1160) is a bare await call('workflow/state') with no try/catch, and neither caller handles a rejection — void refresh().then(...) at :1174, plus the two polls. A failing host therefore ends as an unhandled rejection and the stage renders Stage's empty state (No workflows yet., :1340) — the opposite of the truth.

Measured on the same probe: PROBE alerts: [], PROBE unhandled: ["The workflow store is unreachable"], PROBE section text: "Workflows…No workflows yet.Hand a task to a Workflow". The node test runner fails the case with failureType: 'unhandledRejection', stack client.js:1158 ← :1176.

Expected

• One poll per 2s tick while a run is active — collapse the two effects, no double read.
• A failed read is reported as a failure (alert plus retry, the shape #1504 gives the Settings page), not as "No workflows yet.".
• refresh() settles into a value instead of rejecting: reuse settledCall (client.js:135, added by #1522) instead of introducing a second helper shape.

Scope

• plugins/workflow/client.js — collapse the two poll effects and normalise refresh(). No server change, no new endpoint.
• plugins/workflow/test/client.test.js — one case for "one read per tick" and one for "a failed read is not shown as no workflows"; the withUnhandled helper from #1522 is already there.
• changelog.d/.

Out of scope: #1504's Settings page (WorkflowSettings), the Runs row metadata (P-97), and any core/ or host change.

Risk: T2 (plugins/**).