From 9da38ef3e58970e9b60e58acb694a266285f00e6 Mon Sep 17 00:00:00 2001
From: abccodes <abayercalvert@gmail.com>
Date: Tue, 23 Jun 2026 00:02:52 -0700
Subject: [PATCH] [aidan] fix/run-status: sync ongoing runs and heal
 stuck/interrupted runs

---
 backend/apps/workflows/executor.py          | 13 +++++++++++++
 backend/apps/workflows/scheduler.py         |  8 ++++++++
 frontend/src/shared/state/workflowsSlice.ts | 11 +++++++++++
 3 files changed, 32 insertions(+)

diff --git a/backend/apps/workflows/executor.py b/backend/apps/workflows/executor.py
index 66bc194a..f5f51011 100644
--- a/backend/apps/workflows/executor.py
+++ b/backend/apps/workflows/executor.py
@@ -215,6 +215,19 @@ async def execute(
         "last_run_id": run.id,
     })
 
+    # Announce the run as running the instant it claims execution, not at the
+    # first step. Without this a run that fails fast (e.g. no runnable steps) or
+    # hasn't streamed yet never hits the Home "Ongoing runs" list, so a batch of
+    # missed re-runs only ever surfaced whichever one reached its first step.
+    try:
+        from backend.apps.agents.core.ws_manager import ws_manager as _wsm_start
+        await _wsm_start.broadcast_global("workflow:run", {
+            "workflow_id": wf.id,
+            "run": run.model_dump(mode="json"),
+        })
+    except Exception:
+        pass
+
     session = None
     try:
         steps = [s for s in wf.steps if s.enabled and s.text and s.text.strip()]
diff --git a/backend/apps/workflows/scheduler.py b/backend/apps/workflows/scheduler.py
index 6fd0386e..650b40f1 100644
--- a/backend/apps/workflows/scheduler.py
+++ b/backend/apps/workflows/scheduler.py
@@ -362,6 +362,14 @@ def _mark_stuck_runs_failed() -> None:
                     error="OpenSwarm closed before this run finished.",
                     finished_at=now,
                 )
+                # The run row is fixed, but the workflow still summarizes this
+                # dead run as 'running' (that's what the detail header reads), so
+                # heal the summary too when this was the latest run.
+                if wf.last_run_id == r.id and wf.last_run_status == "running":
+                    executor._persist_run_fields(wf, {
+                        "last_run_status": "failure",
+                        "last_run_at": now,
+                    })
 
 
 def record_skipped(wf: Workflow, scheduled_for: datetime, error: str) -> WorkflowRun:
diff --git a/frontend/src/shared/state/workflowsSlice.ts b/frontend/src/shared/state/workflowsSlice.ts
index 82c954df..f24e4a4b 100644
--- a/frontend/src/shared/state/workflowsSlice.ts
+++ b/frontend/src/shared/state/workflowsSlice.ts
@@ -230,6 +230,17 @@ function mergeRunIntoState(state: State, r: WorkflowRun) {
     wf.last_run_status = r.status === 'skipped' ? wf.last_run_status : (r.status as Workflow['last_run_status']);
     wf.last_run_id = r.id;
   }
+  // Keep the live "Ongoing runs" list in sync off the WS stream: a run that's no
+  // longer running drops out of `active`, a freshly-running one joins. Without
+  // this, `active` only refreshed on the one-shot mount fetch, so finished runs
+  // lingered as "Working…" while the monitor already showed them done.
+  const activeIdx = state.active.findIndex((a) => a.run_id === r.id);
+  if (r.status === 'running') {
+    const entry: ActiveRun = { workflow_id: r.workflow_id, run_id: r.id, title: wf?.title || '', started_at: r.started_at };
+    if (activeIdx >= 0) state.active[activeIdx] = entry; else state.active.unshift(entry);
+  } else if (activeIdx >= 0) {
+    state.active.splice(activeIdx, 1);
+  }
   // Auto-flip the card view on run state transitions so the user sees
   // Running while it streams, Completed on success, Failed on failure.
   // Only nudge from views that the user hasn't actively navigated away