Merge claudedo/81b054800dc4446598685fd04fefddb5

docs/explore-notes/worker-task-pipeline.md

@@ -1,6 +1,6 @@
 > **Explore-note — verify before trusting.** Distilled map of a subsystem, not authoritative.
-> Last verified against commit `896d4b5` (2026-07-23).
-> Drift check: `git log --oneline 896d4b5..HEAD -- src/ClaudeDo.Worker`
+> Last verified against commit `58741c2` (2026-08-06).
+> Drift check: `git log --oneline 58741c2..HEAD -- src/ClaudeDo.Worker`
 > Stable structure only (no line numbers). See docs/explore-notes/README.md.
 
 # Worker: Task Execution Pipeline
@@ -108,7 +108,11 @@ read-only "## Reference files" section.
 - `TaskStateService` — all task status transitions; guards preconditions; signals queue/hub.
 
 **Lifecycle/** (startup recovery)
-- `StaleTaskRecovery` — tasks stuck Running after a crash/restart → Failed.
+- `StaleTaskRecovery` — tasks stuck Running after a crash/restart → Failed. The underlying
+  `TaskStateService.RecoverStaleRunningAsync` bulk-flips Running→Failed, then re-runs the same
+  chain/parent-advance side effects as a normal `FailAsync` (per recovered id, best-effort) so a
+  crash mid-chain-child or mid-improvement-child doesn't leave a successor blocked forever or a
+  `WaitingForChildren` parent wedged.
 - `OrphanRecovery` — dequeues children whose parent is no longer planning (stays attached).
 - `AttachmentOrphanRecovery` — cleans orphaned attachment files.
 - `TaskResetService` — manual reset to Idle.

src/ClaudeDo.Worker/State/TaskStateService.cs

@@ -325,7 +325,9 @@ public sealed class TaskStateService : ITaskStateService
 
     // Unconditional status write — bypasses transition rules. Used by the UI's
     // "set status freely" affordance; intentionally no guards (caller may strand
-    // the runner if used while a task is executing).
+    // the runner if used while a task is executing). It also bypasses chain/parent
+    // advancement — forcing a chain child or a WaitingForChildren parent's child to a
+    // terminal status here does not unblock its successor or re-check the parent.
     public async Task<TransitionResult> ForceSetStatusAsync(string taskId, TaskStatus status, CancellationToken ct)
     {
         await using var ctx = await _dbFactory.CreateDbContextAsync(ct);
@@ -418,13 +420,32 @@ public sealed class TaskStateService : ITaskStateService
     {
         var resultText = "[stale] " + reason;
         var now = DateTime.UtcNow;
-        await using var ctx = await _dbFactory.CreateDbContextAsync(ct);
-        return await ctx.Tasks
-            .Where(t => t.Status == TaskStatus.Running)
-            .ExecuteUpdateAsync(s => s
-                .SetProperty(t => t.Status, TaskStatus.Failed)
-                .SetProperty(t => t.FinishedAt, now)
-                .SetProperty(t => t.Result, resultText), ct);
+        List<string> recoveredIds;
+        int affected;
+        await using (var ctx = await _dbFactory.CreateDbContextAsync(ct))
+        {
+            recoveredIds = await ctx.Tasks
+                .Where(t => t.Status == TaskStatus.Running)
+                .Select(t => t.Id)
+                .ToListAsync(ct);
+
+            affected = await ctx.Tasks
+                .Where(t => t.Status == TaskStatus.Running)
+                .ExecuteUpdateAsync(s => s
+                    .SetProperty(t => t.Status, TaskStatus.Failed)
+                    .SetProperty(t => t.FinishedAt, now)
+                    .SetProperty(t => t.Result, resultText), ct);
+        }
+
+        // A recovered task may have been a planning/improvement chain child or the last
+        // non-terminal child of a WaitingForChildren parent. The bulk flip above skips the
+        // usual terminal-transition side effects, so without this a crash mid-run would
+        // leave the chain successor blocked forever and the parent wedged in
+        // WaitingForChildren with nothing left to re-check it.
+        foreach (var taskId in recoveredIds)
+            await OnChildTerminalAsync(taskId, TaskStatus.Failed);
+
+        return affected;
     }
 
     // A subtask is "draft" only while its planning parent has an open (Active) session.

tests/ClaudeDo.Worker.Tests/State/TaskStateServiceTests.cs

@@ -456,6 +456,25 @@ public sealed class TaskStateServiceTests : IDisposable
         Assert.StartsWith("[stale] ", t.Result);
     }
 
+    [Fact]
+    public async Task RecoverStaleRunningAsync_ResolvesBlockedSuccessor_AndAdvancesParent()
+    {
+        var parent = await SeedTaskAsync(TaskStatus.WaitingForChildren, phase: PlanningPhase.Finalized);
+        var c0 = await SeedTaskAsync(TaskStatus.Running, parentId: parent, sortOrder: 0);
+        var c1 = await SeedTaskAsync(TaskStatus.Queued, parentId: parent, sortOrder: 1, blockedBy: c0);
+
+        var count = await _sut.RecoverStaleRunningAsync("worker restart", default);
+
+        Assert.Equal(1, count);
+        Assert.Equal(TaskStatus.Failed, await GetStatusAsync(c0));
+        // c1 was blocked on the crashed task; the chain coordinator resolves it (cancels it,
+        // since the predecessor didn't finish successfully) instead of leaving it Queued with
+        // a dangling BlockedByTaskId that the picker would skip forever.
+        Assert.Equal(TaskStatus.Cancelled, await GetStatusAsync(c1));
+        // Both children are now terminal, so the WaitingForChildren parent advances.
+        Assert.Equal(TaskStatus.WaitingForReview, await GetStatusAsync(parent));
+    }
+
     // ─── Child terminal → chain advance ───────────────────────────────────
 
     [Fact]