fix: prevent stale duplicate planning tasks from blocking phase completion

Three fixes for phases getting stuck when a detail task crashes and is retried:

1. detailPhase mutation (architect.ts): clean up orphaned pending/in_progress
   detail tasks before creating new ones, preventing duplicates at the source
2. orchestrator recovery: detect and complete stale duplicate planning tasks
   (same category+phase, one completed, one pending)
3. ensureBranch: catch "already exists" TOCTOU race instead of blocking phase

apps/server/execution/orchestrator.ts

@@ -20,7 +20,7 @@ import type { ProjectRepository } from '../db/repositories/project-repository.js
 import type { AgentRepository } from '../db/repositories/agent-repository.js';
 import type { DispatchManager, PhaseDispatchManager } from '../dispatch/types.js';
 import type { ConflictResolutionService } from '../coordination/conflict-resolution-service.js';
-import { phaseBranchName, taskBranchName } from '../git/branch-naming.js';
+import { phaseBranchName, taskBranchName, isPlanningCategory } from '../git/branch-naming.js';
 import { ensureProjectClone } from '../git/project-clones.js';
 import { createModuleLogger } from '../logger/index.js';
 import { phaseMetaCache, fileDiffCache } from '../review/diff-cache.js';
@@ -637,6 +637,23 @@ export class ExecutionOrchestrator {
             }
           }
 
+          // Clean up stale duplicate planning tasks (e.g. a crashed detail task
+          // that was reset to pending, then a new detail task was created and completed).
+          const tasksAfterRecovery = await this.taskRepository.findByPhaseId(phase.id);
+          const completedPlanningNames = new Set<string>();
+          for (const t of tasksAfterRecovery) {
+            if (isPlanningCategory(t.category) && t.status === 'completed') {
+              completedPlanningNames.add(`${t.category}:${t.phaseId}`);
+            }
+          }
+          for (const t of tasksAfterRecovery) {
+            if (isPlanningCategory(t.category) && t.status === 'pending' && completedPlanningNames.has(`${t.category}:${t.phaseId}`)) {
+              await this.taskRepository.update(t.id, { status: 'completed', summary: 'Superseded by retry' });
+              tasksRecovered++;
+              log.info({ taskId: t.id, category: t.category }, 'recovered stale duplicate planning task');
+            }
+          }
+
           // Re-read tasks after recovery updates and check if phase is now fully done
           const updatedTasks = await this.taskRepository.findByPhaseId(phase.id);
           const allDone = updatedTasks.every((t) => t.status === 'completed');

apps/server/git/simple-git-branch-manager.ts

@@ -46,8 +46,17 @@ export class SimpleGitBranchManager implements BranchManager {
       return;
     }
 
+    try {
       await git.branch([branch, baseBranch]);
       log.info({ repoPath, branch, baseBranch }, 'branch created');
+    } catch (err) {
+      // Handle TOCTOU race: branch may have been created between the exists check and now
+      if (err instanceof Error && err.message.includes('already exists')) {
+        log.debug({ repoPath, branch }, 'branch created by concurrent process');
+        return;
+      }
+      throw err;
+    }
   }
 
   async mergeBranch(repoPath: string, sourceBranch: string, targetBranch: string): Promise<MergeResult> {

apps/server/trpc/routers/architect.ts

@@ -337,6 +337,14 @@ export function architectProcedures(publicProcedure: ProcedureBuilder) {
           });
         }
 
+        // Clean up orphaned pending/in_progress detail tasks from previous failed attempts
+        const phaseTasks = await taskRepo.findByPhaseId(input.phaseId);
+        for (const t of phaseTasks) {
+          if (t.category === 'detail' && (t.status === 'pending' || t.status === 'in_progress')) {
+            await taskRepo.update(t.id, { status: 'completed', summary: 'Superseded by retry' });
+          }
+        }
+
         const detailTaskName = input.taskName ?? `Detail: ${phase.name}`;
         const task = await taskRepo.create({
           phaseId: phase.id,

docs/dispatch-events.md

@@ -149,8 +149,11 @@ When an agent crashes (`agent:crashed` event), the orchestrator automatically re
 On server restart, `recoverDispatchQueues()` also recovers:
 - Stuck `in_progress` tasks whose agents are dead (status is not `running` or `waiting_for_input`) — reset to `pending` and re-queued
 - Erroneously `blocked` tasks whose agents completed successfully (status is `idle` or `stopped`) — marked `completed` so the phase can progress. This handles the legacy case where conflict resolution incorrectly blocked already-completed tasks.
+- Stale duplicate planning tasks — if a phase has both a completed and a pending task of the same planning category (e.g. two `detail` tasks from a crash-and-retry), the pending one is marked `completed` with summary "Superseded by retry"
 - Fully-completed `in_progress` phases — after task recovery, if all tasks in an `in_progress` phase are completed, triggers `handlePhaseAllTasksDone` to complete/review the phase
 
+The `detailPhase` mutation in `architect.ts` also cleans up orphaned pending/in_progress detail tasks before creating new ones, preventing duplicates at the source.
+
 Manual retry via `retryBlockedTask()` resets `retryCount` to 0, giving the task a fresh set of automatic retries.
 
 ### Coalesced Scheduling