fix(agents): increment failed instead of skipped for stale orphaned subagent runs (fixes #90766)

zenglingbiao · zenglingbiao · commit 7a6cf2583986 · 2026-06-21T10:11:52.000+08:00
When a stale unended subagent run is detected, increment 'failed' instead of
'skipped' so scheduleOrphanRecovery can retry finalization. Previously the stale
branch only incremented 'skipped' and the fire-and-forget Promise.allSettled
finalization would silently drop errors.

ClawSweeper P2: 'Make stale finalization retry-visible'
diff --git a/src/agents/subagent-orphan-recovery.ts b/src/agents/subagent-orphan-recovery.ts
@@ -256,8 +256,11 @@ export async function recoverOrphanedSubagentSessions(params: {
 
         // Age out stale runs so long-downtime orphans are retired instead
         // of being resurrected past the staleness window.
+        // Increment `failed` (not `skipped`) so scheduleOrphanRecovery retries
+        // when finalization yields zero-update or throws; otherwise a stale
+        // active run can be left without retry-visible accounting.
         if (isStaleUnendedSubagentRun(runRecord, now)) {
-          result.skipped++;
+          result.failed++;
           result.failedRuns.push({
             runId,
             childSessionKey,
