fix(paired): cap consecutive reviewer silent-failure retries

src/paired-execution-context-reviewer.ts

@@ -2,7 +2,10 @@ import { ARBITER_DEADLOCK_THRESHOLD } from './config.js';
 import { getPairedWorkspace } from './db.js';
 import { logger } from './logger.js';
 import { requestArbiterOrEscalate } from './paired-arbiter-request.js';
-import { transitionPairedTaskStatus } from './paired-task-status.js';
+import {
+  applyPairedTaskPatch,
+  transitionPairedTaskStatus,
+} from './paired-task-status.js';
 import {
   resolveReviewerCompletionSignal,
   resolveReviewerFailureSignal,
@@ -11,6 +14,17 @@ import { resolveCanonicalSourceRef } from './paired-source-ref.js';
 import { parseVisibleVerdict } from './paired-verdict.js';
 import type { PairedTask } from './types.js';
 
+/**
+ * Maximum consecutive reviewer executions that may produce no verdict
+ * before we stop the ping-pong loop. Mirrors the owner-side failure cap and
+ * closes the "reviewer silently fails forever" hole: codex sometimes emits a
+ * final result with no visible text, the runtime marks it failed, the
+ * next-turn scheduler re-queues another reviewer-turn, and the original
+ * round_trip_count never increments — so the existing round-trip / arbiter
+ * caps never fire.
+ */
+const REVIEWER_FAILURE_ESCALATION_THRESHOLD = 2;
+
 export function handleFailedReviewerExecution(args: {
   task: PairedTask;
   taskId: string;
@@ -75,6 +89,35 @@ export function handleFailedReviewerExecution(args: {
     }
   }
 
+  // Reviewer produced no usable verdict and no recognised infrastructure
+  // failure. Track consecutive flakes — after the threshold, stop the
+  // ping-pong loop instead of preserving review_ready forever.
+  const nextFailureCount = (task.reviewer_failure_count ?? 0) + 1;
+
+  if (nextFailureCount >= REVIEWER_FAILURE_ESCALATION_THRESHOLD) {
+    requestArbiterOrEscalate({
+      taskId,
+      currentStatus: task.status,
+      expectedUpdatedAt: task.updated_at,
+      now,
+      arbiterLogMessage:
+        'Reviewer failed repeatedly without a visible verdict — requesting arbiter',
+      escalateLogMessage:
+        'Reviewer failed repeatedly without a visible verdict — escalating to user',
+      logContext: {
+        taskId,
+        role: 'reviewer',
+        previousStatus: task.status,
+        reviewerFailureCount: nextFailureCount,
+        summary: summary?.slice(0, 160),
+      },
+      patch: {
+        reviewer_failure_count: nextFailureCount,
+      },
+    });
+    return;
+  }
+
   const fallbackStatus =
     task.status === 'in_review' || task.status === 'review_ready'
       ? 'review_ready'
@@ -86,17 +129,30 @@ export function handleFailedReviewerExecution(args: {
       nextStatus: fallbackStatus,
       expectedUpdatedAt: task.updated_at,
       updatedAt: now,
-    });
-    logger.warn(
-      {
-        taskId,
-        role: 'reviewer',
-        previousStatus: task.status,
-        nextStatus: fallbackStatus,
+      patch: {
+        reviewer_failure_count: nextFailureCount,
       },
-      'Preserved reviewer task in review-ready state after failed execution',
-    );
+    });
+  } else {
+    applyPairedTaskPatch({
+      taskId,
+      expectedUpdatedAt: task.updated_at,
+      updatedAt: now,
+      patch: {
+        reviewer_failure_count: nextFailureCount,
+      },
+    });
   }
+  logger.warn(
+    {
+      taskId,
+      role: 'reviewer',
+      previousStatus: task.status,
+      reviewerFailureCount: nextFailureCount,
+      summary: summary?.slice(0, 160),
+    },
+    'Preserved reviewer task in review-ready state after failed execution',
+  );
 }
 
 export function handleReviewerCompletion(args: {
@@ -128,6 +184,7 @@ export function handleReviewerCompletion(args: {
         patch: {
           source_ref: approvedSourceRef,
           owner_failure_count: 0,
+          reviewer_failure_count: 0,
           owner_step_done_streak: 0,
           finalize_step_done_count: 0,
           empty_step_done_streak: 0,
@@ -164,6 +221,7 @@ export function handleReviewerCompletion(args: {
         arbiterLogMessage,
         escalateLogMessage,
         logContext: { taskId, verdict, summary: summary?.slice(0, 100) },
+        patch: { reviewer_failure_count: 0 },
       });
       return;
     }
@@ -175,6 +233,7 @@ export function handleReviewerCompletion(args: {
         nextStatus: 'active',
         expectedUpdatedAt: task.updated_at,
         updatedAt: now,
+        patch: { reviewer_failure_count: 0 },
       });
       logger.info(
         { taskId, verdict },