diff --git a/.opencode/agents/ca-continuous-pr-reviewer.md b/.opencode/agents/ca-continuous-pr-reviewer.md
index 1c1589de3..190368c5a 100644
--- a/.opencode/agents/ca-continuous-pr-reviewer.md
+++ b/.opencode/agents/ca-continuous-pr-reviewer.md
@@ -321,8 +321,30 @@ LOOP FOREVER:
     
     stale_count = 0
     
+    # ── Step 2b: Prioritize work items ───────────────────────────
+    # Sort work items so the most impactful PRs are reviewed first.
+    # This ensures earlier milestones (which are overdue) get reviewed
+    # before later milestones, and merge retries (which take seconds)
+    # are processed before full reviews (which take minutes).
+    work_items.sort(key=lambda item: (
+        # Priority 1: Merge retries first (already approved, just need merge)
+        0 if item.type == "merge_retry" else 1,
+        # Priority 2: Earlier milestones first (v3.2.0 before v3.7.0)
+        #   Extract version tuple from milestone title, e.g. "v3.2.0" → (3, 2, 0)
+        #   PRs without a milestone sort last (999, 999, 999)
+        milestone_sort_key(item.pr.milestone if hasattr(item, 'pr') else None),
+        # Priority 3: PRs with passing CI before PRs with pending CI
+        0 if item.type == "ci_fix" else 1,
+        # Priority 4: Smaller PRs first (faster to review)
+        (item.pr.additions + item.pr.deletions) if hasattr(item, 'pr') else 0,
+    ))
+    # Helper: milestone_sort_key(milestone) →
+    #   if milestone is None: return (999, 999, 999)
+    #   parse version from milestone.title, e.g. "v3.2.0" → (3, 2, 0)
+    #   return parsed tuple, or (999, 999, 999) if unparseable
+    
     # ── Step 3: Dispatch parallel reviewers via prompt_async ─────
-    # Take up to N work items
+    # Take up to N work items (now sorted by priority)
     batch = work_items[:N]
     
     # Claim all PRs in the batch using two-phase locking protocol