fix(review): use ParseVerdict for pass detection in single-result formatting

strings.Contains(marker) was too loose — showed Review Passed even
when output also contained severity-labeled findings. Now delegates
to storage.ParseVerdict which checks severity labels first.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: cpcloud

internal/review/synthesize.go

@@ -5,12 +5,12 @@ import (
 	"errors"
 	"fmt"
 	"log"
-	"strings"
 	"time"
 
 	"github.com/roborev-dev/roborev/internal/agent"
 	"github.com/roborev-dev/roborev/internal/config"
 	"github.com/roborev-dev/roborev/internal/git"
+	"github.com/roborev-dev/roborev/internal/storage"
 )
 
 // ErrAllFailed is returned by Synthesize when every review job
@@ -96,7 +96,7 @@ func formatSingleResult(
 ) string {
 	var header string
 	if r.Output == "" || r.Output == "No issues found." ||
-		strings.Contains(r.Output, config.SeverityThresholdMarker) {
+		storage.ParseVerdict(r.Output) == "P" {
 		header = fmt.Sprintf(
 			"## roborev: Review Passed (`%s`)\n\n",
 			git.ShortSHA(headSHA))

internal/review/synthesize_test.go

@@ -121,6 +121,21 @@ func TestSynthesize_Formatting(t *testing.T) {
 				"Review Passed",
 			},
 		},
+		{
+			name: "ThresholdMetWithFindings",
+			results: []ReviewResult{
+				{
+					Agent:      "codex",
+					ReviewType: "review",
+					Status:     "done",
+					Output:     "SEVERITY_THRESHOLD_MET\n\n- High: critical bug found",
+				},
+			},
+			expectedErr: nil,
+			expectedTexts: []string{
+				"Review Complete",
+			},
+		},
 		{
 			name: "AllQuota",
 			results: []ReviewResult{