diff --git a/pkg/git/rawobj.go b/pkg/git/rawobj.go
index fa34c1636..4e583ed75 100644
--- a/pkg/git/rawobj.go
+++ b/pkg/git/rawobj.go
@@ -113,7 +113,7 @@ type TagSig struct {
 // or gpgsig-sha256 headers, because either is ambiguous about which
 // signature to extract.
 func SplitCommit(r io.Reader) (*CommitSig, error) {
-	scanner := bufio.NewScanner(r)
+	scanner := newRawObjectScanner(r)
 
 	var (
 		payloadBuf bytes.Buffer
@@ -245,7 +245,7 @@ func JoinCommit(c *CommitSig) ([]byte, error) {
 // signatures (gpgsig, gpgsig-sha256) are stripped from the header section
 // the same way SplitCommit does.
 func SplitTag(r io.Reader) (*TagSig, error) {
-	scanner := bufio.NewScanner(r)
+	scanner := newRawObjectScanner(r)
 
 	var (
 		payloadBuf   bytes.Buffer
@@ -392,6 +392,26 @@ func sigOrNil(b *bytes.Buffer) []byte {
 	return b.Bytes()
 }
 
+// newRawObjectScanner splits on '\n' like bufio.ScanLines but keeps a trailing
+// '\r' in the token. git treats the '\r' as object content, so dropping it
+// would make what we split diverge from the bytes git hashes.
+func newRawObjectScanner(r io.Reader) *bufio.Scanner {
+	s := bufio.NewScanner(r)
+	s.Split(func(data []byte, atEOF bool) (int, []byte, error) {
+		if atEOF && len(data) == 0 {
+			return 0, nil, nil
+		}
+		if i := bytes.IndexByte(data, '\n'); i >= 0 {
+			return i + 1, data[:i], nil
+		}
+		if atEOF {
+			return len(data), data, nil
+		}
+		return 0, nil, nil
+	})
+	return s
+}
+
 // commitSingletons / tagSingletons name the headers a well-formed object
 // carries at most once. parent is intentionally absent from the commit set
 // (merge commits have several); mergetag and encoding are intentionally
@@ -453,7 +473,7 @@ func checkUniqueHeaders(obj plumbing.EncodedObject, singletons map[string]bool)
 	defer r.Close() // nolint:errcheck
 
 	seen := make(map[string]bool, len(singletons))
-	scanner := bufio.NewScanner(r)
+	scanner := newRawObjectScanner(r)
 	for scanner.Scan() {
 		line := scanner.Bytes()
 		if len(line) == 0 {
diff --git a/pkg/git/rawobj_test.go b/pkg/git/rawobj_test.go
index cb44787e4..df7f77781 100644
--- a/pkg/git/rawobj_test.go
+++ b/pkg/git/rawobj_test.go
@@ -415,6 +415,36 @@ func TestSplitCommit_LenientContinuation(t *testing.T) {
 	}
 }
 
+// TestSplitCommit_TrailingCR confirms a trailing CR stays in the payload. git
+// hashes it as object content, so dropping it would collapse two distinct
+// commits onto one payload and let a single signature verify for both.
+func TestSplitCommit_TrailingCR(t *testing.T) {
+	sig := "gpgsig -----BEGIN SIGNED MESSAGE-----\n ZmFrZQ==\n -----END SIGNED MESSAGE-----\n"
+	base := "tree b333504b8cf3d9c314fed2cc242c5c38e89534a5\n" +
+		"author Alice <alice@example.com> 1700000000 +0000\n" +
+		"committer Alice <alice@example.com> 1700000000 +0000\n" + sig + "\n"
+	raw := []byte(base + "release v1.0\n")
+	rawCR := []byte(base + "release v1.0\r\n")
+
+	hash, _ := ObjectHash(raw)
+	hashCR, _ := ObjectHash(rawCR)
+	if hash == hashCR {
+		t.Fatalf("setup: expected distinct objects, both hashed to %s", hash)
+	}
+
+	c, err := SplitCommit(bytes.NewReader(raw))
+	if err != nil {
+		t.Fatalf("SplitCommit: %v", err)
+	}
+	cCR, err := SplitCommit(bytes.NewReader(rawCR))
+	if err != nil {
+		t.Fatalf("SplitCommit (CR variant): %v", err)
+	}
+	if bytes.Equal(c.Payload, cCR.Payload) {
+		t.Errorf("payloads should differ, both are %q", c.Payload)
+	}
+}
+
 // TestJoinCommit_RoundTrip uses the real signed HEAD commit: split it, join
 // it back, and confirm the bytes are identical.
 func TestJoinCommit_RoundTrip(t *testing.T) {
@@ -623,6 +653,29 @@ func TestSplitTag_NoHeaderTerminator(t *testing.T) {
 	}
 }
 
+// TestSplitTag_TrailingCR is the tag counterpart to
+// TestSplitCommit_TrailingCR.
+func TestSplitTag_TrailingCR(t *testing.T) {
+	sig := "-----BEGIN SIGNED MESSAGE-----\nZmFrZQ==\n-----END SIGNED MESSAGE-----\n"
+	base := "object 2d9cff2bad7132c586e128bcc23322dbb5297e8e\n" +
+		"type commit\ntag v1\n" +
+		"tagger Alice <alice@example.com> 1700000000 +0000\n\n"
+	raw := []byte(base + "release v1.0\n" + sig)
+	rawCR := []byte(base + "release v1.0\r\n" + sig)
+
+	tag, err := SplitTag(bytes.NewReader(raw))
+	if err != nil {
+		t.Fatalf("SplitTag: %v", err)
+	}
+	tagCR, err := SplitTag(bytes.NewReader(rawCR))
+	if err != nil {
+		t.Fatalf("SplitTag (CR variant): %v", err)
+	}
+	if bytes.Equal(tag.Payload, tagCR.Payload) {
+		t.Errorf("payloads should differ, both are %q", tag.Payload)
+	}
+}
+
 func TestJoinTag_RoundTrip(t *testing.T) {
 	raw := loadObject(t, "tag.txt")
 	tag, err := SplitTag(bytes.NewReader(raw))
@@ -663,6 +716,7 @@ func TestValidateCommit(t *testing.T) {
 		{"duplicate gpgsig-sha256", base + "gpgsig-sha256 sig1\ngpgsig-sha256 sig2\n\nmsg\n", true},
 		{"gpgsig with continuation lines does not count as duplicate", base + "gpgsig line1\n line2\n line3\n\nmsg\n", false},
 		{"duplicate headers in message body are ignored", base + "\ntree fake\ntree alsofake\n", false},
+		{"duplicate tree behind a bare-CR line", "tree aaaa\n" + authorLine + commLine + "\r\ntree bbbb\n\nmsg\n", true},
 	}
 	for _, tc := range tcs {
 		t.Run(tc.name, func(t *testing.T) {
@@ -693,6 +747,7 @@ func TestValidateTag(t *testing.T) {
 		{"duplicate tag", "object aaaa\ntype commit\ntag v1\ntag v2\n" + taggerLine + "\nmsg\n", true},
 		{"duplicate tagger", "object aaaa\ntype commit\ntag v1\n" + taggerLine + taggerLine + "\nmsg\n", true},
 		{"duplicate headers in message body are ignored", base + "\nobject fake\nobject alsofake\n", false},
+		{"duplicate object behind a bare-CR line", "object aaaa\ntype commit\ntag v1\n" + taggerLine + "\r\nobject bbbb\n\nmsg\n", true},
 	}
 	for _, tc := range tcs {
 		t.Run(tc.name, func(t *testing.T) {