Fix Responses API image forwarding

responses.go

@@ -80,7 +80,17 @@ func respToUpstream(instructions string, input json.RawMessage) (string, []upMes
 	}
 	for _, it := range items {
 		switch it.Type {
-		case "message":
+		case "message", "":
+			// Hermes (and some other Responses clients) send input items in the
+			// compact chat form {"role": "user", "content": "..."} with no
+			// `type` field. Treat a missing/empty type as a message so the user
+			// text is not silently dropped. See: ds4f "model can't see user"
+			// reports through Hermes -> sub2api -> qodercli2api.
+			if it.Type == "" && it.Role == "" && it.Content == nil && it.CallID == "" && it.Output == "" {
+				// not a recognizable message item; keep old default behavior
+				flushAssistant()
+				continue
+			}
 			flushAssistant()
 			if it.Role == "system" || it.Role == "developer" {
 				sysParts = append(sysParts, respPartsText(it.Content))
@@ -92,7 +102,9 @@ func respToUpstream(instructions string, input json.RawMessage) (string, []upMes
 			} else {
 				um.Content = respPartsToOAI(it.Content)
 			}
-			if txt := respPartsText(it.Content); txt != "" {
+			if it.Role == "user" {
+				um.Contents = respPartsToUpstream(it.Content)
+			} else if txt := respPartsText(it.Content); txt != "" {
 				um.Contents = []upPart{{Type: "text", Text: txt}}
 			}
 			msgs = append(msgs, um)
@@ -167,6 +179,23 @@ func respPartsToOAI(raw json.RawMessage) any {
 	return out
 }
 
+func respPartsToUpstream(raw json.RawMessage) []upPart {
+	parts := parseRespParts(raw)
+	out := make([]upPart, 0, len(parts))
+	for _, p := range parts {
+		switch p.Type {
+		case "input_text", "output_text", "text":
+			out = append(out, upPart{Type: "text", Text: p.Text})
+		case "input_image":
+			out = append(out, upPart{
+				Type:     "image_url",
+				ImageURL: &upImageURL{URL: p.ImageURL},
+			})
+		}
+	}
+	return out
+}
+
 func parseRespParts(raw json.RawMessage) []respContentPart {
 	if len(raw) == 0 {
 		return nil

responses_test.go

@@ -0,0 +1,61 @@
+package main
+
+import (
+	"encoding/json"
+	"testing"
+)
+
+func TestRespToUpstreamPreservesInputImage(t *testing.T) {
+	input := json.RawMessage(`[
+		{
+			"type": "message",
+			"role": "user",
+			"content": [
+				{"type": "input_text", "text": "describe this"},
+				{"type": "input_image", "image_url": "data:image/png;base64,abc"}
+			]
+		}
+	]`)
+
+	_, msgs, err := respToUpstream("", input)
+	if err != nil {
+		t.Fatalf("respToUpstream returned error: %v", err)
+	}
+	if len(msgs) != 1 {
+		t.Fatalf("got %d messages, want 1", len(msgs))
+	}
+	if len(msgs[0].Contents) != 2 {
+		t.Fatalf("got %d upstream content parts, want 2", len(msgs[0].Contents))
+	}
+	if got := msgs[0].Contents[0].Text; got != "describe this" {
+		t.Fatalf("text content = %q, want %q", got, "describe this")
+	}
+	if got := msgs[0].Contents[1].ImageURL.URL; got != "data:image/png;base64,abc" {
+		t.Fatalf("image URL = %q, want data URL", got)
+	}
+}
+
+func TestRespToUpstreamAcceptsCompactRoleContentItem(t *testing.T) {
+	// Hermes sends Responses input in this compact form: it has role/content
+	// but omits the optional type:"message" discriminator.
+	input := json.RawMessage(`[
+		{"role": "user", "content": "只回复四个字：测试成功"}
+	]`)
+
+	_, msgs, err := respToUpstream("", input)
+	if err != nil {
+		t.Fatalf("respToUpstream returned error: %v", err)
+	}
+	if len(msgs) != 1 {
+		t.Fatalf("got %d messages, want 1", len(msgs))
+	}
+	if got := msgs[0].Role; got != "user" {
+		t.Fatalf("role = %q, want user", got)
+	}
+	if got := msgs[0].Content; got != "只回复四个字：测试成功" {
+		t.Fatalf("content = %q, want compact user text", got)
+	}
+	if len(msgs[0].Contents) != 1 || msgs[0].Contents[0].Text != "只回复四个字：测试成功" {
+		t.Fatalf("upstream contents = %#v, want compact user text", msgs[0].Contents)
+	}
+}