fix: auto-generate prompt_tokens_details from prompt_cache_hit_tokens in MarshalJSON

When only the legacy DeepSeek fields (prompt_cache_hit_tokens) are set but
the OpenAI-standard prompt_tokens_details is nil, MarshalJSON now auto-
generates the nested object. This ensures dsh (which reads the standard
format first) sees cache hit data regardless of which adapter format the
upstream uses.

internal/types/types.go

@@ -107,6 +107,14 @@ type PromptTokensDetails struct {
 // standard keys (prompt_tokens/completion_tokens/total_tokens). Standard
 // clients such as DSH and DevEco Code read the *_tokens fields.
 func (t TokenUsage) MarshalJSON() ([]byte, error) {
+	// Auto-generate prompt_tokens_details from legacy DeepSeek fields when
+	// the upstream adapter only set the standalone hit count (openai.lua
+	// does this in Lua, but other adapters or the standardSSEChunk fallback
+	// may not). dsh reads prompt_tokens_details.cached_tokens first.
+	pdetails := t.PromptTokensDetails
+	if pdetails == nil && t.PromptCacheHit > 0 {
+		pdetails = &PromptTokensDetails{CachedTokens: t.PromptCacheHit}
+	}
 	return json.Marshal(struct {
 		PromptTokens          int                   `json:"prompt_tokens"`
 		CompletionTokens       int                   `json:"completion_tokens"`
@@ -124,7 +132,7 @@ func (t TokenUsage) MarshalJSON() ([]byte, error) {
 		Prompt:               t.Prompt,
 		Completion:           t.Completion,
 		Total:                t.Total,
-		PromptTokensDetails:  t.PromptTokensDetails,
+		PromptTokensDetails:  pdetails,
 		PromptCacheHitTokens:  t.PromptCacheHit,
 		PromptCacheMissTokens: t.PromptCacheMiss,
 	})

internal/types/types_test.go

@@ -0,0 +1,70 @@
+package types
+
+import (
+	"encoding/json"
+	"testing"
+)
+
+func TestTokenUsageMarshalCacheFields(t *testing.T) {
+	// DeepSeek-style: prompt_cache_hit_tokens + prompt_tokens_details auto-generated
+	u := TokenUsage{
+		Prompt:          100,
+		Completion:      50,
+		Total:           150,
+		PromptCacheHit:  80,
+		PromptCacheMiss: 20,
+	}
+	b, _ := json.Marshal(u)
+	var out map[string]interface{}
+	json.Unmarshal(b, &out)
+	check := func(name string, want interface{}) {
+		got := out[name]
+		if got != want {
+			t.Errorf("%s = %v (want %v)", name, got, want)
+		}
+	}
+	check("prompt_tokens", 100.0)
+	check("completion_tokens", 50.0)
+	check("prompt_cache_hit_tokens", 80.0)
+	check("prompt_cache_miss_tokens", 20.0)
+	details := out["prompt_tokens_details"]
+	if details == nil {
+		t.Error("prompt_tokens_details is missing (should be set from prompt_cache_hit_tokens?)")
+	} else {
+		d := details.(map[string]interface{})
+		if d["cached_tokens"] != 80.0 {
+			t.Errorf("cached_tokens = %v (want 80)", d["cached_tokens"])
+		}
+	}
+
+	// OpenAI-style: prompt_tokens_details.cached_tokens
+	u2 := TokenUsage{
+		Prompt:     200,
+		Completion: 100,
+		Total:      300,
+		PromptTokensDetails: &PromptTokensDetails{
+			CachedTokens: 150,
+		},
+	}
+	b2, _ := json.Marshal(u2)
+	json.Unmarshal(b2, &out)
+	check2 := func(name string, want interface{}) {
+		got := out[name]
+		if got != want {
+			t.Errorf("OpenAI %s = %v (want %v)", name, got, want)
+		}
+	}
+	check2("prompt_tokens", 200.0)
+	details2 := out["prompt_tokens_details"]
+	if details2 == nil {
+		t.Error("OpenAI prompt_tokens_details is missing")
+	} else {
+		d := details2.(map[string]interface{})
+		if d["cached_tokens"] != 150.0 {
+			t.Errorf("OpenAI cached_tokens = %v (want 150)", d["cached_tokens"])
+		}
+	}
+
+	t.Logf("DeepSeek JSON: %s", string(b))
+	t.Logf("OpenAI JSON: %s", string(b2))
+}