feat: token used 展示拆分成 input 和 output

cmd/opencodereview/review_cmd.go

@@ -122,7 +122,7 @@ func runReview(args []string) error {
 	if len(comments) > 0 {
 		telemetry.RecordCommentsGenerated(ctx, int64(len(comments)))
 	}
-	telemetry.PrintTraceSummary(ag.FilesReviewed(), int64(len(comments)), ag.TotalTokensUsed(), duration)
+	telemetry.PrintTraceSummary(ag.FilesReviewed(), int64(len(comments)), ag.TotalInputTokens(), ag.TotalOutputTokens(), ag.TotalTokensUsed(), duration)
 
 	if opts.outputFormat == "json" {
 		return outputJSON(comments)

internal/agent/agent.go

@@ -94,7 +94,9 @@ type Agent struct {
 	totalDeletions  int64
 	currentDate     string
 	session         *session.SessionHistory
-	totalTokensUsed int64 // accumulated tokens from all LLM calls, accessed atomically
+	totalTokensUsed     int64 // accumulated total tokens from all LLM calls, accessed atomically
+	totalInputTokens    int64 // accumulated input/prompt tokens, accessed atomically
+	totalOutputTokens   int64 // accumulated completion tokens, accessed atomically
 }
 
 // CommentWorkerPool manages a fixed-size pool of workers dedicated to
@@ -225,6 +227,16 @@ func (a *Agent) TotalTokensUsed() int64 {
 	return atomic.LoadInt64(&a.totalTokensUsed)
 }
 
+// TotalInputTokens returns the accumulated input/prompt tokens from all LLM calls.
+func (a *Agent) TotalInputTokens() int64 {
+	return atomic.LoadInt64(&a.totalInputTokens)
+}
+
+// TotalOutputTokens returns the accumulated completion tokens from all LLM calls.
+func (a *Agent) TotalOutputTokens() int64 {
+	return atomic.LoadInt64(&a.totalOutputTokens)
+}
+
 // loadDiffs populates the diff-related fields.
 func (a *Agent) loadDiffs() error {
 	var provider *diff.Provider
@@ -460,6 +472,8 @@ func (a *Agent) executePlanPhase(ctx context.Context, newPath, rawDiff, changeFi
 	rec.SetResponse(resp, time.Since(startTime))
 	if resp.Usage != nil {
 		atomic.AddInt64(&a.totalTokensUsed, int64(resp.Usage.TotalTokens))
+		atomic.AddInt64(&a.totalInputTokens, int64(resp.Usage.PromptTokens+resp.Usage.CacheReadTokens))
+		atomic.AddInt64(&a.totalOutputTokens, int64(resp.Usage.CompletionTokens+resp.Usage.CacheWriteTokens))
 	}
 	fmt.Printf("[ocr] Plan completed for %s\n", newPath)
 	return resp.Content(), nil
@@ -538,6 +552,8 @@ func (a *Agent) performLlmCodeReview(ctx context.Context, messages []llm.Message
 		totalTokens := int64(0)
 		if resp.Usage != nil {
 			totalTokens = resp.Usage.TotalTokens
+			atomic.AddInt64(&a.totalInputTokens, int64(resp.Usage.PromptTokens+resp.Usage.CacheReadTokens))
+			atomic.AddInt64(&a.totalOutputTokens, int64(resp.Usage.CompletionTokens+resp.Usage.CacheWriteTokens))
 		}
 		telemetry.RecordLLMRequest(ctx, a.args.Model, duration, totalTokens, "ok")
 		atomic.AddInt64(&a.totalTokensUsed, totalTokens)
@@ -764,6 +780,8 @@ func (a *Agent) compressAndRecord(msgs []llm.Message, filePath string) []llm.Mes
 	rec.SetResponse(resp, duration)
 	if resp.Usage != nil {
 		atomic.AddInt64(&a.totalTokensUsed, int64(resp.Usage.TotalTokens))
+		atomic.AddInt64(&a.totalInputTokens, int64(resp.Usage.PromptTokens+resp.Usage.CacheReadTokens))
+		atomic.AddInt64(&a.totalOutputTokens, int64(resp.Usage.CompletionTokens+resp.Usage.CacheWriteTokens))
 	}
 
 	summary := resp.Content()

internal/llm/usage_resolver.go

@@ -7,7 +7,35 @@ import (
 
 // UsageInfo holds token usage extracted from an LLM API response.
 type UsageInfo struct {
-	TotalTokens int64 `json:"total_tokens"`
+	TotalTokens      int64 `json:"total_tokens"`
+	PromptTokens     int64 `json:"prompt_tokens"`
+	CompletionTokens int64 `json:"completion_tokens"`
+	CacheReadTokens  int64 `json:"cache_read_tokens,omitempty"`
+	CacheWriteTokens int64 `json:"cache_write_tokens,omitempty"`
+}
+
+var promptTokensPaths = []string{
+	"usage.prompt_tokens",        // OpenAI standard
+	"prompt_tokens",              // flat at root
+	"data.usage.prompt_tokens",   // wrapped in data layer
+}
+
+var completionTokensPaths = []string{
+	"usage.completion_tokens",      // OpenAI standard
+	"completion_tokens",            // flat at root
+	"data.usage.completion_tokens", // wrapped in data layer
+}
+
+var cacheReadTokensPaths = []string{
+	"usage.cache_read_input_tokens",   // Anthropic
+	"cache_read_input_tokens",          // flat at root
+	"usage.prompt_tokens_details.cache_tokens_hit", // some providers
+	"usage.prompt_tokens_details.cache_tokens",     // some providers
+}
+
+var cacheWriteTokensPaths = []string{
+	"usage.cache_creation_input_tokens",   // Anthropic / proxy
+	"cache_creation_input_tokens",          // flat at root
 }
 
 // totalTokensPaths is an ordered list of JSON paths to try when extracting
@@ -20,18 +48,37 @@ var totalTokensPaths = []string{
 }
 
 // resolveUsage parses raw JSON bytes into a map and extracts token usage
-// by probing totalTokensPaths sequentially. Returns nil if no path matches.
+// by probing configured paths sequentially. Returns nil if no total_tokens found.
 func resolveUsage(raw []byte) *UsageInfo {
 	var rawBody map[string]any
 	if err := json.Unmarshal(raw, &rawBody); err != nil {
 		return nil
 	}
 
-	total, ok := probePath(rawBody, totalTokensPaths)
-	if !ok {
+	total, hasAny := probePath(rawBody, totalTokensPaths)
+	prompt, _ := probePath(rawBody, promptTokensPaths)
+	completion, _ := probePath(rawBody, completionTokensPaths)
+	cacheRead, _ := probePath(rawBody, cacheReadTokensPaths)
+	cacheWrite, _ := probePath(rawBody, cacheWriteTokensPaths)
+
+	if !hasAny && prompt == 0 && completion == 0 {
 		return nil
 	}
-	return &UsageInfo{TotalTokens: total}
+
+	ui := &UsageInfo{
+		TotalTokens:      total,
+		PromptTokens:     prompt,
+		CompletionTokens: completion,
+		CacheReadTokens:  cacheRead,
+		CacheWriteTokens: cacheWrite,
+	}
+
+	// If TotalTokens wasn't explicitly available but we have prompt+completion, compute it.
+	if total == 0 && (prompt > 0 || completion > 0) {
+		ui.TotalTokens = prompt + completion + cacheRead + cacheWrite
+	}
+
+	return ui
 }
 
 // probePath walks through each candidate path in order, returning the first

internal/telemetry/events.go

@@ -60,9 +60,14 @@ func FormatDuration(dur time.Duration) string {
 }
 
 // PrintTraceSummary prints a one-line summary of the review to stdout.
-func PrintTraceSummary(filesReviewed, commentsGenerated int64, totalTokens int64, duration time.Duration) {
-	fmt.Printf("[ocr] Summary: %d file(s) reviewed, %d comment(s), ~%d token(s) used, %s elapsed\n",
-		filesReviewed, commentsGenerated, totalTokens, FormatDuration(duration))
+func PrintTraceSummary(filesReviewed, commentsGenerated int64, inputTokens, outputTokens, totalTokens int64, duration time.Duration) {
+	if inputTokens > 0 || outputTokens > 0 {
+		fmt.Printf("[ocr] Summary: %d file(s) reviewed, %d comment(s), ~%d token(s) used (input: ~%d, output: ~%d), %s elapsed\n",
+			filesReviewed, commentsGenerated, totalTokens, inputTokens, outputTokens, FormatDuration(duration))
+	} else {
+		fmt.Printf("[ocr] Summary: %d file(s) reviewed, %d comment(s), ~%d token(s) used, %s elapsed\n",
+			filesReviewed, commentsGenerated, totalTokens, FormatDuration(duration))
+	}
 }
 
 // PrintToolCallStarted prints a line when a tool begins execution.