Method handleAdminAPI

proxy/handler.go:2147–2271

==================== 管理 API ====================

(w http.ResponseWriter, r *http.Request)

				"delta":         map[string]interface{}{},
				"finish_reason": finishReason,
			}},
			"usage": map[string]int{
				"prompt_tokens":     inputTokens,
				"completion_tokens": outputTokens,
				"total_tokens":      inputTokens + outputTokens,
			},
		}
		data, _ := json.Marshal(chunk)
		fmt.Fprintf(w, "data: %s\n\n", string(data))
		fmt.Fprintf(w, "data: [DONE]\n\n")
		flusher.Flush()
		return
	}

	if lastErr == nil {
		h.sendOpenAIError(w, 503, "server_error", "No available accounts")
		return
	}

	h.recordFailureWithDetails("openai", model, "", lastErr)
	h.sendOpenAIError(w, 500, "server_error", lastErr.Error())
}

// handleOpenAINonStream OpenAI 非流式响应
func (h *Handler) handleOpenAINonStream(ctx context.Context, w http.ResponseWriter, payload *KiroPayload, model string, thinking bool, estimatedInputTokens int, apiKeyID string) {
	excluded := make(map[string]bool)
	var lastErr error
	reqStart := time.Now()

	for attempt := 0; attempt < maxAccountRetryAttempts; attempt++ {
		account := h.pool.GetNextForModelExcluding(model, excluded)
		if account == nil {
			break
		}
		if err := h.ensureValidToken(account); err != nil {
			lastErr = err
			excluded[account.ID] = true
			h.handleAccountFailure(account, err)
			continue
		}

		var content string
		var reasoningContent string
		var toolUses []KiroToolUse
		var inputTokens, outputTokens int
		var credits float64
		var realInputTokens int
		var upstreamStopReason string

		callback := &KiroStreamCallback{
			OnText: func(text string, isThinking bool) {
				if isThinking {
					reasoningContent += text
				} else {
					content += text
				}
			},
			OnToolUse:  func(tu KiroToolUse) { toolUses = append(toolUses, tu) },