agent-Specialization/server/chat_flow_runner.py

from __future__ import annotations

import asyncio
import json
import time
import re
import zipfile
from collections import defaultdict, Counter, deque
from datetime import datetime, timedelta
from pathlib import Path
from typing import Any, Dict, List, Optional, Tuple

from werkzeug.utils import secure_filename

from config import (
    OUTPUT_FORMATS,
    AUTO_FIX_TOOL_CALL,
    AUTO_FIX_MAX_ATTEMPTS,
    MAX_ITERATIONS_PER_TASK,
    MAX_CONSECUTIVE_SAME_TOOL,
    MAX_TOTAL_TOOL_CALLS,
    TOOL_CALL_COOLDOWN,
    MAX_UPLOAD_SIZE,
    DEFAULT_CONVERSATIONS_LIMIT,
    MAX_CONVERSATIONS_LIMIT,
    CONVERSATIONS_DIR,
    DEFAULT_RESPONSE_MAX_TOKENS,
    DEFAULT_PROJECT_PATH,
    LOGS_DIR,
    AGENT_VERSION,
    THINKING_FAST_INTERVAL,
    PROJECT_MAX_STORAGE_MB,
    PROJECT_MAX_STORAGE_BYTES,
    UPLOAD_SCAN_LOG_SUBDIR,
)
from modules.personalization_manager import (
    load_personalization_config,
    save_personalization_config,
    THINKING_INTERVAL_MIN,
    THINKING_INTERVAL_MAX,
)
from modules.upload_security import UploadSecurityError
from modules.user_manager import UserWorkspace
from modules.usage_tracker import QUOTA_DEFAULTS
from core.web_terminal import WebTerminal
from utils.tool_result_formatter import format_tool_result_for_context
from utils.conversation_manager import ConversationManager
from config.model_profiles import get_model_context_window, get_model_profile

from .auth_helpers import api_login_required, resolve_admin_policy, get_current_user_record, get_current_username
from .context import with_terminal, get_gui_manager, get_upload_guard, build_upload_error_response, ensure_conversation_loaded, reset_system_state, get_user_resources, get_or_create_usage_tracker
from .utils_common import (
    build_review_lines,
    debug_log,
    log_backend_chunk,
    log_frontend_chunk,
    log_streaming_debug_entry,
    brief_log,
    DEBUG_LOG_FILE,
    CHUNK_BACKEND_LOG_FILE,
    CHUNK_FRONTEND_LOG_FILE,
    STREAMING_DEBUG_LOG_FILE,
)
from .security import rate_limited, format_tool_result_notice, compact_web_search_result, consume_socket_token, prune_socket_tokens, validate_csrf_request, requires_csrf_protection, get_csrf_token
from .monitor import cache_monitor_snapshot, get_cached_monitor_snapshot
from .extensions import socketio
from .state import (
    MONITOR_FILE_TOOLS,
    MONITOR_MEMORY_TOOLS,
    MONITOR_SNAPSHOT_CHAR_LIMIT,
    MONITOR_MEMORY_ENTRY_LIMIT,
    RATE_LIMIT_BUCKETS,
    FAILURE_TRACKERS,
    pending_socket_tokens,
    usage_trackers,
    MONITOR_SNAPSHOT_CACHE,
    MONITOR_SNAPSHOT_CACHE_LIMIT,
    PROJECT_STORAGE_CACHE,
    PROJECT_STORAGE_CACHE_TTL_SECONDS,
    RECENT_UPLOAD_EVENT_LIMIT,
    RECENT_UPLOAD_FEED_LIMIT,
    THINKING_FAILURE_KEYWORDS,
    TITLE_PROMPT_PATH,
    get_last_active_ts,
    user_manager,
    container_manager,
    custom_tool_registry,
    user_terminals,
    terminal_rooms,
    connection_users,
    stop_flags,
    get_stop_flag,
    set_stop_flag,
    clear_stop_flag,
)
from .chat_flow_helpers import (
    detect_malformed_tool_call as _detect_malformed_tool_call,
    detect_tool_failure,
    get_thinking_state,
    mark_force_thinking as _mark_force_thinking,
    mark_suppress_thinking,
    apply_thinking_schedule as _apply_thinking_schedule,
    update_thinking_after_call as _update_thinking_after_call,
    maybe_mark_failure_from_message as _maybe_mark_failure_from_message,
    generate_conversation_title_background as _generate_conversation_title_background,
)


from .chat_flow_runner_helpers import (
    extract_intent_from_partial,
    resolve_monitor_path,
    resolve_monitor_memory,
    capture_monitor_snapshot,
)


def generate_conversation_title_background(web_terminal: WebTerminal, conversation_id: str, user_message: str, username: str):
    return _generate_conversation_title_background(
        web_terminal=web_terminal,
        conversation_id=conversation_id,
        user_message=user_message,
        username=username,
        socketio_instance=socketio,
        title_prompt_path=TITLE_PROMPT_PATH,
        debug_logger=debug_log,
    )


def mark_force_thinking(terminal: WebTerminal, reason: str = ""):
    return _mark_force_thinking(terminal, reason=reason, debug_logger=debug_log)


def apply_thinking_schedule(terminal: WebTerminal):
    return _apply_thinking_schedule(terminal, default_interval=THINKING_FAST_INTERVAL, debug_logger=debug_log)


def update_thinking_after_call(terminal: WebTerminal):
    return _update_thinking_after_call(terminal, debug_logger=debug_log)


def maybe_mark_failure_from_message(terminal: WebTerminal, content: Optional[str]):
    return _maybe_mark_failure_from_message(
        terminal,
        content,
        failure_keywords=THINKING_FAILURE_KEYWORDS,
        debug_logger=debug_log,
    )


def detect_malformed_tool_call(text):
    return _detect_malformed_tool_call(text)


async def handle_task_with_sender(terminal: WebTerminal, workspace: UserWorkspace, message, images, sender, client_sid, username: str, videos=None):
    """处理任务并发送消息 - 集成token统计版本"""
    web_terminal = terminal
    conversation_id = getattr(web_terminal.context_manager, "current_conversation_id", None)
    videos = videos or []
    raw_sender = sender

    def sender(event_type, data):
        """为关键事件补充会话标识，便于前端定位报错归属。"""
        if not isinstance(data, dict):
            raw_sender(event_type, data)
            return
        payload = data
        if event_type in {"error", "quota_exceeded", "task_stopped", "task_complete"}:
            payload = dict(data)
            current_conv = conversation_id or getattr(web_terminal.context_manager, "current_conversation_id", None)
            if current_conv:
                payload.setdefault("conversation_id", current_conv)
            task_id = getattr(web_terminal, "task_id", None) or client_sid
            if task_id:
                payload.setdefault("task_id", task_id)
            if client_sid:
                payload.setdefault("client_sid", client_sid)
        raw_sender(event_type, payload)
    
    # 如果是思考模式，重置状态
    if web_terminal.thinking_mode:
        web_terminal.api_client.start_new_task(force_deep=web_terminal.deep_thinking_mode)
        state = get_thinking_state(web_terminal)
        state["fast_streak"] = 0
        state["force_next"] = False
        state["suppress_next"] = False
    
    # 添加到对话历史
    history_len_before = len(getattr(web_terminal.context_manager, "conversation_history", []) or [])
    is_first_user_message = history_len_before == 0
    web_terminal.context_manager.add_conversation("user", message, images=images, videos=videos)
    
    if is_first_user_message and getattr(web_terminal, "context_manager", None):
        try:
            personal_config = load_personalization_config(workspace.data_dir)
        except Exception:
            personal_config = {}
        auto_title_enabled = personal_config.get("auto_generate_title", True)
        if auto_title_enabled:
            conv_id = getattr(web_terminal.context_manager, "current_conversation_id", None)
            socketio.start_background_task(
                generate_conversation_title_background,
                web_terminal,
                conv_id,
                message,
                username
            )
    
    # === 移除：不在这里计算输入token，改为在每次API调用前计算 ===
    
    # 构建上下文和消息（用于API调用）
    context = web_terminal.build_context()
    messages = web_terminal.build_messages(context, message)
    tools = web_terminal.define_tools()
    try:
        profile = get_model_profile(getattr(web_terminal, "model_key", None) or "kimi-k2.5")
        web_terminal.apply_model_profile(profile)
    except Exception as exc:
        debug_log(f"更新模型配置失败: {exc}")

    # === 上下文预算与安全校验（避免超出模型上下文） ===
    max_context_tokens = get_model_context_window(getattr(web_terminal, "model_key", None) or "kimi-k2.5")
    current_tokens = web_terminal.context_manager.get_current_context_tokens(conversation_id)
    # 提前同步给底层客户端，动态收缩 max_tokens
    web_terminal.api_client.update_context_budget(current_tokens, max_context_tokens)
    if max_context_tokens:
        if current_tokens >= max_context_tokens:
            err_msg = (
                f"当前对话上下文已达 {current_tokens} tokens，超过模型上限 "
                f"{max_context_tokens}，请先使用压缩功能或清理对话后再试。"
            )
            debug_log(err_msg)
            web_terminal.context_manager.add_conversation("system", err_msg)
            sender('error', {
                'message': err_msg,
                'status_code': 400,
                'error_type': 'context_overflow'
            })
            return
        usage_percent = (current_tokens / max_context_tokens) * 100
        warned = web_terminal.context_manager.conversation_metadata.get("context_warning_sent", False)
        if usage_percent >= 70 and not warned:
            warn_msg = (
                f"当前对话上下文约占 {usage_percent:.1f}%（{current_tokens}/{max_context_tokens}），"
                "建议使用压缩功能。"
            )
            web_terminal.context_manager.conversation_metadata["context_warning_sent"] = True
            web_terminal.context_manager.auto_save_conversation(force=True)
            sender('context_warning', {
                'title': '上下文过长',
                'message': warn_msg,
                'type': 'warning',
                'conversation_id': conversation_id
            })
    
    # 开始新的AI消息
    sender('ai_message_start', {})
    
    # 增量保存相关变量
    accumulated_response = ""   # 累积的响应内容
    is_first_iteration = True   # 是否是第一次迭代
    
    # 统计和限制变量
    total_iterations = 0
    total_tool_calls = 0
    consecutive_same_tool = defaultdict(int)
    last_tool_name = ""
    auto_fix_attempts = 0
    last_tool_call_time = 0
    detected_tool_intent: Dict[str, str] = {}
    
    # 设置最大迭代次数（API 可覆盖）；None 表示不限制
    max_iterations_override = getattr(web_terminal, "max_iterations_override", None)
    max_iterations = max_iterations_override if max_iterations_override is not None else MAX_ITERATIONS_PER_TASK
    max_api_retries = 4
    retry_delay_seconds = 10
    
    pending_append = None  # {"path": str, "tool_call_id": str, "buffer": str, ...}
    append_probe_buffer = ""
    pending_modify = None  # {"path": str, "tool_call_id": str, "buffer": str, ...}
    modify_probe_buffer = ""

    async def finalize_pending_append(response_text: str, stream_completed: bool, finish_reason: str = None) -> Dict:
        """在流式输出结束后处理追加写入"""
        nonlocal pending_append, append_probe_buffer
        
        result = {
            "handled": False,
            "success": False,
            "summary": None,
            "summary_message": None,
            "tool_content": None,
            "tool_call_id": None,
            "path": None,
            "forced": False,
            "error": None,
            "assistant_content": response_text,
            "lines": 0,
            "bytes": 0,
            "finish_reason": finish_reason,
            "appended_content": "",
            "assistant_metadata": None
        }
        
        if not pending_append:
            return result
        
        state = pending_append
        path = state.get("path")
        tool_call_id = state.get("tool_call_id")
        buffer = state.get("buffer", "")
        start_marker = state.get("start_marker")
        end_marker = state.get("end_marker")
        start_idx = state.get("content_start")
        end_idx = state.get("end_index")
        
        display_id = state.get("display_id")
        
        result.update({
            "handled": True,
            "path": path,
            "tool_call_id": tool_call_id,
            "display_id": display_id
        })
        
        if path is None or tool_call_id is None:
            error_msg = "append_to_file 状态不完整，缺少路径或ID。"
            debug_log(error_msg)
            result["error"] = error_msg
            result["summary_message"] = error_msg
            result["tool_content"] = json.dumps({
                "success": False,
                "error": error_msg
            }, ensure_ascii=False)
            if display_id:
                sender('update_action', {
                    'id': display_id,
                    'status': 'failed',
                    'preparing_id': tool_call_id,
                    'message': error_msg
                })
            pending_append = None
            return result
        
        if start_idx is None:
            error_msg = f"未检测到格式正确的开始标识 {start_marker}。"
            debug_log(error_msg)
            result["error"] = error_msg
            result["summary_message"] = error_msg
            result["tool_content"] = json.dumps({
                "success": False,
                "path": path,
                "error": error_msg
            }, ensure_ascii=False)
            if display_id:
                sender('update_action', {
                    'id': display_id,
                    'status': 'failed',
                    'preparing_id': tool_call_id,
                    'message': error_msg
                })
            pending_append = None
            return result
        
        forced = False
        if end_idx is None:
            forced = True
            # 查找下一个<<<，否则使用整个缓冲结尾
            remaining = buffer[start_idx:]
            next_marker = remaining.find("<<<", len(end_marker))
            if next_marker != -1:
                end_idx = start_idx + next_marker
            else:
                end_idx = len(buffer)
        
        content = buffer[start_idx:end_idx]
        if content.startswith('\n'):
            content = content[1:]
        
        if not content:
            error_msg = "未检测到需要追加的内容，请严格按照<<<APPEND:path>>>...<<<END_APPEND>>>格式输出。"
            debug_log(error_msg)
            result["error"] = error_msg
            result["forced"] = forced
            result["tool_content"] = json.dumps({
                "success": False,
                "path": path,
                "error": error_msg
            }, ensure_ascii=False)
            if display_id:
                sender('update_action', {
                    'id': display_id,
                    'status': 'failed',
                    'preparing_id': tool_call_id,
                    'message': error_msg
                })
            pending_append = None
            return result
        
        assistant_message_lines = []
        if start_marker:
            assistant_message_lines.append(start_marker)
        assistant_message_lines.append(content)
        if not forced and end_marker:
            assistant_message_lines.append(end_marker)
        assistant_message_text = "\n".join(assistant_message_lines)
        result["assistant_content"] = assistant_message_text
        assistant_metadata = {
            "append_payload": {
                "path": path,
                "tool_call_id": tool_call_id,
                "forced": forced,
                "has_end_marker": not forced
            }
        }
        result["assistant_metadata"] = assistant_metadata
        
        write_result = web_terminal.file_manager.append_file(path, content)
        if write_result.get("success"):
            bytes_written = len(content.encode('utf-8'))
            line_count = content.count('\n')
            if content and not content.endswith('\n'):
                line_count += 1
            
            summary = f"已向 {path} 追加 {line_count} 行（{bytes_written} 字节）"
            if forced:
                summary += "。未检测到 <<<END_APPEND>>> 标记，系统已在流结束处完成写入。如内容未完成，请重新调用 append_to_file 并按标准格式补充；如已完成，可继续后续步骤。"
            
            result.update({
                "success": True,
                "summary": summary,
                "summary_message": summary,
                "forced": forced,
                "lines": line_count,
                "bytes": bytes_written,
                "appended_content": content,
                "tool_content": json.dumps({
                    "success": True,
                    "path": path,
                    "lines": line_count,
                    "bytes": bytes_written,
                    "forced": forced,
                    "message": summary,
                    "finish_reason": finish_reason
                }, ensure_ascii=False)
            })
            
            assistant_meta_payload = result["assistant_metadata"]["append_payload"]
            assistant_meta_payload["lines"] = line_count
            assistant_meta_payload["bytes"] = bytes_written
            assistant_meta_payload["success"] = True
            
            summary_payload = {
                "success": True,
                "path": path,
                "lines": line_count,
                "bytes": bytes_written,
                "forced": forced,
                "message": summary
            }
            
            if display_id:
                sender('update_action', {
                    'id': display_id,
                    'status': 'completed',
                    'result': summary_payload,
                    'preparing_id': tool_call_id,
                    'message': summary
                })
            
            debug_log(f"追加写入完成: {summary}")
        else:
            error_msg = write_result.get("error", "追加写入失败")
            result.update({
                "error": error_msg,
                "summary_message": error_msg,
                "forced": forced,
                "appended_content": content,
                "tool_content": json.dumps({
                    "success": False,
                    "path": path,
                    "error": error_msg,
                    "finish_reason": finish_reason
                }, ensure_ascii=False)
            })
            debug_log(f"追加写入失败: {error_msg}")
            
            if result["assistant_metadata"]:
                assistant_meta_payload = result["assistant_metadata"]["append_payload"]
                assistant_meta_payload["lines"] = content.count('\n') + (0 if content.endswith('\n') or not content else 1)
                assistant_meta_payload["bytes"] = len(content.encode('utf-8'))
                assistant_meta_payload["success"] = False
            
            failure_payload = {
                "success": False,
                "path": path,
                "error": error_msg,
                "forced": forced
            }
            
            if display_id:
                sender('update_action', {
                    'id': display_id,
                    'status': 'completed',
                    'result': failure_payload,
                    'preparing_id': tool_call_id,
                    'message': error_msg
                })
        
        pending_append = None
        append_probe_buffer = ""
        if hasattr(web_terminal, "pending_append_request"):
            web_terminal.pending_append_request = None
        return result
    
    async def finalize_pending_modify(response_text: str, stream_completed: bool, finish_reason: str = None) -> Dict:
        """在流式输出结束后处理修改写入"""
        nonlocal pending_modify, modify_probe_buffer
        
        result = {
            "handled": False,
            "success": False,
            "path": None,
            "tool_call_id": None,
            "display_id": None,
            "total_blocks": 0,
            "completed_blocks": [],
            "failed_blocks": [],
            "forced": False,
            "details": [],
            "error": None,
            "assistant_content": response_text,
            "assistant_metadata": None,
            "tool_content": None,
            "summary_message": None,
            "finish_reason": finish_reason
        }
        
        if not pending_modify:
            return result
        
        state = pending_modify
        path = state.get("path")
        tool_call_id = state.get("tool_call_id")
        display_id = state.get("display_id")
        start_marker = state.get("start_marker")
        end_marker = state.get("end_marker")
        buffer = state.get("buffer", "")
        raw_buffer = state.get("raw_buffer", "")
        end_index = state.get("end_index")
        
        result.update({
            "handled": True,
            "path": path,
            "tool_call_id": tool_call_id,
            "display_id": display_id
        })
        
        if not state.get("start_seen"):
            error_msg = "未检测到格式正确的 <<<MODIFY:path>>> 标记。"
            debug_log(error_msg)
            result["error"] = error_msg
            result["summary_message"] = error_msg
            result["tool_content"] = json.dumps({
                "success": False,
                "path": path,
                "error": error_msg,
                "finish_reason": finish_reason
            }, ensure_ascii=False)
            if display_id:
                sender('update_action', {
                    'id': display_id,
                    'status': 'failed',
                    'preparing_id': tool_call_id,
                    'message': error_msg
                })
            if hasattr(web_terminal, "pending_modify_request"):
                web_terminal.pending_modify_request = None
            pending_modify = None
            modify_probe_buffer = ""
            return result
        
        forced = end_index is None
        apply_text = buffer if forced else buffer[:end_index]
        raw_content = raw_buffer if forced else raw_buffer[:len(start_marker) + end_index + len(end_marker)]
        if raw_content:
            result["assistant_content"] = raw_content
        
        blocks_info = []
        block_reports = {}
        detected_indices = set()
        block_pattern = re.compile(r"\[replace:(\d+)\](.*?)\[/replace\]", re.DOTALL)
        structure_warnings: List[str] = []
        structure_detail_entries: List[Dict] = []

        def record_structure_warning(message: str, hint: Optional[str] = None):
            """记录结构性缺陷，便于给出更具体的反馈。"""
            if message in structure_warnings:
                return
            structure_warnings.append(message)
            structure_detail_entries.append({
                "index": 0,
                "status": "failed",
                "reason": message,
                "removed_lines": 0,
                "added_lines": 0,
                "hint": hint or "请严格按照模板输出：[replace:n] + <<OLD>>/<<NEW>> + [/replace]，并使用 <<<END_MODIFY>>> 收尾。"
            })
        
        def extract_segment(body: str, tag: str):
            marker = f"<<{tag}>>"
            end_tag = "<<END>>"
            start_pos = body.find(marker)
            if start_pos == -1:
                return None, f"缺少 {marker}"
            start_pos += len(marker)
            if body[start_pos:start_pos+2] == "\r\n":
                start_pos += 2
            elif body[start_pos:start_pos+1] == "\n":
                start_pos += 1
            end_pos = body.find(end_tag, start_pos)
            if end_pos == -1:
                return None, f"缺少 {end_tag}"
            segment = body[start_pos:end_pos]
            return segment, None
        
        for match in block_pattern.finditer(apply_text):
            try:
                index = int(match.group(1))
            except ValueError:
                continue
            body = match.group(2)
            if index in detected_indices:
                continue
            detected_indices.add(index)
            block_reports[index] = {
                "index": index,
                "status": "pending",
                "reason": None,
                "removed_lines": 0,
                "added_lines": 0,
                "hint": None
            }
            old_content, old_error = extract_segment(body, "OLD")
            new_content, new_error = extract_segment(body, "NEW")
            if old_error or new_error:
                reason = old_error or new_error
                block_reports[index]["status"] = "failed"
                block_reports[index]["reason"] = reason
            blocks_info.append({
                "index": index,
                "old": old_content,
                "new": new_content,
                "error": old_error or new_error
            })
        
        if not blocks_info:
            has_replace_start = bool(re.search(r"\[replace:\s*\d+\]", apply_text))
            has_replace_end = "[/replace]" in apply_text
            has_old_tag = "<<OLD>>" in apply_text
            has_new_tag = "<<NEW>>" in apply_text

            if has_replace_start and not has_replace_end:
                record_structure_warning("检测到 [replace:n] 标记但缺少对应的 [/replace] 结束标记。")
            if has_replace_end and not has_replace_start:
                record_structure_warning("检测到 [/replace] 结束标记但缺少对应的 [replace:n] 起始标记。")

            old_tags = len(re.findall(r"<<OLD>>", apply_text))
            completed_old_tags = len(re.findall(r"<<OLD>>[\s\S]*?<<END>>", apply_text))
            if old_tags and completed_old_tags < old_tags:
                record_structure_warning("检测到 <<OLD>> 段落但未看到对应的 <<END>> 结束标记。")

            new_tags = len(re.findall(r"<<NEW>>", apply_text))
            completed_new_tags = len(re.findall(r"<<NEW>>[\s\S]*?<<END>>", apply_text))
            if new_tags and completed_new_tags < new_tags:
                record_structure_warning("检测到 <<NEW>> 段落但未看到对应的 <<END>> 结束标记。")

            if (has_replace_start or has_replace_end or has_old_tag or has_new_tag) and not structure_warnings:
                record_structure_warning("检测到部分补丁标记，但整体结构不完整，请严格按照模板填写所有标记。")

        total_blocks = len(blocks_info)
        result["total_blocks"] = total_blocks
        if forced:
            debug_log("未检测到 <<<END_MODIFY>>>，将在流结束处执行已识别的修改块。")
            result["forced"] = True
        
        blocks_to_apply = [
            {"index": block["index"], "old": block["old"], "new": block["new"]}
            for block in blocks_info
            if block["error"] is None and block["old"] is not None and block["new"] is not None
        ]
        
        # 记录格式残缺的块
        for block in blocks_info:
            if block["error"]:
                idx = block["index"]
                block_reports[idx]["status"] = "failed"
                block_reports[idx]["reason"] = block["error"]
                block_reports[idx]["hint"] = "请检查补丁块的 OLD/NEW 标记是否完整，必要时复用 terminal_snapshot 或终端命令重新调整。"
        
        apply_result = {}
        if blocks_to_apply:
            apply_result = web_terminal.file_manager.apply_modify_blocks(path, blocks_to_apply)
        else:
            apply_result = {"success": False, "completed": [], "failed": [], "results": [], "write_performed": False, "error": None}
        
        block_result_map = {item["index"]: item for item in apply_result.get("results", [])}
        
        for block in blocks_info:
            idx = block["index"]
            report = block_reports.get(idx)
            if report is None:
                continue
            if report["status"] == "failed":
                continue
            block_apply = block_result_map.get(idx)
            if not block_apply:
                report["status"] = "failed"
                report["reason"] = "未执行，可能未找到匹配原文"
                report["hint"] = report.get("hint") or "请确认 OLD 文本与文件内容完全一致；若多次失败，可改用终端命令/Python 进行精准替换。"
                continue
            status = block_apply.get("status")
            report["removed_lines"] = block_apply.get("removed_lines", 0)
            report["added_lines"] = block_apply.get("added_lines", 0)
            if block_apply.get("hint"):
                report["hint"] = block_apply.get("hint")
            if status == "success":
                report["status"] = "completed"
            elif status == "not_found":
                report["status"] = "failed"
                report["reason"] = block_apply.get("reason") or "未找到匹配的原文"
                if not report.get("hint"):
                    report["hint"] = "请使用 terminal_snapshot/grep -n 校验原文，或在说明后改用 run_command/python 精确替换。"
            else:
                report["status"] = "failed"
                report["reason"] = block_apply.get("reason") or "替换失败"
                if not report.get("hint"):
                    report["hint"] = block_apply.get("hint") or "若多次尝试仍失败，可考虑利用终端命令或 Python 小脚本完成此次修改。"
        
        completed_blocks = sorted([idx for idx, rep in block_reports.items() if rep["status"] == "completed"])
        failed_blocks = sorted([idx for idx, rep in block_reports.items() if rep["status"] != "completed"])
        
        result["completed_blocks"] = completed_blocks
        result["failed_blocks"] = failed_blocks
        details = sorted(block_reports.values(), key=lambda x: x["index"])
        if structure_detail_entries:
            details = structure_detail_entries + details
        result["details"] = details
        
        summary_parts = []
        if total_blocks == 0:
            summary_parts.append("未检测到有效的修改块，未执行任何修改。")
            summary_parts.extend(structure_warnings)
        else:
            if not completed_blocks and failed_blocks:
                summary_parts.append(f"共检测到 {total_blocks} 个修改块，全部未执行。")
            elif completed_blocks and not failed_blocks:
                summary_parts.append(f"共 {total_blocks} 个修改块全部完成。")
            else:
                summary_parts.append(
                    f"共检测到 {total_blocks} 个修改块，其中成功 {len(completed_blocks)} 个，失败 {len(failed_blocks)} 个。"
                )
        if forced:
            summary_parts.append("未检测到 <<<END_MODIFY>>> 标记，系统已在流结束处执行补丁。")
        if apply_result.get("error"):
            summary_parts.append(apply_result["error"])
        
        matching_note = "提示：补丁匹配基于完整文本，包含注释和空白符，请确保 <<<OLD>>> 段落与文件内容逐字一致。如果修改成功，请忽略，如果失败，请明确原文后再次尝试。"
        summary_parts.append(matching_note)
        summary_message = " ".join(summary_parts).strip()
        result["summary_message"] = summary_message
        result["success"] = bool(completed_blocks) and not failed_blocks and apply_result.get("error") is None
        
        tool_payload = {
            "success": result["success"],
            "path": path,
            "total_blocks": total_blocks,
            "completed": completed_blocks,
            "failed": [
                {
                    "index": rep["index"],
                    "reason": rep.get("reason"),
                    "hint": rep.get("hint")
                }
                for rep in result["details"] if rep["status"] != "completed"
            ],
            "forced": forced,
            "message": summary_message,
            "finish_reason": finish_reason,
            "details": result["details"]
        }
        if apply_result.get("error"):
            tool_payload["error"] = apply_result["error"]
        
        result["tool_content"] = json.dumps(tool_payload, ensure_ascii=False)
        result["assistant_metadata"] = {
            "modify_payload": {
                "path": path,
                "total_blocks": total_blocks,
                "completed": completed_blocks,
                "failed": failed_blocks,
                "forced": forced,
                "details": result["details"]
            }
        }
        
        if display_id:
            sender('update_action', {
                'id': display_id,
                'status': 'completed' if result["success"] else 'failed',
                'result': tool_payload,
                'preparing_id': tool_call_id,
                'message': summary_message
            })
        
        pending_modify = None
        modify_probe_buffer = ""
        if hasattr(web_terminal, "pending_modify_request"):
            web_terminal.pending_modify_request = None
        return result

    async def process_sub_agent_updates(
        messages: List[Dict],
        inline: bool = False,
        after_tool_call_id: Optional[str] = None
    ):
        """轮询子智能体任务并通知前端，并把结果插入当前对话上下文。"""
        manager = getattr(web_terminal, "sub_agent_manager", None)
        if not manager:
            return
        try:
            updates = manager.poll_updates()
            debug_log(f"[SubAgent] poll inline={inline} updates={len(updates)}")
        except Exception as exc:
            debug_log(f"子智能体状态检查失败: {exc}")
            return
        for update in updates:
            message = update.get("system_message")
            if not message:
                continue
            task_id = update.get("task_id")
            debug_log(f"[SubAgent] update task={task_id} inline={inline} msg={message}")
            web_terminal._record_sub_agent_message(message, task_id, inline=inline)
            debug_log(f"[SubAgent] recorded task={task_id}, 计算插入位置")

            insert_index = len(messages)
            if after_tool_call_id:
                for idx, msg in enumerate(messages):
                    if msg.get("role") == "tool" and msg.get("tool_call_id") == after_tool_call_id:
                        insert_index = idx + 1
                        break

            messages.insert(insert_index, {
                "role": "system",
                "content": message,
                "metadata": {"sub_agent_notice": True, "inline": inline, "task_id": task_id}
            })
            debug_log(f"[SubAgent] 插入系统消息位置: {insert_index}")
            sender('system_message', {
                'content': message,
                'inline': inline
            })
            maybe_mark_failure_from_message(web_terminal, message)

    async def _wait_retry_delay(delay_seconds: int) -> bool:
        """等待重试间隔，同时检查是否收到停止请求。"""
        if delay_seconds <= 0:
            return False
        deadline = time.time() + delay_seconds
        while time.time() < deadline:
            client_stop_info = get_stop_flag(client_sid, username)
            if client_stop_info:
                stop_requested = client_stop_info.get('stop', False) if isinstance(client_stop_info, dict) else client_stop_info
                if stop_requested:
                    sender('task_stopped', {
                        'message': '命令执行被用户取消',
                        'reason': 'user_stop'
                    })
                    clear_stop_flag(client_sid, username)
                    return True
            await asyncio.sleep(0.2)
        return False
    
    iteration = 0
    while max_iterations is None or iteration < max_iterations:
        current_iteration = iteration + 1
        iteration += 1
        total_iterations += 1
        iteration_limit_label = max_iterations if max_iterations is not None else "∞"
        debug_log(f"\n--- 迭代 {current_iteration}/{iteration_limit_label} 开始 ---")
        
        # 检查是否超过总工具调用限制
        if MAX_TOTAL_TOOL_CALLS is not None and total_tool_calls >= MAX_TOTAL_TOOL_CALLS:
            debug_log(f"已达到最大工具调用次数限制 ({MAX_TOTAL_TOOL_CALLS})")
            sender('system_message', {
                'content': f'⚠️ 已达到最大工具调用次数限制 ({MAX_TOTAL_TOOL_CALLS})，任务结束。'
            })
            mark_force_thinking(web_terminal, reason="tool_limit")
            break
        
        apply_thinking_schedule(web_terminal)

        full_response = ""
        tool_calls = []
        current_thinking = ""
        detected_tools = {}
        last_usage_payload = None
        
        # 状态标志
        in_thinking = False
        thinking_started = False
        thinking_ended = False
        text_started = False
        text_has_content = False
        text_streaming = False
        text_chunk_index = 0
        last_text_chunk_time: Optional[float] = None
        
        # 计数器
        chunk_count = 0
        reasoning_chunks = 0
        content_chunks = 0
        tool_chunks = 0
        append_break_triggered = False
        append_result = {"handled": False}
        modify_break_triggered = False
        modify_result = {"handled": False}
        last_finish_reason = None

        def _cancel_pending_tools(tool_calls_list):
            """为尚未返回结果的工具生成取消结果，防止缺失 tool_call_id 造成后续 400。"""
            if not tool_calls_list:
                return
            for tc in tool_calls_list:
                tc_id = tc.get("id")
                func_name = tc.get("function", {}).get("name")
                sender('update_action', {
                    'preparing_id': tc_id,
                    'status': 'cancelled',
                    'result': {
                        "success": False,
                        "status": "cancelled",
                        "message": "命令执行被用户取消",
                        "tool": func_name
                    }
                })
                if tc_id:
                    messages.append({
                        "role": "tool",
                        "tool_call_id": tc_id,
                        "name": func_name,
                        "content": "命令执行被用户取消",
                        "metadata": {"status": "cancelled"}
                    })
        
        thinking_expected = web_terminal.api_client.get_current_thinking_mode()
        debug_log(f"思考模式: {thinking_expected}")
        quota_allowed = True
        quota_info = {}
        if hasattr(web_terminal, "record_model_call"):
            quota_allowed, quota_info = web_terminal.record_model_call(bool(thinking_expected))
        if not quota_allowed:
            quota_type = 'thinking' if thinking_expected else 'fast'
            socketio.emit('quota_notice', {
                'type': quota_type,
                'reset_at': quota_info.get('reset_at'),
                'limit': quota_info.get('limit'),
                'count': quota_info.get('count')
            }, room=f"user_{getattr(web_terminal, 'username', '')}")
            sender('quota_exceeded', {
                'type': quota_type,
                'reset_at': quota_info.get('reset_at')
            })
            sender('error', {
                'message': "配额已达到上限，暂时无法继续调用模型。",
                'quota': quota_info
            })
            return
        
        tool_call_limit_label = MAX_TOTAL_TOOL_CALLS if MAX_TOTAL_TOOL_CALLS is not None else "∞"
        print(f"[API] 第{current_iteration}次调用 (总工具调用: {total_tool_calls}/{tool_call_limit_label})")
        
        api_error = None
        for api_attempt in range(max_api_retries + 1):
            api_error = None
            if api_attempt > 0:
                full_response = ""
                tool_calls = []
                current_thinking = ""
                detected_tools = {}
                last_usage_payload = None
                in_thinking = False
                thinking_started = False
                thinking_ended = False
                text_started = False
                text_has_content = False
                text_streaming = False
                text_chunk_index = 0
                last_text_chunk_time = None
                chunk_count = 0
                reasoning_chunks = 0
                content_chunks = 0
                tool_chunks = 0
                append_break_triggered = False
                append_result = {"handled": False}
                modify_break_triggered = False
                modify_result = {"handled": False}
                last_finish_reason = None

            # 收集流式响应
            async for chunk in web_terminal.api_client.chat(messages, tools, stream=True):
                chunk_count += 1
            
                # 检查停止标志
                client_stop_info = get_stop_flag(client_sid, username)
                if client_stop_info:
                    stop_requested = client_stop_info.get('stop', False) if isinstance(client_stop_info, dict) else client_stop_info
                    if stop_requested:
                        debug_log(f"检测到停止请求，中断流处理")
                        if pending_append:
                            append_result = await finalize_pending_append(full_response, False, finish_reason="user_stop")
                        if pending_modify:
                            modify_result = await finalize_pending_modify(full_response, False, finish_reason="user_stop")
                        _cancel_pending_tools(tool_calls)
                        sender('task_stopped', {
                            'message': '命令执行被用户取消',
                            'reason': 'user_stop'
                        })
                        clear_stop_flag(client_sid, username)
                        return
            
                if isinstance(chunk, dict) and chunk.get("error"):
                    api_error = chunk.get("error")
                    break

                # 先尝试记录 usage（有些平台会在最后一个 chunk 里携带 usage 但 choices 为空）
                usage_info = chunk.get("usage")
                if usage_info:
                    last_usage_payload = usage_info

                if "choices" not in chunk:
                    debug_log(f"Chunk {chunk_count}: 无choices字段")
                    continue
                if not chunk.get("choices"):
                    debug_log(f"Chunk {chunk_count}: choices为空列表")
                    continue
                choice = chunk["choices"][0]
                if not usage_info and isinstance(choice, dict) and choice.get("usage"):
                    # 兼容部分供应商将 usage 放在 choice 内的格式（例如部分 Kimi/Qwen 返回）
                    last_usage_payload = choice.get("usage")
                delta = choice.get("delta", {})
                finish_reason = choice.get("finish_reason")
                if finish_reason:
                    last_finish_reason = finish_reason
            
                # 处理思考内容（兼容 reasoning_content / reasoning_details）
                reasoning_content = ""
                if "reasoning_content" in delta:
                    reasoning_content = delta.get("reasoning_content") or ""
                elif "reasoning_details" in delta:
                    details = delta.get("reasoning_details")
                    if isinstance(details, list):
                        parts = []
                        for item in details:
                            if isinstance(item, dict):
                                text = item.get("text")
                                if text:
                                    parts.append(text)
                        if parts:
                            reasoning_content = "".join(parts)
                if reasoning_content:
                    reasoning_chunks += 1
                    debug_log(f"  思考内容 #{reasoning_chunks}: {len(reasoning_content)} 字符")
                
                    if not thinking_started:
                        in_thinking = True
                        thinking_started = True
                        sender('thinking_start', {})
                        await asyncio.sleep(0.05)
                
                    current_thinking += reasoning_content
                    sender('thinking_chunk', {'content': reasoning_content})
            
                # 处理正常内容
                if "content" in delta:
                    content = delta["content"]
                    if content:
                        content_chunks += 1
                        debug_log(f"  正式内容 #{content_chunks}: {repr(content[:100] if content else 'None')}")
                    
                        if in_thinking and not thinking_ended:
                            in_thinking = False
                            thinking_ended = True
                            sender('thinking_end', {'full_content': current_thinking})
                            await asyncio.sleep(0.1)
                        
                    
                        expecting_modify = bool(pending_modify) or bool(getattr(web_terminal, "pending_modify_request", None))
                        expecting_append = bool(pending_append) or bool(getattr(web_terminal, "pending_append_request", None))

                        if pending_modify:
                            if not pending_modify.get("start_seen"):
                                probe_buffer = pending_modify.get("probe_buffer", "") + content
                                if len(probe_buffer) > 10000:
                                    probe_buffer = probe_buffer[-10000:]
                                marker = pending_modify.get("start_marker")
                                marker_index = probe_buffer.find(marker)
                                if marker_index == -1:
                                    pending_modify["probe_buffer"] = probe_buffer
                                    continue
                                after_marker = marker_index + len(marker)
                                remainder = probe_buffer[after_marker:]
                                pending_modify["buffer"] = remainder
                                pending_modify["raw_buffer"] = marker + remainder
                                pending_modify["start_seen"] = True
                                pending_modify["detected_blocks"] = set()
                                pending_modify["probe_buffer"] = ""
                                if pending_modify.get("display_id"):
                                    sender('update_action', {
                                        'id': pending_modify["display_id"],
                                        'status': 'running',
                                        'preparing_id': pending_modify.get("tool_call_id"),
                                        'message': f"正在修改 {pending_modify['path']}..."
                                    })
                            else:
                                pending_modify["buffer"] += content
                                pending_modify["raw_buffer"] += content
                        
                            if pending_modify.get("start_seen"):
                                block_text = pending_modify["buffer"]
                                for match in re.finditer(r"\[replace:(\d+)\]", block_text):
                                    try:
                                        block_index = int(match.group(1))
                                    except ValueError:
                                        continue
                                    detected_blocks = pending_modify.setdefault("detected_blocks", set())
                                    if block_index not in detected_blocks:
                                        detected_blocks.add(block_index)
                                        if pending_modify.get("display_id"):
                                            sender('update_action', {
                                                'id': pending_modify["display_id"],
                                                'status': 'running',
                                                'preparing_id': pending_modify.get("tool_call_id"),
                                                'message': f"正在对 {pending_modify['path']} 进行第 {block_index} 处修改..."
                                            })
                        
                            if pending_modify.get("start_seen"):
                                end_pos = pending_modify["buffer"].find(pending_modify["end_marker"])
                                if end_pos != -1:
                                    pending_modify["end_index"] = end_pos
                                    modify_break_triggered = True
                                    debug_log("检测到<<<END_MODIFY>>>，即将终止流式输出并应用修改")
                                    break
                            continue
                        elif expecting_modify:
                            modify_probe_buffer += content
                            if len(modify_probe_buffer) > 10000:
                                modify_probe_buffer = modify_probe_buffer[-10000:]
                        
                            marker_match = re.search(r"<<<MODIFY:\s*([\s\S]*?)>>>", modify_probe_buffer)
                            if marker_match:
                                detected_raw_path = marker_match.group(1)
                                detected_path = detected_raw_path.strip()
                                marker_full = marker_match.group(0)
                                after_marker_index = modify_probe_buffer.find(marker_full) + len(marker_full)
                                remainder = modify_probe_buffer[after_marker_index:]
                                modify_probe_buffer = ""
                            
                                if not detected_path:
                                    debug_log("检测到 MODIFY 起始标记但路径为空，忽略。")
                                    continue
                            
                                pending_modify = {
                                    "path": detected_path,
                                    "tool_call_id": None,
                                    "buffer": remainder,
                                    "raw_buffer": marker_full + remainder,
                                    "start_marker": marker_full,
                                    "end_marker": "<<<END_MODIFY>>>",
                                    "start_seen": True,
                                    "end_index": None,
                                    "display_id": None,
                                    "detected_blocks": set()
                                }
                                if hasattr(web_terminal, "pending_modify_request"):
                                    web_terminal.pending_modify_request = {"path": detected_path}
                                debug_log(f"直接检测到modify起始标记，构建修改缓冲: {detected_path}")
                            
                                end_pos = pending_modify["buffer"].find(pending_modify["end_marker"])
                                if end_pos != -1:
                                    pending_modify["end_index"] = end_pos
                                    modify_break_triggered = True
                                    debug_log("检测到<<<END_MODIFY>>>，即将终止流式输出并应用修改")
                                    break
                            continue
                    
                        if pending_append:
                            pending_append["buffer"] += content
                        
                            if pending_append.get("content_start") is None:
                                marker_index = pending_append["buffer"].find(pending_append["start_marker"])
                                if marker_index != -1:
                                    pending_append["content_start"] = marker_index + len(pending_append["start_marker"])
                                    debug_log(f"检测到追加起始标识: {pending_append['start_marker']}")
                        
                            if pending_append.get("content_start") is not None:
                                end_index = pending_append["buffer"].find(
                                    pending_append["end_marker"],
                                    pending_append["content_start"]
                                )
                                if end_index != -1:
                                    pending_append["end_index"] = end_index
                                    append_break_triggered = True
                                    debug_log("检测到<<<END_APPEND>>>，即将终止流式输出并写入文件")
                                    break
                        
                            # 继续累积追加内容
                            continue
                        elif expecting_append:
                            append_probe_buffer += content
                            # 限制缓冲区大小防止过长
                            if len(append_probe_buffer) > 10000:
                                append_probe_buffer = append_probe_buffer[-10000:]

                            marker_match = re.search(r"<<<APPEND:\s*([\s\S]*?)>>>", append_probe_buffer)
                            if marker_match:
                                detected_raw_path = marker_match.group(1)
                                detected_path = detected_raw_path.strip()
                                if not detected_path:
                                    append_probe_buffer = append_probe_buffer[marker_match.end():]
                                    continue
                                marker_full = marker_match.group(0)
                                after_marker_index = append_probe_buffer.find(marker_full) + len(marker_full)
                                remainder = append_probe_buffer[after_marker_index:]
                                append_probe_buffer = ""
                                pending_append = {
                                    "path": detected_path,
                                    "tool_call_id": None,
                                    "buffer": remainder,
                                    "start_marker": marker_full,
                                    "end_marker": "<<<END_APPEND>>>",
                                    "content_start": 0,
                                    "end_index": None,
                                    "display_id": None
                                }
                                if hasattr(web_terminal, "pending_append_request"):
                                    web_terminal.pending_append_request = {"path": detected_path}
                                debug_log(f"直接检测到append起始标记，构建追加缓冲: {detected_path}")
                                # 检查是否立即包含结束标记
                                if pending_append["buffer"]:
                                    end_index = pending_append["buffer"].find(pending_append["end_marker"], pending_append["content_start"])
                                    if end_index != -1:
                                        pending_append["end_index"] = end_index
                                        append_break_triggered = True
                                        debug_log("检测到<<<END_APPEND>>>，即将终止流式输出并写入文件")
                                        break
                                continue
                    
                        if not text_started:
                            text_started = True
                            text_streaming = True
                            sender('text_start', {})
                            brief_log("模型输出了内容")
                            await asyncio.sleep(0.05)
                    
                        if not pending_append:
                            full_response += content
                            accumulated_response += content
                            text_has_content = True
                            emit_time = time.time()
                            elapsed = 0.0 if last_text_chunk_time is None else emit_time - last_text_chunk_time
                            last_text_chunk_time = emit_time
                            text_chunk_index += 1
                            log_backend_chunk(
                                conversation_id,
                                current_iteration,
                                text_chunk_index,
                                elapsed,
                                len(content),
                                content[:32]
                            )
                            sender('text_chunk', {
                                'content': content,
                                'index': text_chunk_index,
                                'elapsed': elapsed
                            })
            
                # 收集工具调用 - 实时发送准备状态
                if "tool_calls" in delta:
                    tool_chunks += 1
                    for tc in delta["tool_calls"]:
                        found = False
                        for existing in tool_calls:
                            if existing.get("index") == tc.get("index"):
                                if "function" in tc and "arguments" in tc["function"]:
                                    arg_chunk = tc["function"]["arguments"]
                                    existing_fn = existing.get("function", {})
                                    existing_args = existing_fn.get("arguments", "")
                                    existing_fn["arguments"] = (existing_args or "") + arg_chunk
                                    existing["function"] = existing_fn

                                    combined_args = existing_fn.get("arguments", "")
                                    tool_id = existing.get("id") or tc.get("id")
                                    tool_name = (
                                        existing_fn.get("name")
                                        or tc.get("function", {}).get("name", "")
                                    )
                                    intent_value = extract_intent_from_partial(combined_args)
                                    if (
                                        intent_value
                                        and tool_id
                                        and detected_tool_intent.get(tool_id) != intent_value
                                    ):
                                        detected_tool_intent[tool_id] = intent_value
                                        brief_log(f"[intent] 增量提取 {tool_name}: {intent_value}")
                                        sender('tool_intent', {
                                            'id': tool_id,
                                            'name': tool_name,
                                            'intent': intent_value,
                                            'conversation_id': conversation_id
                                        })
                                        debug_log(f"    发送工具意图: {tool_name} -> {intent_value}")
                                        await asyncio.sleep(0.01)
                                found = True
                                break
                    
                        if not found and tc.get("id"):
                            tool_id = tc["id"]
                            tool_name = tc.get("function", {}).get("name", "")
                            arguments_str = tc.get("function", {}).get("arguments", "") or ""
                        
                            # 新工具检测到，立即发送准备事件
                            if tool_id not in detected_tools and tool_name:
                                detected_tools[tool_id] = tool_name
                            
                                # 尝试提前提取 intent
                                intent_value = None
                                if arguments_str:
                                    intent_value = extract_intent_from_partial(arguments_str)
                                    if intent_value:
                                        detected_tool_intent[tool_id] = intent_value
                                        brief_log(f"[intent] 预提取 {tool_name}: {intent_value}")
                            
                                # 立即发送工具准备中事件
                                brief_log(f"[tool] 准备调用 {tool_name} (id={tool_id}) intent={intent_value or '-'}")
                                sender('tool_preparing', {
                                    'id': tool_id,
                                    'name': tool_name,
                                    'message': f'准备调用 {tool_name}...',
                                    'intent': intent_value,
                                    'conversation_id': conversation_id
                                })
                                debug_log(f"    发送工具准备事件: {tool_name}")
                                await asyncio.sleep(0.1)
                        
                            tool_calls.append({
                                "id": tool_id,
                                "index": tc.get("index"),
                                "type": "function",
                                "function": {
                                    "name": tool_name,
                                    "arguments": arguments_str
                                }
                            })
                            # 尝试从增量参数中抽取 intent，并单独推送
                            if tool_id and arguments_str:
                                intent_value = extract_intent_from_partial(arguments_str)
                                if intent_value and detected_tool_intent.get(tool_id) != intent_value:
                                    detected_tool_intent[tool_id] = intent_value
                                    sender('tool_intent', {
                                        'id': tool_id,
                                        'name': tool_name,
                                        'intent': intent_value,
                                        'conversation_id': conversation_id
                                    })
                                    debug_log(f"    发送工具意图: {tool_name} -> {intent_value}")
                                    await asyncio.sleep(0.01)
                            debug_log(f"    新工具: {tool_name}")
        
            # 检查是否被停止
            client_stop_info = get_stop_flag(client_sid, username)
            if client_stop_info:
                stop_requested = client_stop_info.get('stop', False) if isinstance(client_stop_info, dict) else client_stop_info
                if stop_requested:
                    debug_log("任务在流处理完成后检测到停止状态")
                    sender('task_stopped', {
                        'message': '命令执行被用户取消',
                        'reason': 'user_stop'
                    })
                    _cancel_pending_tools(tool_calls)
                    clear_stop_flag(client_sid, username)
                    return
        
            # === API响应完成后只计算输出token ===
            if last_usage_payload:
                try:
                    web_terminal.context_manager.apply_usage_statistics(last_usage_payload)
                    debug_log(
                        f"Usage统计: prompt={last_usage_payload.get('prompt_tokens', 0)}, "
                        f"completion={last_usage_payload.get('completion_tokens', 0)}, "
                        f"total={last_usage_payload.get('total_tokens', 0)}"
                    )
                except Exception as e:
                    debug_log(f"Usage统计更新失败: {e}")
            else:
                debug_log("未获取到usage字段，跳过token统计更新")
        

            if api_error:
                try:
                    debug_log(f"API错误原始数据: {json.dumps(api_error, ensure_ascii=False)}")
                except Exception:
                    debug_log(f"API错误原始数据(不可序列化): {repr(api_error)}")
                error_message = ""
                error_status = None
                error_type = None
                error_code = None
                error_text = ""
                request_dump = None
                error_base_url = None
                error_model_id = None
                if isinstance(api_error, dict):
                    error_status = api_error.get("status_code")
                    error_type = api_error.get("error_type") or api_error.get("type")
                    error_code = api_error.get("error_code") or api_error.get("code")
                    error_text = api_error.get("error_text") or ""
                    error_message = (
                        api_error.get("error_message")
                        or api_error.get("message")
                        or error_text
                        or ""
                    )
                    request_dump = api_error.get("request_dump")
                    error_base_url = api_error.get("base_url")
                    error_model_id = api_error.get("model_id")
                elif isinstance(api_error, str):
                    error_message = api_error
                if not error_message:
                    if error_status:
                        error_message = f"API 请求失败（HTTP {error_status}）"
                    else:
                        error_message = "API 请求失败"
                # 若命中阿里云配额错误，立即写入状态并切换到官方 API
                try:
                    from utils.aliyun_fallback import compute_disabled_until, set_disabled_until
                    disabled_until, reason = compute_disabled_until(error_message)
                    if disabled_until and reason:
                        set_disabled_until(getattr(web_terminal, "model_key", None) or "kimi-k2.5", disabled_until, reason)
                        profile = get_model_profile(getattr(web_terminal, "model_key", None) or "kimi-k2.5")
                        web_terminal.apply_model_profile(profile)
                except Exception as exc:
                    debug_log(f"处理阿里云配额回退失败: {exc}")
                can_retry = (
                    api_attempt < max_api_retries
                    and not full_response
                    and not tool_calls
                    and not current_thinking
                    and not pending_append
                    and not pending_modify
                )
                sender('error', {
                    'message': error_message,
                    'status_code': error_status,
                    'error_type': error_type,
                    'error_code': error_code,
                    'error_text': error_text,
                    'request_dump': request_dump,
                    'base_url': error_base_url,
                    'model_id': error_model_id,
                    'retry': bool(can_retry),
                    'retry_in': retry_delay_seconds if can_retry else None,
                    'attempt': api_attempt + 1,
                    'max_attempts': max_api_retries + 1
                })
                if can_retry:
                    try:
                        profile = get_model_profile(getattr(web_terminal, "model_key", None) or "kimi-k2.5")
                        web_terminal.apply_model_profile(profile)
                    except Exception as exc:
                        debug_log(f"重试前更新模型配置失败: {exc}")
                    cancelled = await _wait_retry_delay(retry_delay_seconds)
                    if cancelled:
                        return
                    continue
                _cancel_pending_tools(tool_calls)
                return
            break

        # 流结束后的处理
        debug_log(f"\n流结束统计:")
        debug_log(f"  总chunks: {chunk_count}")
        debug_log(f"  思考chunks: {reasoning_chunks}")
        debug_log(f"  内容chunks: {content_chunks}")
        debug_log(f"  工具chunks: {tool_chunks}")
        debug_log(f"  收集到的思考: {len(current_thinking)} 字符")
        debug_log(f"  收集到的正文: {len(full_response)} 字符")
        debug_log(f"  收集到的工具: {len(tool_calls)} 个")
        
        if not append_result["handled"] and pending_append:
            append_result = await finalize_pending_append(full_response, True, finish_reason=last_finish_reason)
        if not modify_result["handled"] and pending_modify:
            modify_result = await finalize_pending_modify(full_response, True, finish_reason=last_finish_reason)
        
        # 结束未完成的流
        if in_thinking and not thinking_ended:
            sender('thinking_end', {'full_content': current_thinking})
            await asyncio.sleep(0.1)
            
        
        # 确保text_end事件被发送
        if text_started and text_has_content and not append_result["handled"] and not modify_result["handled"]:
            debug_log(f"发送text_end事件，完整内容长度: {len(full_response)}")
            sender('text_end', {'full_content': full_response})
            await asyncio.sleep(0.1)
            text_streaming = False
            
            if full_response.strip():
                debug_log(f"流式文本内容长度: {len(full_response)} 字符")
        
        if append_result["handled"]:
            append_metadata = append_result.get("assistant_metadata")
            append_content_text = append_result.get("assistant_content")
            if append_content_text:
                web_terminal.context_manager.add_conversation(
                    "assistant",
                    append_content_text,
                    metadata=append_metadata
                )
                debug_log("💾 增量保存：追加正文快照")
                
                payload_info = append_metadata.get("append_payload") if append_metadata else {}
                sender('append_payload', {
                    'path': payload_info.get("path") or append_result.get("path"),
                    'forced': payload_info.get("forced", False),
                    'lines': payload_info.get("lines"),
                    'bytes': payload_info.get("bytes"),
                    'tool_call_id': payload_info.get("tool_call_id") or append_result.get("tool_call_id"),
                    'success': payload_info.get("success", append_result.get("success", False)),
                    'conversation_id': conversation_id
                })
            
            if append_result["tool_content"]:
                tool_call_id = append_result.get("tool_call_id") or f"append_{int(time.time() * 1000)}"
                system_notice = format_tool_result_notice("append_to_file", tool_call_id, append_result["tool_content"])
                web_terminal.context_manager.add_conversation("system", system_notice)
                append_result["tool_call_id"] = tool_call_id
                debug_log("💾 增量保存：append_to_file 工具结果（system 通知）")
            
            finish_reason = append_result.get("finish_reason")
            path_for_prompt = append_result.get("path")
            need_follow_prompt = (
                finish_reason == "length" or
                append_result.get("forced") or
                not append_result.get("success")
            )

            if need_follow_prompt and path_for_prompt:
                prompt_lines = [
                    f"append_to_file 在处理 {path_for_prompt} 时未完成，需要重新发起写入。"
                ]
                if finish_reason == "length":
                    prompt_lines.append(
                        "上一次输出达到系统单次输出上限，已写入的内容已保存。"
                    )
                if append_result.get("forced"):
                    prompt_lines.append(
                        "收到的内容缺少 <<<END_APPEND>>> 标记，系统依据流式结束位置落盘。"
                    )
                if not append_result.get("success"):
                    prompt_lines.append("系统未能识别有效的追加标记。")
                prompt_lines.append(
                    "请再次调用 append_to_file 工具获取新的写入窗口，并在工具调用的输出中遵循以下格式："
                )
                prompt_lines.append(f"<<<APPEND:{path_for_prompt}>>>")
                prompt_lines.append("...填写剩余正文，如内容已完成可留空...")
                prompt_lines.append("<<<END_APPEND>>>")
                prompt_lines.append("不要在普通回复中粘贴上述标记，必须通过 append_to_file 工具发送。")
                follow_prompt = "\n".join(prompt_lines)
                messages.append({
                    "role": "system",
                    "content": follow_prompt
                })
                web_terminal.context_manager.add_conversation("system", follow_prompt)
                debug_log("已注入追加任务提示")

        if append_result["handled"] and append_result.get("forced") and append_result.get("success"):
            mark_force_thinking(web_terminal, reason="append_forced_finish")
        if append_result["handled"] and not append_result.get("success"):
            sender('system_message', {
                'content': f'⚠️ 追加写入失败：{append_result.get("error")}'
            })
            maybe_mark_failure_from_message(web_terminal, f'⚠️ 追加写入失败：{append_result.get("error")}')
            mark_force_thinking(web_terminal, reason="append_failed")
        
        if modify_result["handled"]:
            modify_metadata = modify_result.get("assistant_metadata")
            modify_content_text = modify_result.get("assistant_content")
            if modify_content_text:
                web_terminal.context_manager.add_conversation(
                    "assistant",
                    modify_content_text,
                    metadata=modify_metadata
                )
                debug_log("💾 增量保存：修改正文快照")
                
                payload_info = modify_metadata.get("modify_payload") if modify_metadata else {}
                sender('modify_payload', {
                    'path': payload_info.get("path") or modify_result.get("path"),
                    'total': payload_info.get("total_blocks") or modify_result.get("total_blocks"),
                    'completed': payload_info.get("completed") or modify_result.get("completed_blocks"),
                    'failed': payload_info.get("failed") or modify_result.get("failed_blocks"),
                    'forced': payload_info.get("forced", modify_result.get("forced", False)),
                    'success': modify_result.get("success", False),
                    'conversation_id': conversation_id
                })
            
            if modify_result["tool_content"]:
                tool_call_id = modify_result.get("tool_call_id") or f"modify_{int(time.time() * 1000)}"
                system_notice = format_tool_result_notice("modify_file", tool_call_id, modify_result["tool_content"])
                web_terminal.context_manager.add_conversation("system", system_notice)
                modify_result["tool_call_id"] = tool_call_id
                debug_log("💾 增量保存：modify_file 工具结果（system 通知）")
            
            path_for_prompt = modify_result.get("path")
            failed_blocks = modify_result.get("failed_blocks") or []
            need_follow_prompt = modify_result.get("forced") or bool(failed_blocks)
            
            if need_follow_prompt and path_for_prompt:
                prompt_lines = [
                    f"modify_file 在处理 {path_for_prompt} 时未完成，需要重新发起补丁。"
                ]
                if modify_result.get("forced"):
                    prompt_lines.append(
                        "刚才的内容缺少 <<<END_MODIFY>>> 标记，系统仅应用了已识别的部分。"
                    )
                if failed_blocks:
                    failed_text = "、".join(str(idx) for idx in failed_blocks)
                    prompt_lines.append(f"以下补丁未成功：第 {failed_text} 处。")
                prompt_lines.append(
                    "请再次调用 modify_file 工具，并在新的工具调用中按以下模板提供完整补丁："
                )
                prompt_lines.append(f"<<<MODIFY:{path_for_prompt}>>>")
                prompt_lines.append("[replace:序号]")
                prompt_lines.append("<<OLD>>")
                prompt_lines.append("...原文（必须逐字匹配，包含全部缩进、空格和换行）...")
                prompt_lines.append("<<END>>")
                prompt_lines.append("<<NEW>>")
                prompt_lines.append("...新内容，可留空表示清空，注意保持结构完整...")
                prompt_lines.append("<<END>>")
                prompt_lines.append("[/replace]")
                prompt_lines.append("<<<END_MODIFY>>>")
                prompt_lines.append("请勿在普通回复中直接粘贴补丁，必须通过 modify_file 工具发送。")
                follow_prompt = "\n".join(prompt_lines)
                messages.append({
                    "role": "system",
                    "content": follow_prompt
                })
                web_terminal.context_manager.add_conversation("system", follow_prompt)
                debug_log("已注入修改任务提示")

            if modify_result["handled"] and modify_result.get("failed_blocks"):
                mark_force_thinking(web_terminal, reason="modify_partial_failure")
            if modify_result["handled"] and modify_result.get("forced") and modify_result.get("success"):
                mark_force_thinking(web_terminal, reason="modify_forced_finish")
            if modify_result["handled"] and not modify_result.get("success"):
                error_message = modify_result.get("summary_message") or modify_result.get("error") or "修改操作未成功，请根据提示重新执行。"
                sender('system_message', {
                    'content': f'⚠️ 修改操作存在未完成的内容：{error_message}'
                })
                maybe_mark_failure_from_message(web_terminal, f'⚠️ 修改操作存在未完成的内容：{error_message}')
                mark_force_thinking(web_terminal, reason="modify_failed")
            
        if web_terminal.api_client.last_call_used_thinking and current_thinking:
            web_terminal.api_client.current_task_thinking = current_thinking or ""
        if web_terminal.api_client.current_task_first_call:
            web_terminal.api_client.current_task_first_call = False
        update_thinking_after_call(web_terminal)
        
        # 检测是否有格式错误的工具调用
        if not tool_calls and full_response and AUTO_FIX_TOOL_CALL and not append_result["handled"] and not modify_result["handled"]:
            if detect_malformed_tool_call(full_response):
                auto_fix_attempts += 1
                
                if auto_fix_attempts <= AUTO_FIX_MAX_ATTEMPTS:
                    debug_log(f"检测到格式错误的工具调用，尝试自动修复 (尝试 {auto_fix_attempts}/{AUTO_FIX_MAX_ATTEMPTS})")
                    
                    fix_message = "你使用了错误的格式输出工具调用。请使用正确的工具调用格式而不是直接输出JSON。根据当前进度继续执行任务。"
                    
                    sender('system_message', {
                        'content': f'⚠️ 自动修复: {fix_message}'
                    })
                    maybe_mark_failure_from_message(web_terminal, f'⚠️ 自动修复: {fix_message}')
                    
                    messages.append({
                        "role": "user",
                        "content": fix_message
                    })
                    
                    await asyncio.sleep(1)
                    continue
                else:
                    debug_log(f"自动修复尝试已达上限 ({AUTO_FIX_MAX_ATTEMPTS})")
                    sender('system_message', {
                        'content': f'⌘ 工具调用格式错误，自动修复失败。请手动检查并重试。'
                    })
                    maybe_mark_failure_from_message(web_terminal, '⌘ 工具调用格式错误，自动修复失败。请手动检查并重试。')
                    break
        
        # 构建助手消息（用于API继续对话）
        assistant_content_parts = []
        
        if full_response:
            assistant_content_parts.append(full_response)
        elif append_result["handled"] and append_result["assistant_content"]:
            assistant_content_parts.append(append_result["assistant_content"])
        elif modify_result["handled"] and modify_result.get("assistant_content"):
            assistant_content_parts.append(modify_result["assistant_content"])
        
        assistant_content = "\n".join(assistant_content_parts) if assistant_content_parts else ""
        
        # 添加到消息历史（用于API继续对话，不保存到文件）
        assistant_message = {
            "role": "assistant",
            "content": assistant_content,
            "tool_calls": tool_calls
        }
        if current_thinking:
            assistant_message["reasoning_content"] = current_thinking
        
        messages.append(assistant_message)
        if assistant_content or current_thinking or tool_calls:
            web_terminal.context_manager.add_conversation(
                "assistant",
                assistant_content,
                tool_calls=tool_calls if tool_calls else None,
                reasoning_content=current_thinking or None
            )
        
        # 为下一轮迭代重置流状态标志，但保留 full_response 供上面保存使用
        text_streaming = False
        text_started = False
        text_has_content = False
        full_response = ""
        
        if append_result["handled"] and append_result.get("tool_content"):
            tool_call_id = append_result.get("tool_call_id") or f"append_{int(time.time() * 1000)}"
            system_notice = format_tool_result_notice("append_to_file", tool_call_id, append_result["tool_content"])
            messages.append({
                "role": "system",
                "content": system_notice
            })
            append_result["tool_call_id"] = tool_call_id
            debug_log("已将 append_to_file 工具结果以 system 形式追加到对话上下文")
        if modify_result["handled"] and modify_result.get("tool_content"):
            tool_call_id = modify_result.get("tool_call_id") or f"modify_{int(time.time() * 1000)}"
            system_notice = format_tool_result_notice("modify_file", tool_call_id, modify_result["tool_content"])
            messages.append({
                "role": "system",
                "content": system_notice
            })
            modify_result["tool_call_id"] = tool_call_id
            debug_log("已将 modify_file 工具结果以 system 形式追加到对话上下文")
        
        force_continue = append_result["handled"] or modify_result["handled"]
        if force_continue:
            if append_result["handled"]:
                debug_log("append_to_file 已处理，继续下一轮以让模型返回确认回复")
            elif modify_result["handled"]:
                debug_log("modify_file 已处理，继续下一轮以让模型返回确认回复")
            else:
                debug_log("补丁处理完成，继续下一轮以获取模型回复")
            continue

        if not tool_calls:
            debug_log("没有工具调用，结束迭代")
            break
        
        # 检查连续相同工具调用
        for tc in tool_calls:
            tool_name = tc["function"]["name"]
            
            if tool_name == last_tool_name:
                consecutive_same_tool[tool_name] += 1
                
                if (
                    MAX_CONSECUTIVE_SAME_TOOL is not None
                    and consecutive_same_tool[tool_name] >= MAX_CONSECUTIVE_SAME_TOOL
                ):
                    debug_log(f"警告: 连续调用相同工具 {tool_name} 已达 {MAX_CONSECUTIVE_SAME_TOOL} 次")
                    sender('system_message', {
                        'content': f'⚠️ 检测到重复调用 {tool_name} 工具 {MAX_CONSECUTIVE_SAME_TOOL} 次，可能存在循环。'
                    })
                    maybe_mark_failure_from_message(web_terminal, f'⚠️ 检测到重复调用 {tool_name} 工具 {MAX_CONSECUTIVE_SAME_TOOL} 次，可能存在循环。')
                    
                    if consecutive_same_tool[tool_name] >= MAX_CONSECUTIVE_SAME_TOOL + 2:
                        debug_log(f"终止: 工具 {tool_name} 调用次数过多")
                        sender('system_message', {
                            'content': f'⌘ 工具 {tool_name} 重复调用过多，任务终止。'
                        })
                        maybe_mark_failure_from_message(web_terminal, f'⌘ 工具 {tool_name} 重复调用过多，任务终止。')
                        break
            else:
                consecutive_same_tool.clear()
                consecutive_same_tool[tool_name] = 1
            
            last_tool_name = tool_name
        # 更新统计
        total_tool_calls += len(tool_calls)

        # 执行每个工具
        for tool_call in tool_calls:
            # 检查停止标志
            client_stop_info = get_stop_flag(client_sid, username)
            if client_stop_info:
                stop_requested = client_stop_info.get('stop', False) if isinstance(client_stop_info, dict) else client_stop_info
                if stop_requested:
                    debug_log("在工具调用过程中检测到停止状态")
                    tool_call_id = tool_call.get("id")
                    function_name = tool_call.get("function", {}).get("name")
                    # 通知前端该工具已被取消，避免界面卡住
                    sender('update_action', {
                        'preparing_id': tool_call_id,
                        'status': 'cancelled',
                        'result': {
                            "success": False,
                            "status": "cancelled",
                            "message": "命令执行被用户取消",
                            "tool": function_name
                        }
                    })
                    # 在消息列表中记录取消结果，防止重新加载时仍显示运行中
                    if tool_call_id:
                        messages.append({
                            "role": "tool",
                            "tool_call_id": tool_call_id,
                            "name": function_name,
                            "content": "命令执行被用户取消",
                            "metadata": {"status": "cancelled"}
                        })
                    sender('task_stopped', {
                        'message': '命令执行被用户取消',
                        'reason': 'user_stop'
                    })
                    clear_stop_flag(client_sid, username)
                    return

            # 工具调用间隔控制
            current_time = time.time()
            if last_tool_call_time > 0:
                elapsed = current_time - last_tool_call_time
                if elapsed < TOOL_CALL_COOLDOWN:
                    await asyncio.sleep(TOOL_CALL_COOLDOWN - elapsed)
            last_tool_call_time = time.time()
            
            function_name = tool_call["function"]["name"]
            arguments_str = tool_call["function"]["arguments"]
            tool_call_id = tool_call["id"]
            

            debug_log(f"准备解析JSON，工具: {function_name}, 参数长度: {len(arguments_str)}")
            debug_log(f"JSON参数前200字符: {arguments_str[:200]}")
            debug_log(f"JSON参数后200字符: {arguments_str[-200:]}")
            
            # 使用改进的参数解析方法
            if hasattr(web_terminal, 'api_client') and hasattr(web_terminal.api_client, '_safe_tool_arguments_parse'):
                success, arguments, error_msg = web_terminal.api_client._safe_tool_arguments_parse(arguments_str, function_name)
                if not success:
                    debug_log(f"安全解析失败: {error_msg}")
                    error_text = f'工具参数解析失败: {error_msg}'
                    error_payload = {
                        "success": False,
                        "error": error_text,
                        "error_type": "parameter_format_error",
                        "tool_name": function_name,
                        "tool_call_id": tool_call_id,
                        "message": error_text
                    }
                    sender('error', {'message': error_text})
                    sender('update_action', {
                        'preparing_id': tool_call_id,
                        'status': 'completed',
                        'result': error_payload,
                        'message': error_text
                    })
                    error_content = json.dumps(error_payload, ensure_ascii=False)
                    web_terminal.context_manager.add_conversation(
                        "tool",
                        error_content,
                        tool_call_id=tool_call_id,
                        name=function_name
                    )
                    messages.append({
                        "role": "tool",
                        "tool_call_id": tool_call_id,
                        "name": function_name,
                        "content": error_content
                    })
                    continue
                debug_log(f"使用安全解析成功，参数键: {list(arguments.keys())}")
            else:
                # 回退到带有基本修复逻辑的解析
                try:
                    arguments = json.loads(arguments_str) if arguments_str.strip() else {}
                    debug_log(f"直接JSON解析成功，参数键: {list(arguments.keys())}")
                except json.JSONDecodeError as e:
                    debug_log(f"原始JSON解析失败: {e}")
                    # 尝试基本的JSON修复
                    repaired_str = arguments_str.strip()
                    repair_attempts = []
                    
                    # 修复1: 未闭合字符串
                    if repaired_str.count('"') % 2 == 1:
                        repaired_str += '"'
                        repair_attempts.append("添加闭合引号")
                    
                    # 修复2: 未闭合JSON对象
                    if repaired_str.startswith('{') and not repaired_str.rstrip().endswith('}'):
                        repaired_str = repaired_str.rstrip() + '}'
                        repair_attempts.append("添加闭合括号")
                    
                    # 修复3: 截断的JSON（移除不完整的最后一个键值对）
                    if not repair_attempts:  # 如果前面的修复都没用上
                        last_comma = repaired_str.rfind(',')
                        if last_comma > 0:
                            repaired_str = repaired_str[:last_comma] + '}'
                            repair_attempts.append("移除不完整的键值对")
                    
                    # 尝试解析修复后的JSON
                    try:
                        arguments = json.loads(repaired_str)
                        debug_log(f"JSON修复成功: {', '.join(repair_attempts)}")
                        debug_log(f"修复后参数键: {list(arguments.keys())}")
                    except json.JSONDecodeError as repair_error:
                        debug_log(f"JSON修复也失败: {repair_error}")
                        debug_log(f"修复尝试: {repair_attempts}")
                        debug_log(f"修复后内容前100字符: {repaired_str[:100]}")
                        error_text = f'工具参数解析失败: {e}'
                        error_payload = {
                            "success": False,
                            "error": error_text,
                            "error_type": "parameter_format_error",
                            "tool_name": function_name,
                            "tool_call_id": tool_call_id,
                            "message": error_text
                        }
                        sender('error', {'message': error_text})
                        sender('update_action', {
                            'preparing_id': tool_call_id,
                            'status': 'completed',
                            'result': error_payload,
                            'message': error_text
                        })
                        error_content = json.dumps(error_payload, ensure_ascii=False)
                        web_terminal.context_manager.add_conversation(
                            "tool",
                            error_content,
                            tool_call_id=tool_call_id,
                            name=function_name
                        )
                        messages.append({
                            "role": "tool",
                            "tool_call_id": tool_call_id,
                            "name": function_name,
                            "content": error_content
                        })
                        continue            
            
            debug_log(f"执行工具: {function_name} (ID: {tool_call_id})")
            
            # 发送工具开始事件
            tool_display_id = f"tool_{iteration}_{function_name}_{time.time()}"
            monitor_snapshot = None
            snapshot_path = None
            memory_snapshot_type = None
            if function_name in MONITOR_FILE_TOOLS:
                snapshot_path = resolve_monitor_path(arguments)
                monitor_snapshot = capture_monitor_snapshot(web_terminal.file_manager, snapshot_path, MONITOR_SNAPSHOT_CHAR_LIMIT, debug_log)
                if monitor_snapshot:
                    cache_monitor_snapshot(tool_display_id, 'before', monitor_snapshot)
            elif function_name in MONITOR_MEMORY_TOOLS:
                memory_snapshot_type = (arguments.get('memory_type') or 'main').lower()
                before_entries = None
                try:
                    before_entries = resolve_monitor_memory(web_terminal.memory_manager._read_entries(memory_snapshot_type), MONITOR_MEMORY_ENTRY_LIMIT)
                except Exception as exc:
                    debug_log(f"[MonitorSnapshot] 读取记忆失败: {memory_snapshot_type} ({exc})")
                if before_entries is not None:
                    monitor_snapshot = {
                        'memory_type': memory_snapshot_type,
                        'entries': before_entries
                    }
                    cache_monitor_snapshot(tool_display_id, 'before', monitor_snapshot)
            
            sender('tool_start', {
                'id': tool_display_id,
                'name': function_name,
                'arguments': arguments,
                'preparing_id': tool_call_id,
                'monitor_snapshot': monitor_snapshot,
                'conversation_id': conversation_id
            })
            brief_log(f"调用了工具: {function_name}")
            
            await asyncio.sleep(0.3)
            start_time = time.time()
            
            # 执行工具
            tool_result = await web_terminal.handle_tool_call(function_name, arguments)
            debug_log(f"工具结果: {tool_result[:200]}...")
            
            execution_time = time.time() - start_time
            if execution_time < 1.5:
                await asyncio.sleep(1.5 - execution_time)
            
            # 更新工具状态
            result_data = {}
            try:
                result_data = json.loads(tool_result)
            except:
                result_data = {'output': tool_result}
            tool_failed = detect_tool_failure(result_data)

            action_status = 'completed'
            action_message = None
            awaiting_flag = False
            
            if function_name in {"write_file", "edit_file"}:
                diff_path = result_data.get("path") or arguments.get("file_path")
                summary = result_data.get("summary") or result_data.get("message")
                if summary:
                    action_message = summary
                debug_log(f"{function_name} 执行完成: {summary or '无摘要'}")
            
            if function_name == "wait_sub_agent":
                system_msg = result_data.get("system_message")
                if system_msg:
                    messages.append({
                        "role": "system",
                        "content": system_msg
                    })
                    sender('system_message', {
                        'content': system_msg,
                        'inline': False
                    })
                    maybe_mark_failure_from_message(web_terminal, system_msg)
            monitor_snapshot_after = None
            if function_name in MONITOR_FILE_TOOLS:
                result_path = None
                if isinstance(result_data, dict):
                    result_path = resolve_monitor_path(result_data)
                    if not result_path:
                        candidate_path = result_data.get('path')
                        if isinstance(candidate_path, str) and candidate_path.strip():
                            result_path = candidate_path.strip()
                if not result_path:
                    result_path = resolve_monitor_path(arguments, snapshot_path) or snapshot_path
                monitor_snapshot_after = capture_monitor_snapshot(web_terminal.file_manager, result_path, MONITOR_SNAPSHOT_CHAR_LIMIT, debug_log)
            elif function_name in MONITOR_MEMORY_TOOLS:
                memory_after_type = str(
                    arguments.get('memory_type')
                    or (isinstance(result_data, dict) and result_data.get('memory_type'))
                    or memory_snapshot_type
                    or 'main'
                ).lower()
                after_entries = None
                try:
                    after_entries = resolve_monitor_memory(web_terminal.memory_manager._read_entries(memory_after_type), MONITOR_MEMORY_ENTRY_LIMIT)
                except Exception as exc:
                    debug_log(f"[MonitorSnapshot] 读取记忆失败(after): {memory_after_type} ({exc})")
                if after_entries is not None:
                    monitor_snapshot_after = {
                        'memory_type': memory_after_type,
                        'entries': after_entries
                    }
            
            update_payload = {
                'id': tool_display_id,
                'status': action_status,
                'result': result_data,
                'preparing_id': tool_call_id,
                'conversation_id': conversation_id
            }
            if action_message:
                update_payload['message'] = action_message
            if awaiting_flag:
                update_payload['awaiting_content'] = True
            if monitor_snapshot_after:
                update_payload['monitor_snapshot_after'] = monitor_snapshot_after
                cache_monitor_snapshot(tool_display_id, 'after', monitor_snapshot_after)
            
            sender('update_action', update_payload)
            
            if function_name in ['create_file', 'delete_file', 'rename_file', 'create_folder']:
                if not web_terminal.context_manager._is_host_mode_without_safety():
                    structure = web_terminal.context_manager.get_project_structure()
                    sender('file_tree_update', structure)
            
            # ===== 增量保存：立即保存工具结果 =====
            metadata_payload = None
            tool_images = None
            tool_videos = None
            if isinstance(result_data, dict):
                # 特殊处理 web_search：保留可供前端渲染的精简结构，以便历史记录复现搜索结果
                if function_name == "web_search":
                    try:
                        tool_result_content = json.dumps(compact_web_search_result(result_data), ensure_ascii=False)
                    except Exception:
                        tool_result_content = tool_result
                else:
                    tool_result_content = format_tool_result_for_context(function_name, result_data, tool_result)
                metadata_payload = {"tool_payload": result_data}
            else:
                tool_result_content = tool_result
            tool_message_content = tool_result_content

            # view_image: 将图片直接附加到 tool 结果中（不再插入 user 消息）
            if function_name == "view_image" and getattr(web_terminal, "pending_image_view", None):
                inj = web_terminal.pending_image_view
                web_terminal.pending_image_view = None
                if (
                    not tool_failed
                    and isinstance(result_data, dict)
                    and result_data.get("success") is not False
                ):
                    img_path = inj.get("path") if isinstance(inj, dict) else None
                    if img_path:
                        text_part = tool_result_content if isinstance(tool_result_content, str) else ""
                        tool_message_content = web_terminal.context_manager._build_content_with_images(
                            text_part,
                            [img_path]
                        )
                        tool_images = [img_path]
                        if metadata_payload is None:
                            metadata_payload = {}
                        metadata_payload["tool_image_path"] = img_path
                        sender('system_message', {
                            'content': f'系统已按模型请求将图片附加到工具结果: {img_path}'
                        })

            # view_video: 将视频直接附加到 tool 结果中（不再插入 user 消息）
            if function_name == "view_video" and getattr(web_terminal, "pending_video_view", None):
                inj = web_terminal.pending_video_view
                web_terminal.pending_video_view = None
                if (
                    not tool_failed
                    and isinstance(result_data, dict)
                    and result_data.get("success") is not False
                ):
                    video_path = inj.get("path") if isinstance(inj, dict) else None
                    if video_path:
                        text_part = tool_result_content if isinstance(tool_result_content, str) else ""
                        video_payload = [video_path]
                        tool_message_content = web_terminal.context_manager._build_content_with_images(
                            text_part,
                            [],
                            video_payload
                        )
                        tool_videos = [video_path]
                        if metadata_payload is None:
                            metadata_payload = {}
                        metadata_payload["tool_video_path"] = video_path
                        sender('system_message', {
                            'content': f'系统已按模型请求将视频附加到工具结果: {video_path}'
                        })
            
            # 立即保存工具结果
            web_terminal.context_manager.add_conversation(
                "tool",
                tool_result_content,
                tool_call_id=tool_call_id,
                name=function_name,
                metadata=metadata_payload,
                images=tool_images,
                videos=tool_videos
            )
            debug_log(f"💾 增量保存：工具结果 {function_name}")
            system_message = result_data.get("system_message") if isinstance(result_data, dict) else None
            if system_message:
                web_terminal._record_sub_agent_message(system_message, result_data.get("task_id"), inline=False)
                maybe_mark_failure_from_message(web_terminal, system_message)
            
            # 添加到消息历史（用于API继续对话）
            messages.append({
                "role": "tool",
                "tool_call_id": tool_call_id,
                "name": function_name,
                "content": tool_message_content
            })

            if function_name not in {'write_file', 'edit_file'}:
                await process_sub_agent_updates(messages, inline=True, after_tool_call_id=tool_call_id)
            
            await asyncio.sleep(0.2)
            
            if tool_failed:
                mark_force_thinking(web_terminal, reason=f"{function_name}_failed")
        
        # 标记不再是第一次迭代
        is_first_iteration = False

    
    # 最终统计
    debug_log(f"\n{'='*40}")
    debug_log(f"任务完成统计:")
    debug_log(f"  总迭代次数: {total_iterations}")
    debug_log(f"  总工具调用: {total_tool_calls}")
    debug_log(f"  自动修复尝试: {auto_fix_attempts}")
    debug_log(f"  累积响应: {len(accumulated_response)} 字符")
    debug_log(f"{'='*40}\n")
    
    # 发送完成事件
    sender('task_complete', {
        'total_iterations': total_iterations,
        'total_tool_calls': total_tool_calls,
        'auto_fix_attempts': auto_fix_attempts
    })