diff --git a/modules/goal_state_manager.py b/modules/goal_state_manager.py
index 7eb562e4..d2acad1b 100644
--- a/modules/goal_state_manager.py
+++ b/modules/goal_state_manager.py
@@ -1,23 +1,28 @@
-"""目标模式（Goal Mode）的工作区级状态管理。
+"""目标模式（Goal Mode）的对话级状态管理。
 
-每个工作区/项目最多存在一个活动目标。状态落盘到 `{data_dir}/goal_state.json`，
-以便在切换对话、对话压缩（conversation_id 变化）、进程重启后仍能维持。
+每个对话各自持有独立的目标状态，互不影响。状态落盘到
+`{data_dir}/goal_states/<conversation_id>.json`，在切换对话、对话压缩、
+进程重启后仍能维持（压缩不会改变 conversation_id，压缩 handoff 重入时
+由主循环入口按当前 conversation_id 重新加载本对话状态并续注提示词）。
 
 设计要点：
-- 不依赖 web_terminal，只接受一个 data_dir，便于单测与复用。
+- 不依赖 web_terminal，只接受 data_dir 与 conversation_id，便于单测与复用。
 - token 基线对齐工作区级累计 `{data_dir}/token_totals.json`（input/output/total）。
+  注意：该累计是整个工作区所有对话共享的，多对话并发运行时 max_tokens
+  边界判定是近似值（可能因其他对话的消耗而提前触发）。
 - review_history 保存历轮 {main_output, review_reply}，用于构建交叉结构审核文本。
 """
 
 from __future__ import annotations
 
 import json
+import re
 import time
 from copy import deepcopy
 from pathlib import Path
 from typing import Any, Dict, List, Optional, Union
 
-GOAL_STATE_FILENAME = "goal_state.json"
+GOAL_STATES_DIRNAME = "goal_states"
 
 # 状态机
 STATUS_RUNNING = "running"
@@ -36,6 +41,16 @@ REVIEW_MODE_ACTIVE = "active"
 
 PathLike = Union[str, Path]
 
+_SAFE_CONVERSATION_ID = re.compile(r"^[A-Za-z0-9_-]+$")
+
+
+def _validate_conversation_id(conversation_id: str) -> str:
+    """conversation_id 用作状态文件名，必须是安全字符，防止路径穿越。"""
+    cid = str(conversation_id or "").strip()
+    if not cid or not _SAFE_CONVERSATION_ID.match(cid):
+        raise ValueError(f"非法 conversation_id: {conversation_id!r}")
+    return cid
+
 
 def _empty_state() -> Dict[str, Any]:
     return {
@@ -45,7 +60,6 @@ def _empty_state() -> Dict[str, Any]:
         "review_mode": REVIEW_MODE_READONLY,
         "max_turns": 5,
         "max_tokens": None,
-        "start_conversation_id": None,
         "started_at": None,
         "turn_count": 0,
         "token_baseline": {"input_tokens": 0, "output_tokens": 0, "total_tokens": 0},
@@ -58,21 +72,22 @@ def _empty_state() -> Dict[str, Any]:
 
 
 class GoalStateManager:
-    """工作区级目标状态。一个实例对应一个工作区的 goal_state.json。"""
+    """对话级目标状态。一个实例对应一个对话的 goal_states/<conversation_id>.json。"""
 
-    def __init__(self, data_dir: PathLike):
+    def __init__(self, data_dir: PathLike, conversation_id: str):
         self.data_dir = Path(data_dir).expanduser()
+        self.conversation_id = _validate_conversation_id(conversation_id)
         self.state: Dict[str, Any] = self.load()
 
     # ------------------------------------------------------------------ 路径/持久化
 
     def _path(self) -> Path:
-        return self.data_dir / GOAL_STATE_FILENAME
+        return self.data_dir / GOAL_STATES_DIRNAME / f"{self.conversation_id}.json"
 
     @classmethod
-    def load_from(cls, data_dir: PathLike) -> "GoalStateManager":
-        """便捷构造：等价于 GoalStateManager(data_dir)。"""
-        return cls(data_dir)
+    def load_from(cls, data_dir: PathLike, conversation_id: str) -> "GoalStateManager":
+        """便捷构造：等价于 GoalStateManager(data_dir, conversation_id)。"""
+        return cls(data_dir, conversation_id)
 
     def load(self) -> Dict[str, Any]:
         path = self._path()
@@ -112,7 +127,6 @@ class GoalStateManager:
         review_mode: str,
         max_turns: Optional[int],
         max_tokens: Optional[int],
-        conversation_id: Optional[str],
         token_baseline: Optional[Dict[str, int]] = None,
         tool_call_baseline: int = 0,
     ) -> Dict[str, Any]:
@@ -131,7 +145,6 @@ class GoalStateManager:
             "review_mode": rm,
             "max_turns": int(max_turns) if max_turns else None,
             "max_tokens": int(max_tokens) if max_tokens else None,
-            "start_conversation_id": conversation_id,
             "started_at": time.time(),
             "turn_count": 0,
             "token_baseline": baseline,
@@ -194,8 +207,15 @@ class GoalStateManager:
         return deepcopy(self.state)
 
     def clear(self) -> None:
-        """完全清除目标状态（用户取消）。"""
+        """完全清除本对话的目标状态。"""
         self.state = _empty_state()
+        path = self._path()
+        try:
+            if path.exists():
+                path.unlink()
+            return
+        except OSError:
+            pass
         self.save()
 
     # ------------------------------------------------------------------ 边界判定
@@ -267,6 +287,7 @@ class GoalStateManager:
             except Exception:
                 pass
         return {
+            "conversation_id": self.conversation_id,
             "goal": self.get_goal(),
             "status": self.state.get("status"),
             "turn_count": self.get_turn(),
diff --git a/server/chat_flow_task_main.py b/server/chat_flow_task_main.py
index b67e9281..ae865227 100644
--- a/server/chat_flow_task_main.py
+++ b/server/chat_flow_task_main.py
@@ -1868,12 +1868,11 @@ async def handle_task_with_sender(
                 goal_text=message,
                 current_tool_calls=0,
             )
-            # 本标记只表示“本次用户显式开启目标模式”。目标状态已写入
-            # goal_state.json 后必须立即消费掉，否则自动深层压缩后的递归
-            # handoff 会再次进入 handle_task_with_sender，把压缩引导语误当作
-            # 新目标并重置目标文本、轮数、token、工具次数和开始时间。
+            # 本标记只表示“本次用户显式开启目标模式”，写入对话级目标状态后
+            # 必须立即消费掉：自动深层压缩会递归重入 handle_task_with_sender，
+            # 若不消费，压缩引导语会被误当作新目标再次启动目标模式。
             setattr(web_terminal, "_goal_mode_requested", False)
-        if goal_is_active(workspace):
+        if goal_is_active(workspace, conversation_id):
             inject_goal_prompt(
                 web_terminal=web_terminal,
                 messages=messages,
@@ -1884,6 +1883,7 @@ async def handle_task_with_sender(
                 web_terminal=web_terminal,
                 workspace=workspace,
                 sender=sender,
+                conversation_id=conversation_id,
                 total_tool_calls=0,
             )
     except Exception as exc:
@@ -1969,6 +1969,7 @@ async def handle_task_with_sender(
                     web_terminal=web_terminal,
                     workspace=workspace,
                     sender=sender,
+                    conversation_id=conversation_id,
                     total_tool_calls=total_tool_calls,
                 )
             except Exception as exc:
@@ -2227,7 +2228,7 @@ async def handle_task_with_sender(
             debug_log("没有工具调用，结束迭代")
             # === 目标模式：turn 结束拦截 ===
             try:
-                if goal_is_active(workspace):
+                if goal_is_active(workspace, conversation_id):
                     goal_result = await handle_goal_after_turn(
                         web_terminal=web_terminal,
                         workspace=workspace,
@@ -2311,11 +2312,12 @@ async def handle_task_with_sender(
         # 更新统计
         total_tool_calls += len(tool_calls)
         try:
-            if goal_is_active(workspace):
+            if goal_is_active(workspace, conversation_id):
                 emit_goal_progress(
                     web_terminal=web_terminal,
                     workspace=workspace,
                     sender=sender,
+                    conversation_id=conversation_id,
                     total_tool_calls=total_tool_calls,
                 )
         except Exception as exc:
diff --git a/server/goal_flow.py b/server/goal_flow.py
index d35f4a16..e98eafc6 100644
--- a/server/goal_flow.py
+++ b/server/goal_flow.py
@@ -1,7 +1,8 @@
 """目标模式（Goal Mode）在 Web 任务主循环中的编排逻辑。
 
-将启动、提示词注入、turn 结束后的审核/续命/停止判定集中在此，
-尽量减少对 chat_flow_task_main.py 的侵入。
+目标状态是对话级的（见 modules/goal_state_manager.py）：每个对话独立持有
+自己的目标，多对话并发互不影响。本模块将启动、提示词注入、turn 结束后的
+审核/续命/停止判定集中在此，尽量减少对 chat_flow_task_main.py 的侵入。
 """
 
 from __future__ import annotations
@@ -57,17 +58,18 @@ def maybe_start_goal(
     *,
     web_terminal,
     workspace,
-    conversation_id: Optional[str],
+    conversation_id: str,
     goal_text: str,
     current_tool_calls: int,
 ) -> bool:
-    """启动一个新目标。
+    """为本对话启动一个新目标。
 
-    本函数语义是“本次用户显式开启目标模式”而不是“续接已有目标”。
-    因此必须覆盖工作区级 goal_state.json 中的旧目标状态（包括 running/done/stopped、
-    review_history、token 基线、开始时间等），避免新对话复用上一轮目标。
+    目标状态是对话级的：启动会覆盖本对话自己的旧目标状态（running/done/stopped、
+    review_history、token 基线、开始时间等），不影响其他对话的目标。
     """
-    gsm = GoalStateManager(workspace.data_dir)
+    if not conversation_id:
+        return False
+    gsm = GoalStateManager(workspace.data_dir, conversation_id)
     cfg = load_personalization_config(workspace.data_dir)
     review_mode = cfg.get("goal_review_mode") or "readonly"
     end_conditions = cfg.get("goal_end_conditions") or ["max_turns"]
@@ -78,7 +80,6 @@ def maybe_start_goal(
         review_mode=review_mode,
         max_turns=max_turns,
         max_tokens=max_tokens,
-        conversation_id=conversation_id,
         token_baseline=snapshot_token_baseline(web_terminal),
         tool_call_baseline=current_tool_calls,
     )
@@ -105,8 +106,13 @@ def inject_goal_prompt(
     )
 
 
-def goal_is_active(workspace) -> bool:
-    return GoalStateManager(workspace.data_dir).is_active()
+def goal_is_active(workspace, conversation_id: Optional[str]) -> bool:
+    if not conversation_id:
+        return False
+    try:
+        return GoalStateManager(workspace.data_dir, conversation_id).is_active()
+    except ValueError:
+        return False
 
 
 def _emit_snapshot(sender, event: str, gsm: GoalStateManager, *, total_tokens: int, tool_calls: int, extra: Optional[Dict] = None):
@@ -126,11 +132,18 @@ def emit_goal_progress(
     web_terminal,
     workspace,
     sender: Optional[Callable[[str, Dict[str, Any]], None]],
+    conversation_id: Optional[str],
     total_tool_calls: int = 0,
     extra: Optional[Dict[str, Any]] = None,
 ) -> None:
-    """向前端广播当前目标模式进度，用于刷新后恢复与进度弹窗动态数字。"""
-    gsm = GoalStateManager(workspace.data_dir)
+    """广播本对话的目标模式进度（快照自带 conversation_id，前端按对话过滤），
+    用于刷新后恢复与进度弹窗动态数字。"""
+    if not conversation_id:
+        return
+    try:
+        gsm = GoalStateManager(workspace.data_dir, conversation_id)
+    except ValueError:
+        return
     if not gsm.state:
         return
     total_tokens = _load_token_total(web_terminal)
@@ -158,12 +171,17 @@ async def handle_goal_after_turn(
     """主模型某轮结束（无 tool_calls）后的目标处理。
 
     返回 {'action': 'inactive'|'stop'|'done'|'continue', ...}：
-    - inactive：当前无活动目标，调用方照常结束本次任务。
+    - inactive：本对话无活动目标，调用方照常结束本次任务。
     - stop：目标因空转/边界停止，调用方照常结束（事件已发）。
     - done：目标达成，调用方照常结束（事件已发）。
     - continue：已注入续命消息，调用方应 continue 回主循环开下一轮。
     """
-    gsm = GoalStateManager(workspace.data_dir)
+    if not conversation_id:
+        return {"action": "inactive"}
+    try:
+        gsm = GoalStateManager(workspace.data_dir, conversation_id)
+    except ValueError:
+        return {"action": "inactive"}
     if not gsm.is_active():
         return {"action": "inactive"}
 
@@ -192,7 +210,7 @@ async def handle_goal_after_turn(
     payload_text = gsm.build_review_payload_text(assistant_content)
     if callable(sender):
         try:
-            sender("goal_review_progress", {"progress": {"stage": "start", "message": "开始审核"}})
+            sender("goal_review_progress", {"conversation_id": conversation_id, "progress": {"stage": "start", "message": "开始审核"}})
         except Exception:
             pass
     try:
@@ -200,7 +218,7 @@ async def handle_goal_after_turn(
         def _progress(progress: Dict[str, Any]) -> None:
             if callable(sender):
                 try:
-                    sender("goal_review_progress", {"progress": progress})
+                    sender("goal_review_progress", {"conversation_id": conversation_id, "progress": progress})
                 except Exception:
                     pass
 
@@ -240,9 +258,14 @@ async def handle_goal_after_turn(
     return {"action": "continue", "message": message}
 
 
-def stop_goal_user_cancel(*, web_terminal, workspace, sender, total_tool_calls: int = 0) -> bool:
-    """用户停止任务时，若有活动目标则一并停止。返回是否有活动目标被停止。"""
-    gsm = GoalStateManager(workspace.data_dir)
+def stop_goal_user_cancel(*, web_terminal, workspace, sender, conversation_id: Optional[str], total_tool_calls: int = 0) -> bool:
+    """用户停止任务时，若本对话有活动目标则一并停止。返回是否有活动目标被停止。"""
+    if not conversation_id:
+        return False
+    try:
+        gsm = GoalStateManager(workspace.data_dir, conversation_id)
+    except ValueError:
+        return False
     if not gsm.is_active():
         return False
     total_tokens = _load_token_total(web_terminal)
diff --git a/server/tasks/api.py b/server/tasks/api.py
index 9411a78e..b8118c6d 100644
--- a/server/tasks/api.py
+++ b/server/tasks/api.py
@@ -124,16 +124,16 @@ def create_task_api():
     message_source = payload.get("message_source")
     max_iterations = payload.get("max_iterations")
     goal_mode = bool(payload.get("goal_mode"))
-    # 用户显式发送非目标模式消息时，若工作区仍有上一轮残留的活动目标，先清理，
-    # 避免新对话错误继承旧目标状态。
-    if not goal_mode:
+    # 用户显式发送非目标模式消息时，若本对话仍有残留的活动目标，先停止它，
+    # 避免本对话错误延续旧目标。目标状态是对话级的，不影响其他对话。
+    if not goal_mode and conversation_id:
         try:
             _terminal, workspace = get_user_resources(username, workspace_id, conversation_id=conversation_id)
             if workspace:
-                gsm = GoalStateManager(workspace.data_dir)
+                gsm = GoalStateManager(workspace.data_dir, conversation_id)
                 if gsm.is_active():
-                    gsm.mark_stopped("new_conversation")
-                    debug_log(f"[Goal] 新任务未开启目标模式，清理工作区残留目标状态")
+                    gsm.mark_stopped("new_message_without_goal")
+                    debug_log(f"[Goal] 新消息未开启目标模式，停止本对话残留目标状态")
         except Exception as exc:
             debug_log(f"[Goal] 新任务清理残留目标状态失败: {exc}")
     skill_context_messages: List[Dict[str, str]] = []
@@ -287,11 +287,11 @@ def cancel_task_api(task_id: str):
     try:
         if ok and rec.workspace_id:
             _, workspace = get_user_resources(username, rec.workspace_id, conversation_id=rec.conversation_id)
-            if workspace:
-                gsm = GoalStateManager(workspace.data_dir)
+            if workspace and rec.conversation_id:
+                gsm = GoalStateManager(workspace.data_dir, rec.conversation_id)
                 if gsm.is_active():
                     gsm.mark_stopped(REASON_USER_CANCEL)
-                    debug_log(f"[Goal] 用户取消任务 {task_id}，同步停止工作区目标模式")
+                    debug_log(f"[Goal] 用户取消任务 {task_id}，同步停止本对话目标模式")
     except Exception as exc:
         debug_log(f"[Goal] 取消任务时停止目标模式失败: {exc}")
     return jsonify({"success": True})
diff --git a/server/tasks/helpers.py b/server/tasks/helpers.py
index ef5f624c..19ea6ca2 100644
--- a/server/tasks/helpers.py
+++ b/server/tasks/helpers.py
@@ -21,7 +21,6 @@ from server.state import stop_flags
 from server.utils_common import debug_log, log_conn_diag
 from utils.host_workspace_debug import write_host_workspace_debug
 from config import DATA_DIR, WORKSPACE_SKILLS_DIRNAME
-from modules.goal_state_manager import GoalStateManager, REASON_USER_CANCEL
 from server.tasks.models import TaskRecord
 
 
diff --git a/server/tasks/media.py b/server/tasks/media.py
index 8b106c7a..f2074aa9 100644
--- a/server/tasks/media.py
+++ b/server/tasks/media.py
@@ -21,7 +21,6 @@ from server.state import stop_flags
 from server.utils_common import debug_log, log_conn_diag
 from utils.host_workspace_debug import write_host_workspace_debug
 from config import DATA_DIR, WORKSPACE_SKILLS_DIRNAME
-from modules.goal_state_manager import GoalStateManager, REASON_USER_CANCEL
 
 
 SKILL_FRONTMATTER_RE = re.compile(r"^---\s*\n(?P<body>.*?)\n---\s*\n?", re.S)
diff --git a/server/tasks/models.py b/server/tasks/models.py
index a8de5788..1029c983 100644
--- a/server/tasks/models.py
+++ b/server/tasks/models.py
@@ -285,11 +285,11 @@ class TaskManager:
         if rec.workspace_id:
             try:
                 _, workspace = get_user_resources(username, rec.workspace_id, conversation_id=rec.conversation_id)
-                if workspace:
-                    gsm = GoalStateManager(workspace.data_dir)
+                if workspace and rec.conversation_id:
+                    gsm = GoalStateManager(workspace.data_dir, rec.conversation_id)
                     if gsm.is_active():
                         gsm.mark_stopped(REASON_USER_CANCEL)
-                        debug_log(f"[Goal] 用户取消任务 {task_id}，同步停止工作区目标模式")
+                        debug_log(f"[Goal] 用户取消任务 {task_id}，同步停止本对话目标模式")
             except Exception as exc:
                 debug_log(f"[Goal] 取消任务时停止目标模式失败: {exc}")