diff --git a/src/coding/proxy/config/config.default.yaml b/src/coding/proxy/config/config.default.yaml
index f67c542..52950e0 100644
--- a/src/coding/proxy/config/config.default.yaml
+++ b/src/coding/proxy/config/config.default.yaml
@@ -704,3 +704,10 @@ session_policies:
   title_vendor_bindings:
     - prefix: "# 目标"
       vendor: "zhipu"
+  # Session 标题豁免前缀：user 输入（经噪声剥离清洗后）以任一前缀开头则跳过，
+  # 继续向后查找 title 候选。用于过滤注入式 Prompt（如示例语言指令），
+  # 避免其被误用作 Session 标题。大小写敏感 startswith。
+  # ⚠️ 自定义 yaml 中该字段为整体替换（loader 的 list 合并约定，非追加合并）；
+  #    写空列表即关闭豁免。加载时会自动 strip + 去空 + 去重。
+  title_exempt_prefixes:
+    - "Write the title in the language"
diff --git a/src/coding/proxy/config/session_policy.py b/src/coding/proxy/config/session_policy.py
index c25da7a..7217f73 100644
--- a/src/coding/proxy/config/session_policy.py
+++ b/src/coding/proxy/config/session_policy.py
@@ -2,7 +2,7 @@
 
 from __future__ import annotations
 
-from pydantic import BaseModel, Field
+from pydantic import BaseModel, Field, field_validator
 
 
 class SessionPolicyMatch(BaseModel):
@@ -81,3 +81,34 @@ class SessionPoliciesConfig(BaseModel):
             "匹配规则按列表顺序求值，首次匹配生效。"
         ),
     )
+    title_exempt_prefixes: list[str] = Field(
+        default_factory=list,
+        description=(
+            "Session 标题豁免前缀名单。当首条 user TEXT 输入（经噪声剥离清洗后）"
+            "以任一前缀开头时，跳过该输入、继续向后查找合适的 title 候选。"
+            "用于过滤注入式 Prompt（如示例语言指令），避免其被误用作 Session 标题。"
+            "大小写敏感的 startswith 匹配，与 title_vendor_bindings 语义一致。"
+            "加载时会自动 strip + 去空 + 去重；空字符串前缀会被丢弃"
+            "（防空串 startswith 恒真导致全量误豁免）。"
+        ),
+    )
+
+    @field_validator("title_exempt_prefixes", mode="after")
+    @classmethod
+    def _normalize_exempt_prefixes(cls, value: list[str]) -> list[str]:
+        """归一化豁免前缀：strip + 去空 + 去重保序.
+
+        空字符串前缀必须丢弃——``"".startswith`` 恒真会豁免一切用户输入，
+        导致 Level 1 标题提取永久失效。
+        """
+        seen: set[str] = set()
+        result: list[str] = []
+        for item in value:
+            if not isinstance(item, str):
+                continue
+            cleaned = item.strip()
+            if not cleaned or cleaned in seen:
+                continue
+            seen.add(cleaned)
+            result.append(cleaned)
+        return result
diff --git a/src/coding/proxy/routing/executor.py b/src/coding/proxy/routing/executor.py
index 9ade9dd..0569126 100644
--- a/src/coding/proxy/routing/executor.py
+++ b/src/coding/proxy/routing/executor.py
@@ -148,14 +148,30 @@ def _sanitize_user_text(raw: str) -> str:
 # ─────────────────────────────────────────────────────────────────
 
 
-def _extract_title_from_user_text(messages: list[CanonicalMessagePart]) -> str:
-    """Level 1: 从 user TEXT 部分提取经噪声剥离后的首条非空文本."""
+def _extract_title_from_user_text(
+    messages: list[CanonicalMessagePart],
+    exempt_prefixes: list[str] | None = None,
+) -> str:
+    """Level 1: 从 user TEXT 部分提取经噪声剥离后的首条非空文本.
+
+    ``exempt_prefixes`` 非空时，清洗后文本以任一前缀开头则跳过该输入、继续向后
+    查找——用于过滤注入式 Prompt，避免其被误用作 Session 标题。匹配作用于
+    ``_sanitize_user_text`` 清洗后的纯文本（而非带 XML 标签的原始 raw）。
+    """
     for part in messages:
         if part.role != "user" or part.type != CanonicalPartType.TEXT:
             continue
         cleaned = _sanitize_user_text(part.text)
-        if cleaned:
-            return cleaned[:_SESSION_TITLE_MAX_LEN]
+        if not cleaned:
+            continue
+        if exempt_prefixes and any(cleaned.startswith(p) for p in exempt_prefixes):
+            logger.debug(
+                "Title candidate skipped by exempt prefix %r (snippet=%r)",
+                next(p for p in exempt_prefixes if cleaned.startswith(p)),
+                cleaned[:60],
+            )
+            continue
+        return cleaned[:_SESSION_TITLE_MAX_LEN]
     return ""
 
 
@@ -193,15 +209,24 @@ def _extract_title_from_metadata(request: CanonicalRequest) -> str:
     return ""
 
 
-def _extract_session_title(request: CanonicalRequest) -> str:
+def _extract_session_title(
+    request: CanonicalRequest,
+    exempt_prefixes: list[str] | None = None,
+) -> str:
     """从规范化请求中提取 session 标题 — 多层级回退策略。
 
     依次尝试: user TEXT 噪声剥离 → TOOL_RESULT 摘要 → IMAGE 计数 → 元数据兜底。
     任意级别命中即返回,确保 Dashboard 尽可能展示有辨识度的标题。
+
+    ``exempt_prefixes`` 仅作用于 Level 1（用户直接输入的 TEXT）；Level 2/3/4
+    为非用户直接输入的回退来源,不参与豁免。默认 ``None`` 时行为与历史一致。
     """
     messages = request.messages
+    # Level 1 需要豁免前缀；单独调用以传入参数，避免污染 L2/L3 的统一签名。
+    title = _extract_title_from_user_text(messages, exempt_prefixes)
+    if title:
+        return title[:_SESSION_TITLE_MAX_LEN]
     for extractor in (
-        _extract_title_from_user_text,
         _extract_title_from_tool_results,
         _extract_title_from_images,
     ):
@@ -614,6 +639,7 @@ def __init__(
         reauth_coordinator: Any | None = None,
         session_policy_resolver: SessionPolicyResolver | None = None,
         title_vendor_bindings: list[TitleVendorBinding] | None = None,
+        title_exempt_prefixes: list[str] | None = None,
     ) -> None:
         self._router = router
         self._tiers = tiers
@@ -622,6 +648,13 @@ def __init__(
         self._reauth_coordinator = reauth_coordinator
         self._policy_resolver = session_policy_resolver or SessionPolicyResolver()
         self._title_vendor_bindings = title_vendor_bindings or []
+        # 豁免前缀名单：构造期二次归一化（strip + 去空），双重防御配置层 field_validator。
+        # 即便绕过配置层直接构造 executor，也能避免空串 startswith 恒真与纯空白失配。
+        self._title_exempt_prefixes: list[str] = [
+            stripped
+            for p in (title_exempt_prefixes or [])
+            if isinstance(p, str) and (stripped := p.strip())
+        ]
         self._validate_title_vendor_bindings()
 
         # Tier 名称 → OAuth provider 名称的映射
@@ -676,6 +709,17 @@ def _resolve_effective_tiers(self, session_key: str) -> list[VendorTier]:
                 seen.add(tier.name)
         return ordered
 
+    def _extract_session_title(self, request: CanonicalRequest) -> str:
+        """封装模块级标题提取，注入实例持有的豁免前缀名单.
+
+        将 ``self._title_exempt_prefixes`` 透传给模块级 ``_extract_session_title``，
+        使 Level 1 能跳过以豁免前缀开头的注入式 Prompt。4 处标题提取调用点统一
+        经此方法，避免重复透传（零透传污染）。
+        """
+        return _extract_session_title(
+            request, exempt_prefixes=self._title_exempt_prefixes
+        )
+
     def _apply_title_based_policy(self, session_key: str, title: str) -> None:
         """根据 Session 标题前缀自动绑定供应商.
 
@@ -790,7 +834,7 @@ async def execute_stream(
             canonical_request.trace_id,
         )
         if is_new_session:
-            title = _extract_session_title(canonical_request)
+            title = self._extract_session_title(canonical_request)
             if title:
                 await self._recorder.set_session_title(
                     canonical_request.session_key, title
@@ -798,7 +842,7 @@ async def execute_stream(
                 self._apply_title_based_policy(canonical_request.session_key, title)
         else:
             # 延迟标题补写: 若 session 尚无标题,尝试从当前请求中提取并回写。
-            title = _extract_session_title(canonical_request)
+            title = self._extract_session_title(canonical_request)
             if title:
                 await self._recorder.update_empty_session_title(
                     canonical_request.session_key, title
@@ -977,7 +1021,7 @@ async def execute_message(
             canonical_request.trace_id,
         )
         if is_new_session:
-            title = _extract_session_title(canonical_request)
+            title = self._extract_session_title(canonical_request)
             if title:
                 await self._recorder.set_session_title(
                     canonical_request.session_key, title
@@ -985,7 +1029,7 @@ async def execute_message(
                 self._apply_title_based_policy(canonical_request.session_key, title)
         else:
             # 延迟标题补写: 若 session 尚无标题,尝试从当前请求中提取并回写。
-            title = _extract_session_title(canonical_request)
+            title = self._extract_session_title(canonical_request)
             if title:
                 await self._recorder.update_empty_session_title(
                     canonical_request.session_key, title
diff --git a/src/coding/proxy/routing/router.py b/src/coding/proxy/routing/router.py
index a9b5379..3ea162c 100644
--- a/src/coding/proxy/routing/router.py
+++ b/src/coding/proxy/routing/router.py
@@ -40,6 +40,7 @@ def __init__(
         compat_session_store: CompatSessionStore | None = None,
         session_policy_resolver: SessionPolicyResolver | None = None,
         title_vendor_bindings: list[TitleVendorBinding] | None = None,
+        title_exempt_prefixes: list[str] | None = None,
     ) -> None:
         if not tiers:
             raise ValueError("至少需要一个供应商层级")
@@ -59,6 +60,7 @@ def __init__(
             reauth_coordinator=reauth_coordinator,
             session_policy_resolver=session_policy_resolver,
             title_vendor_bindings=title_vendor_bindings,
+            title_exempt_prefixes=title_exempt_prefixes,
         )
 
     def set_pricing_table(self, table: PricingTable) -> None:
diff --git a/src/coding/proxy/server/app.py b/src/coding/proxy/server/app.py
index 396b89f..cf8aad4 100644
--- a/src/coding/proxy/server/app.py
+++ b/src/coding/proxy/server/app.py
@@ -162,6 +162,7 @@ def create_app(config: ProxyConfig | None = None) -> FastAPI:
         compat_session_store,
         session_policy_resolver=SessionPolicyResolver(config.session_policies.policies),
         title_vendor_bindings=config.session_policies.title_vendor_bindings,
+        title_exempt_prefixes=config.session_policies.title_exempt_prefixes,
     )
 
     app = FastAPI(title="coding-proxy", version=__version__, lifespan=lifespan)
diff --git a/tests/test_router_executor.py b/tests/test_router_executor.py
index 2e5c13e..4077271 100644
--- a/tests/test_router_executor.py
+++ b/tests/test_router_executor.py
@@ -2409,6 +2409,80 @@ def test_skips_assistant_role(self):
         req = self._build_request(messages)
         assert _extract_session_title(req) == "新的用户问题"
 
+    # ── 豁免前缀端到端（exempt_prefixes 透传至 Level 1）──
+
+    def test_exempt_prefix_falls_through_to_next_user_text(self):
+        """端到端：L1 首条命中豁免 → 取第二条 user text 作为标题."""
+        messages = [
+            {"role": "user", "content": "Write the title in the language ..."},
+            {"role": "user", "content": "端到端业务标题"},
+        ]
+        req = self._build_request(messages)
+        assert (
+            _extract_session_title(
+                req, exempt_prefixes=["Write the title in the language"]
+            )
+            == "端到端业务标题"
+        )
+
+    def test_exempt_l1_falls_back_to_l2_tool_result(self):
+        """L1 命中豁免且无其他 user text → 回退 Level 2 tool_result."""
+        messages = [
+            {"role": "user", "content": "Write the title in the language ..."},
+            {
+                "role": "user",
+                "content": [
+                    {
+                        "type": "tool_result",
+                        "tool_use_id": "tu_1",
+                        "content": [{"type": "text", "text": "文件内容摘要"}],
+                    }
+                ],
+            },
+        ]
+        req = self._build_request(messages)
+        title = _extract_session_title(
+            req, exempt_prefixes=["Write the title in the language"]
+        )
+        assert title == "[Tool output] 文件内容摘要"
+
+    def test_exempt_l1_falls_back_to_l4_metadata(self):
+        """L1 命中豁免且无 L2/L3 → 回退 Level 4 元数据兜底."""
+        messages = [{"role": "user", "content": "Write the title in the language ..."}]
+        req = self._build_request(messages)
+        assert (
+            _extract_session_title(
+                req, exempt_prefixes=["Write the title in the language"]
+            )
+            == "[Session] test"
+        )
+
+    def test_exempt_does_not_affect_l2(self):
+        """豁免仅作用于 L1：tool_result 文本以豁免前缀开头，L2 仍正常提取."""
+        messages = [
+            {
+                "role": "user",
+                "content": [
+                    {
+                        "type": "tool_result",
+                        "tool_use_id": "tu_1",
+                        "content": [
+                            {
+                                "type": "text",
+                                "text": "Write the title in the language ...",
+                            }
+                        ],
+                    }
+                ],
+            },
+        ]
+        req = self._build_request(messages)
+        title = _extract_session_title(
+            req, exempt_prefixes=["Write the title in the language"]
+        )
+        # L2 不受豁免影响，正常提取并加 [Tool output] 前缀
+        assert title == "[Tool output] Write the title in the language ..."
+
 
 # ═══════════════════════════════════════════════════════════════════
 # 多层级回退标题提取测试
@@ -2453,6 +2527,141 @@ def test_returns_empty_for_noise_only(self):
         ]
         assert _extract_title_from_user_text(msgs) == ""
 
+    # ── 豁免前缀（exempt_prefixes）—— 过滤注入式 Prompt ──
+
+    def test_exempt_prefix_skips_first_falls_through_to_second(self):
+        """豁免前缀命中首条 user text → 跳过、返回第二条."""
+        from coding.proxy.model.compat import CanonicalMessagePart, CanonicalPartType
+
+        msgs = [
+            CanonicalMessagePart(
+                type=CanonicalPartType.TEXT,
+                role="user",
+                text="Write the title in the language the user wrote in, regardless.",
+            ),
+            CanonicalMessagePart(
+                type=CanonicalPartType.TEXT, role="user", text="帮我重构 executor"
+            ),
+        ]
+        assert (
+            _extract_title_from_user_text(
+                msgs, exempt_prefixes=["Write the title in the language"]
+            )
+            == "帮我重构 executor"
+        )
+
+    def test_exempt_prefix_case_sensitive(self):
+        """大小写敏感：小写前缀不命中大写开头的输入."""
+        from coding.proxy.model.compat import CanonicalMessagePart, CanonicalPartType
+
+        msgs = [
+            CanonicalMessagePart(
+                type=CanonicalPartType.TEXT,
+                role="user",
+                text="Write the title in the language the user wrote in.",
+            ),
+        ]
+        # 前缀小写、输入首字母大写 → 不命中，正常返回
+        assert (
+            _extract_title_from_user_text(msgs, exempt_prefixes=["write the title"])
+            == "Write the title in the language the user wrote in."
+        )
+
+    def test_exempt_no_param_backward_compatible(self):
+        """不传 exempt_prefixes → 行为与历史一致（首条非空即返回）."""
+        from coding.proxy.model.compat import CanonicalMessagePart, CanonicalPartType
+
+        msgs = [
+            CanonicalMessagePart(
+                type=CanonicalPartType.TEXT,
+                role="user",
+                text="Write the title in the language the user wrote in.",
+            ),
+        ]
+        assert _extract_title_from_user_text(msgs) == (
+            "Write the title in the language the user wrote in."
+        )
+
+    def test_exempt_empty_list_no_op(self):
+        """传空列表 → 等价于不豁免."""
+        from coding.proxy.model.compat import CanonicalMessagePart, CanonicalPartType
+
+        msgs = [
+            CanonicalMessagePart(
+                type=CanonicalPartType.TEXT,
+                role="user",
+                text="Write the title in the language the user wrote in.",
+            ),
+        ]
+        assert _extract_title_from_user_text(msgs, exempt_prefixes=[]) == (
+            "Write the title in the language the user wrote in."
+        )
+
+    def test_exempt_all_user_inputs_skipped_returns_empty(self):
+        """全部 user text 命中豁免 → 返回空字符串."""
+        from coding.proxy.model.compat import CanonicalMessagePart, CanonicalPartType
+
+        msgs = [
+            CanonicalMessagePart(
+                type=CanonicalPartType.TEXT,
+                role="user",
+                text="Write the title in the language variant A",
+            ),
+            CanonicalMessagePart(
+                type=CanonicalPartType.TEXT,
+                role="user",
+                text="Write the title in the language variant B",
+            ),
+        ]
+        assert (
+            _extract_title_from_user_text(
+                msgs, exempt_prefixes=["Write the title in the language"]
+            )
+            == ""
+        )
+
+    def test_exempt_matches_cleaned_not_raw(self):
+        """豁免判断作用于清洗后文本：raw 含 system-reminder 包裹、清洗后命中前缀 → 跳过."""
+        from coding.proxy.model.compat import CanonicalMessagePart, CanonicalPartType
+
+        raw = (
+            "<system-reminder>注入的系统上下文</system-reminder>"
+            "Write the title in the language the user wrote in."
+        )
+        msgs = [
+            CanonicalMessagePart(type=CanonicalPartType.TEXT, role="user", text=raw),
+            CanonicalMessagePart(
+                type=CanonicalPartType.TEXT, role="user", text="真实业务问题"
+            ),
+        ]
+        assert (
+            _extract_title_from_user_text(
+                msgs, exempt_prefixes=["Write the title in the language"]
+            )
+            == "真实业务问题"
+        )
+
+    def test_exempt_multi_prefix_any_match(self):
+        """多前缀：命中任一即跳过."""
+        from coding.proxy.model.compat import CanonicalMessagePart, CanonicalPartType
+
+        msgs = [
+            CanonicalMessagePart(
+                type=CanonicalPartType.TEXT,
+                role="user",
+                text="IGNORE_PREFIX_A something",
+            ),
+            CanonicalMessagePart(
+                type=CanonicalPartType.TEXT, role="user", text="第二输入"
+            ),
+        ]
+        assert (
+            _extract_title_from_user_text(
+                msgs, exempt_prefixes=["IGNORE_PREFIX_A", "IGNORE_PREFIX_B"]
+            )
+            == "第二输入"
+        )
+
 
 class TestExtractTitleFromToolResults:
     """Level 2 辅助函数 ``_extract_title_from_tool_results``."""
@@ -2946,3 +3155,39 @@ def test_known_vendor_no_startup_warning(self, caplog):
             r for r in caplog.records if "title_vendor_bindings" in r.message
         ]
         assert not binding_warnings
+
+
+class TestExemptPrefixInjection:
+    """``_RouteExecutor._extract_session_title`` 实例方法注入豁免前缀名单."""
+
+    @staticmethod
+    def _build_request(messages: list[dict]):
+        return build_canonical_request({"model": "test", "messages": messages}, {})
+
+    def test_executor_injects_exempt_prefixes_to_l1(self):
+        """executor 持有的豁免前缀经实例方法注入，L1 跳过注入式 Prompt."""
+        req = self._build_request(
+            [
+                {"role": "user", "content": "Write the title in the language ..."},
+                {"role": "user", "content": "注入后的真实标题"},
+            ]
+        )
+        executor = _executor(title_exempt_prefixes=["Write the title in the language"])
+        assert executor._extract_session_title(req) == "注入后的真实标题"
+
+    def test_executor_without_exempt_prefixes_backward_compatible(self):
+        """未配置豁免前缀 → 实例方法行为与模块级默认一致（不跳过）."""
+        req = self._build_request(
+            [{"role": "user", "content": "Write the title in the language ..."}]
+        )
+        executor = _executor()  # 不传 title_exempt_prefixes
+        assert executor._extract_session_title(req) == (
+            "Write the title in the language ..."
+        )
+
+    def test_executor_filters_empty_string_prefix_at_construction(self):
+        """构造期过滤空串：即便传入空串/纯空白也不会豁免一切."""
+        req = self._build_request([{"role": "user", "content": "正常标题"}])
+        executor = _executor(title_exempt_prefixes=["", "  ", "\t"])
+        assert executor._title_exempt_prefixes == []
+        assert executor._extract_session_title(req) == "正常标题"
diff --git a/tests/test_session_aware.py b/tests/test_session_aware.py
index 29518e5..9f3441a 100644
--- a/tests/test_session_aware.py
+++ b/tests/test_session_aware.py
@@ -462,6 +462,30 @@ def test_config_parse():
     assert config.policies[0].tiers == ["anthropic", "copilot"]
 
 
+def test_config_exempt_prefixes_default_empty():
+    """未配置时默认空列表."""
+    config = SessionPoliciesConfig()
+    assert config.title_exempt_prefixes == []
+
+
+def test_config_exempt_prefixes_strips_whitespace():
+    """加载时 strip 每个前缀的前后空白."""
+    config = SessionPoliciesConfig(title_exempt_prefixes=["  Write  ", "\tRead\t"])
+    assert config.title_exempt_prefixes == ["Write", "Read"]
+
+
+def test_config_exempt_prefixes_drops_empty_string():
+    """空字符串/纯空白前缀必须丢弃——防空串 startswith 恒真导致全量误豁免."""
+    config = SessionPoliciesConfig(title_exempt_prefixes=["", "   ", "Write"])
+    assert config.title_exempt_prefixes == ["Write"]
+
+
+def test_config_exempt_prefixes_dedupes_preserves_order():
+    """去重并保留首次出现顺序."""
+    config = SessionPoliciesConfig(title_exempt_prefixes=["a", "b", "a", "b", "c"])
+    assert config.title_exempt_prefixes == ["a", "b", "c"]
+
+
 # ── 9. SessionPolicyResolver 运行时可变性 ────────────────────────