diff --git a/astrbot/core/astr_main_agent.py b/astrbot/core/astr_main_agent.py
index b28959aa62..1a8f68a8c1 100644
--- a/astrbot/core/astr_main_agent.py
+++ b/astrbot/core/astr_main_agent.py
@@ -11,6 +11,7 @@
 from typing import Any, TypeGuard, cast
 
 from astrbot import logger
+from astrbot.core.agent.btw.loop_routes import route_is_available_in_loop
 from astrbot.core.agent.btw.runtime_policy import resolve_computer_runtime
 from astrbot.core.agent.chat_model import ChatModel
 from astrbot.core.agent.handoff import HandoffTool
@@ -560,6 +561,11 @@ def _append_skills_prompt(
     plugin_context: CoreExecutionContext,
 ) -> SkillSnapshot:
     runtime = str(cfg.get("computer_use_runtime", "none") or "none")
+    profile = plugin_context.get_config(umo=event.unified_msg_origin)
+    btw_config = profile.get("btw", {})
+    btw_config = btw_config if isinstance(btw_config, dict) else {}
+    btw_enabled = bool(btw_config.get("enabled", False))
+    loop_mode = "work" if event.get_extra("btw_loop") == "work" else "conversation"
     skill_manager = plugin_context.skill_manager or SkillManager(
         builtin_skill_catalog=plugin_context.catalogs.builtin_skills,
     )
@@ -568,11 +574,22 @@ def _append_skills_prompt(
         cfg,
         plugin_context.catalogs.plugins,
     )
+    if btw_enabled:
+        skills = [
+            skill
+            for skill in skills
+            if route_is_available_in_loop(
+                btw_config.get("skill_routes"),
+                route_key="skill_name",
+                route_id=skill.name,
+                loop_mode=loop_mode,
+            )
+        ]
     workspace_skills = (
         skill_manager.list_workspace_skills(
             _get_workspace_path_for_umo(event.unified_msg_origin)
         )
-        if runtime == "local"
+        if runtime == "local" and (not btw_enabled or loop_mode == "work")
         else []
     )
     if persona and persona.get("skills") is not None:
diff --git a/astrbot/core/config/default.py b/astrbot/core/config/default.py
index 7fb4da990a..e8116a8314 100644
--- a/astrbot/core/config/default.py
+++ b/astrbot/core/config/default.py
@@ -200,6 +200,7 @@
         "work_session": {"max_age_seconds": 3600},
         "plugin_routes": [],
         "mcp_routes": [],
+        "skill_routes": [],
     },
     "provider_stt_settings": {
         "enable": False,
@@ -4760,6 +4761,13 @@
             "_special": "select_mcp_loop_routes",
             "condition": {"btw.enabled": True},
         },
+        "btw.skill_routes": {
+            "description": "Skills 循环分配",
+            "type": "list",
+            "hint": "普通 Skill 默认注入两个循环，可显式限制到单一循环；工作区 Skill 仅在本地工作循环可用。",
+            "_special": "select_skill_loop_routes",
+            "condition": {"btw.enabled": True},
+        },
     },
 }
 
diff --git a/dashboard/src/components/shared/CapabilityLoopSelector.vue b/dashboard/src/components/shared/CapabilityLoopSelector.vue
index 09c20e69ec..7474638534 100644
--- a/dashboard/src/components/shared/CapabilityLoopSelector.vue
+++ b/dashboard/src/components/shared/CapabilityLoopSelector.vue
@@ -40,15 +40,16 @@
 
 <script setup lang="ts">
 import { computed, onMounted, ref } from 'vue';
-import { mcpApi } from '@/api/v1';
+import { mcpApi, skillApi } from '@/api/v1';
 import { useModuleI18n } from '@/i18n/composables';
 
-type CapabilityKind = 'mcp';
+type CapabilityKind = 'mcp' | 'skill';
 type LoopMode = 'conversation' | 'work' | 'both';
-type RouteKey = 'server_name';
+type RouteKey = 'server_name' | 'skill_name';
 
 interface RouteEntry {
   server_name?: unknown;
+  skill_name?: unknown;
   loop?: unknown;
 }
 
@@ -70,10 +71,26 @@ const { tm } = useModuleI18n('features/config');
 const loading = ref(false);
 const capabilities = ref<CapabilityItem[]>([]);
 
-const routeKey = computed<RouteKey>(() => 'server_name');
-const defaultLoop = computed<LoopMode>(() => 'work');
-const hint = computed(() => tm('capabilityLoopSelector.mcpHint'));
-const emptyMessage = computed(() => tm('capabilityLoopSelector.emptyMcp'));
+const routeKey = computed<RouteKey>(() =>
+  props.kind === 'mcp' ? 'server_name' : 'skill_name',
+);
+const defaultLoop = computed<LoopMode>(() =>
+  props.kind === 'mcp' ? 'work' : 'both',
+);
+const hint = computed(() =>
+  tm(
+    props.kind === 'mcp'
+      ? 'capabilityLoopSelector.mcpHint'
+      : 'capabilityLoopSelector.skillHint',
+  ),
+);
+const emptyMessage = computed(() =>
+  tm(
+    props.kind === 'mcp'
+      ? 'capabilityLoopSelector.emptyMcp'
+      : 'capabilityLoopSelector.emptySkill',
+  ),
+);
 const loopOptions = computed(() => [
   { title: tm('capabilityLoopSelector.conversation'), value: 'conversation' },
   { title: tm('capabilityLoopSelector.work'), value: 'work' },
@@ -117,7 +134,7 @@ function normalizeItems(value: unknown): CapabilityItem[] {
       (item): item is Record<string, unknown> =>
         item !== null && typeof item === 'object',
     )
-    .filter((item) => item.active !== false)
+    .filter((item) => item.active !== false && item.plugin_active !== false)
     .map((item) => {
       const id = typeof item.name === 'string' ? item.name.trim() : '';
       const description =
@@ -133,12 +150,26 @@ function normalizeItems(value: unknown): CapabilityItem[] {
   );
 }
 
+function normalizeSkillsPayload(value: unknown): unknown[] {
+  if (value === null || typeof value !== 'object') return [];
+  const skills = (value as { skills?: unknown }).skills;
+  return Array.isArray(skills) ? skills : [];
+}
+
 async function loadCapabilities() {
   loading.value = true;
   try {
-    const response = await mcpApi.list();
+    if (props.kind === 'mcp') {
+      const response = await mcpApi.list();
+      capabilities.value =
+        response.data.status === 'ok' ? normalizeItems(response.data.data) : [];
+      return;
+    }
+
+    const response = await skillApi.list();
+    const skills = normalizeSkillsPayload(response.data.data);
     capabilities.value =
-      response.data.status === 'ok' ? normalizeItems(response.data.data) : [];
+      response.data.status === 'ok' ? normalizeItems(skills) : [];
   } catch {
     capabilities.value = [];
   } finally {
diff --git a/dashboard/src/components/shared/ConfigItemRenderer.vue b/dashboard/src/components/shared/ConfigItemRenderer.vue
index 6ff378dc86..26585f0185 100644
--- a/dashboard/src/components/shared/ConfigItemRenderer.vue
+++ b/dashboard/src/components/shared/ConfigItemRenderer.vue
@@ -76,6 +76,13 @@
         @update:model-value="emitUpdate"
       />
     </template>
+    <template v-else-if="itemMeta?._special === 'select_skill_loop_routes'">
+      <CapabilityLoopSelector
+        kind="skill"
+        :model-value="modelValue"
+        @update:model-value="emitUpdate"
+      />
+    </template>
     <template v-else-if="itemMeta?._special === 't2i_template'">
       <T2ITemplateEditor />
     </template>
diff --git a/dashboard/src/i18n/locales/en-US/features/config-metadata.json b/dashboard/src/i18n/locales/en-US/features/config-metadata.json
index 70c491a8a8..cd76f6fe71 100644
--- a/dashboard/src/i18n/locales/en-US/features/config-metadata.json
+++ b/dashboard/src/i18n/locales/en-US/features/config-metadata.json
@@ -1219,6 +1219,10 @@
         "mcp_routes": {
           "description": "MCP server loop assignments",
           "hint": "MCP tools default to the work loop; explicitly assign an enabled server to the conversation loop or both when needed."
+        },
+        "skill_routes": {
+          "description": "Skill loop assignments",
+          "hint": "Skills default to both loops; explicitly restrict an enabled Skill to one loop when needed. Workspace Skills remain work-only with the local runtime."
         }
       }
     }
diff --git a/dashboard/src/i18n/locales/en-US/features/config.json b/dashboard/src/i18n/locales/en-US/features/config.json
index b1702d5e77..8eb08f1147 100644
--- a/dashboard/src/i18n/locales/en-US/features/config.json
+++ b/dashboard/src/i18n/locales/en-US/features/config.json
@@ -215,6 +215,8 @@
     "conversation": "Conversation only",
     "work": "Work only",
     "both": "Conversation and Work",
-    "emptyMcp": "There are no enabled MCP servers."
+    "emptyMcp": "There are no enabled MCP servers.",
+    "skillHint": "Skills default to both loops. Workspace Skills remain available only to the work loop. Workspace Skills remain work-only with the local runtime.",
+    "emptySkill": "There are no enabled Skills."
   }
 }
diff --git a/dashboard/src/i18n/locales/zh-CN/features/config-metadata.json b/dashboard/src/i18n/locales/zh-CN/features/config-metadata.json
index f8b68a3954..2914c2a466 100644
--- a/dashboard/src/i18n/locales/zh-CN/features/config-metadata.json
+++ b/dashboard/src/i18n/locales/zh-CN/features/config-metadata.json
@@ -1213,6 +1213,10 @@
         "mcp_routes": {
           "description": "MCP 服务器循环分配",
           "hint": "MCP 工具默认仅在工作循环可用；可为每个已启用服务器显式改为对话循环或两者。"
+        },
+        "skill_routes": {
+          "description": "Skills 循环分配",
+          "hint": "Skill 默认注入两个循环；可为每个已启用 Skill 显式限制到单一循环。 工作区 Skill 仅在本地工作循环可用。"
         }
       }
     }
diff --git a/dashboard/src/i18n/locales/zh-CN/features/config.json b/dashboard/src/i18n/locales/zh-CN/features/config.json
index 9f6f86c63f..332efe9e67 100644
--- a/dashboard/src/i18n/locales/zh-CN/features/config.json
+++ b/dashboard/src/i18n/locales/zh-CN/features/config.json
@@ -215,6 +215,8 @@
     "conversation": "仅对话循环",
     "work": "仅工作循环",
     "both": "对话与工作循环",
-    "emptyMcp": "当前没有已启用的 MCP 服务器。"
+    "emptyMcp": "当前没有已启用的 MCP 服务器。",
+    "skillHint": "Skill 默认注入两个循环；工作区 Skill 仍仅在工作循环中可用。 工作区 Skill 仅在本地工作循环可用。",
+    "emptySkill": "当前没有已启用的 Skill。"
   }
 }
diff --git a/dashboard/tests/capabilityLoopSelector.vitest.ts b/dashboard/tests/capabilityLoopSelector.vitest.ts
index c3c709ea8c..9c041e2c2e 100644
--- a/dashboard/tests/capabilityLoopSelector.vitest.ts
+++ b/dashboard/tests/capabilityLoopSelector.vitest.ts
@@ -5,12 +5,16 @@ import { mountWithVuetify } from './utils/mountWithVuetify';
 
 const testState = vi.hoisted(() => ({
   mcpListMock: vi.fn(),
+  skillListMock: vi.fn(),
 }));
 
 vi.mock('@/api/v1', () => ({
   mcpApi: {
     list: testState.mcpListMock,
   },
+  skillApi: {
+    list: testState.skillListMock,
+  },
 }));
 
 describe('CapabilityLoopSelector', () => {
@@ -24,6 +28,22 @@ describe('CapabilityLoopSelector', () => {
         ],
       },
     });
+    testState.skillListMock.mockResolvedValue({
+      data: {
+        status: 'ok',
+        data: {
+          skills: [
+            { name: 'workspace-skill', active: true },
+            { name: 'disabled-skill', active: false },
+            {
+              name: 'disabled-plugin-skill',
+              active: true,
+              plugin_active: false,
+            },
+          ],
+        },
+      },
+    });
   });
 
   it('defaults MCP servers to work and preserves an explicit both override', async () => {
@@ -50,4 +70,35 @@ describe('CapabilityLoopSelector', () => {
     ]);
     wrapper.unmount();
   });
+
+  it('uses Skill names as the assignment key', async () => {
+    const wrapper = mountWithVuetify(CapabilityLoopSelector, {
+      props: {
+        kind: 'skill',
+        modelValue: [],
+      },
+    });
+
+    await flushPromises();
+
+    expect(wrapper.text()).toContain('workspace-skill');
+    expect(wrapper.text()).not.toContain('disabled-skill');
+    expect(wrapper.text()).not.toContain('disabled-plugin-skill');
+
+    const select = wrapper.findComponent({ name: 'VSelect' });
+    expect(select.props('modelValue')).toBe('both');
+    select.vm.$emit('update:modelValue', 'conversation');
+    await wrapper.vm.$nextTick();
+
+    expect(wrapper.emitted('update:modelValue')).toEqual([
+      [[{ skill_name: 'workspace-skill', loop: 'conversation' }]],
+    ]);
+    await wrapper.setProps({
+      modelValue: [{ skill_name: 'workspace-skill', loop: 'conversation' }],
+    });
+    select.vm.$emit('update:modelValue', 'both');
+    await wrapper.vm.$nextTick();
+    expect(wrapper.emitted('update:modelValue')?.at(-1)).toEqual([[]]);
+    wrapper.unmount();
+  });
 });
diff --git a/docs/en/dev/astrbot-config.md b/docs/en/dev/astrbot-config.md
index d33ba97123..78da3c461e 100644
--- a/docs/en/dev/astrbot-config.md
+++ b/docs/en/dev/astrbot-config.md
@@ -211,6 +211,12 @@ With BTW enabled, **MCP server loop assignments** selects conversation, work, or
 
 Assignments are saved per configuration profile. They control tool visibility and do not replace MCP read/write authorization or the existing connection, private-network, and redirect restrictions.
 
+## BTW Skill visibility
+
+With BTW enabled, **Skill loop assignments** chooses conversation, work, or both for each enabled ordinary Skill. Ordinary Skills default to both loops; choosing one loop saves an override, and choosing both removes it. Workspace Skills are available only to the work loop with the `local` runtime. Disabling BTW preserves the standard Skill selection path.
+
+Loop assignments narrow the enabled Skills before the request's Skill snapshot is frozen. The prompt, `read_skill`, and Skill-declared tool candidates therefore use the same selection. Persona and plugin restrictions still apply, including an empty Persona Skill list. A loop assignment never grants execution permission: `read_skill` can read permitted Skill manuals when Computer Use is `none`, while Shell and Python remain unavailable.
+
 ## SubAgents, speech, and knowledge base
 
 - `subagent_orchestrator.main_enable` enables handoffs.
diff --git a/docs/zh/dev/astrbot-config.md b/docs/zh/dev/astrbot-config.md
index 79d69c7967..70591e7eb4 100644
--- a/docs/zh/dev/astrbot-config.md
+++ b/docs/zh/dev/astrbot-config.md
@@ -213,6 +213,12 @@ API Key 属于敏感配置。不要把真实 `cmd_config.json`、截图、日志
 
 分配按配置档保存，只控制工具可见性，不替代 MCP 读写授权，也不改变现有连接、私网访问和重定向限制。
 
+## BTW Skill 循环可见性
+
+启用 BTW 后，可通过 **Skills 循环分配** 为每个已启用的普通 Skill 选择对话循环、工作循环或两者。普通 Skill 默认在两个循环可见；选择单一循环会保存覆盖，重新选择两者会移除覆盖。工作区 Skill 仅在使用 `local` 运行时的工作循环中可用。关闭 BTW 后保留标准 Skill 选择路径。
+
+循环分配在请求 Skill 快照冻结之前筛选已启用的 Skill，因此提示词、`read_skill` 和 Skill 声明的候选工具使用同一选择结果。Persona 与插件限制继续生效，包括 Persona 的空 Skill 列表。循环分配不会授予执行权限：Computer Use 为 `none` 时，`read_skill` 仍可读取允许的 Skill 手册，但 Shell 和 Python 仍不可用。
+
 ## 子代理、语音与知识库
 
 - `subagent_orchestrator.main_enable`：启用 handoff。
diff --git a/tests/unit/test_btw_skill_routes.py b/tests/unit/test_btw_skill_routes.py
new file mode 100644
index 0000000000..0ec8ddc115
--- /dev/null
+++ b/tests/unit/test_btw_skill_routes.py
@@ -0,0 +1,191 @@
+"""Skill loop assignments constrain the frozen request capabilities."""
+
+import json
+from pathlib import Path
+from types import SimpleNamespace
+from unittest.mock import MagicMock
+
+import pytest
+
+import astrbot.core.astr_main_agent as main_agent
+from astrbot.core.agent.llm_types import ProviderRequest
+from astrbot.core.agent.tool import FunctionTool
+from astrbot.core.config.astrbot_config import AstrBotConfig
+from astrbot.core.skills._skill_inventory import SkillInfo
+from astrbot.core.skills._skill_read import (
+    SkillReadError,
+    lookup_frozen_skill,
+    read_host_skill_file,
+)
+from astrbot.core.skills._skill_snapshot import SKILL_SNAPSHOT_EXTRA_KEY
+from astrbot.core.tool_catalog import ToolCatalogInputs, assemble_tool_catalog
+from astrbot.core.tools.computer_tools import ExecuteShellTool
+from astrbot.core.tools.skill_tools import ReadSkillTool
+
+
+def _skill(root: Path, name: str, *, workspace: bool = False) -> SkillInfo:
+    directory = root / ("workspace" if workspace else "ordinary") / name
+    directory.mkdir(parents=True)
+    path = directory / "SKILL.md"
+    path.write_text(
+        f"---\nname: {name}\ndescription: {name} manual\n"
+        f"tools:\n  - {name}_query\n  - astrbot_execute_shell\n---\n# {name}\n",
+        encoding="utf-8",
+    )
+    return SkillInfo(
+        name=name,
+        description=f"{name} manual",
+        path=str(path),
+        host_path=str(path),
+        active=True,
+        source_type="workspace" if workspace else "local_only",
+        declared_tools=(f"{name}_query", "astrbot_execute_shell"),
+    )
+
+
+@pytest.fixture
+def skill_context(tmp_path, monkeypatch):
+    ordinary = [_skill(tmp_path, name) for name in ("shared", "restricted")]
+    workspace = [_skill(tmp_path, "workspace-guide", workspace=True)]
+    manager = SimpleNamespace(
+        list_skills=MagicMock(return_value=ordinary),
+        list_workspace_skills=MagicMock(return_value=workspace),
+    )
+    profile = {"btw": {"enabled": True, "skill_routes": []}}
+    context = SimpleNamespace(
+        skill_manager=manager,
+        catalogs=SimpleNamespace(plugins=SimpleNamespace(all=lambda: [])),
+        get_config=lambda **_kwargs: profile,
+    )
+    extras = {}
+    event = SimpleNamespace(
+        unified_msg_origin="webchat:FriendMessage:test",
+        get_extra=lambda key, default=None: extras.get(key, default),
+        set_extra=lambda key, value: extras.__setitem__(key, value),
+    )
+    monkeypatch.setattr(
+        main_agent, "get_astrbot_workspaces_path", lambda: str(tmp_path)
+    )
+    return context, event, profile, extras
+
+
+@pytest.mark.parametrize(
+    ("enabled", "loop", "routes", "names"),
+    [
+        (True, None, [], {"shared", "restricted"}),
+        (True, "work", [], {"shared", "restricted"}),
+        (
+            True,
+            "conversation",
+            [{"skill_name": "restricted", "loop": "work"}],
+            {"shared"},
+        ),
+        (
+            True,
+            "work",
+            [{"skill_name": "restricted", "loop": "conversation"}],
+            {"shared"},
+        ),
+        (
+            False,
+            "conversation",
+            [{"skill_name": "restricted", "loop": "work"}],
+            {"shared", "restricted"},
+        ),
+        (
+            True,
+            "conversation",
+            [{"skill_name": "restricted", "loop": "invalid"}],
+            {"shared", "restricted"},
+        ),
+        (True, "conversation", {"restricted": "work"}, {"shared", "restricted"}),
+    ],
+)
+def test_skill_prompt_reader_and_tool_catalog_share_filtered_snapshot(
+    skill_context, enabled, loop, routes, names
+):
+    context, event, profile, extras = skill_context
+    profile["btw"] = {"enabled": enabled, "skill_routes": routes}
+    extras["btw_loop"] = loop
+    req = ProviderRequest(prompt="help")
+    snapshot = main_agent._append_skills_prompt(
+        req, {"computer_use_runtime": "none"}, None, event, context
+    )
+    assert snapshot is extras[SKILL_SNAPSHOT_EXTRA_KEY]
+    assert {skill.name for skill in snapshot.skills} == names
+    for name in ("shared", "restricted"):
+        assert (f"**{name}**" in req.system_prompt) == (name in names)
+        if name in names:
+            assert f"# {name}" in read_host_skill_file(
+                lookup_frozen_skill(snapshot, name), "SKILL.md"
+            )
+        else:
+            with pytest.raises(SkillReadError):
+                lookup_frozen_skill(snapshot, name)
+    tools = [ReadSkillTool(), ExecuteShellTool()] + [
+        FunctionTool(
+            name=f"{name}_query",
+            description=name,
+            parameters={"type": "object", "properties": {}},
+            required_actions=("session.read",),
+        )
+        for name in ("shared", "restricted")
+    ]
+    catalog = assemble_tool_catalog(
+        ToolCatalogInputs(
+            snapshot=snapshot,
+            persona_tools=None,
+            surface="im",
+            computer_use_runtime="none",
+            plugin_names=None,
+            registered_tools={tool.name: tool for tool in tools},
+            btw_config=profile["btw"],
+            loop_mode=loop,
+        )
+    )
+    assert set(catalog.names()) == {"read_skill"} | {f"{name}_query" for name in names}
+    context.skill_manager.list_workspace_skills.assert_not_called()
+
+
+@pytest.mark.parametrize(
+    ("enabled", "loop", "runtime", "persona", "workspace_visible"),
+    [
+        (True, "conversation", "local", None, False),
+        (True, None, "local", None, False),
+        (True, "work", "local", None, True),
+        (True, "work", "sandbox", None, False),
+        (True, "work", "none", None, False),
+        (False, "conversation", "local", None, True),
+        (True, "work", "local", {"skills": []}, False),
+    ],
+)
+def test_workspace_skill_requires_local_work_and_respects_persona(
+    skill_context, enabled, loop, runtime, persona, workspace_visible
+):
+    context, event, profile, extras = skill_context
+    profile["btw"]["enabled"] = enabled
+    extras["btw_loop"] = loop
+    snapshot = main_agent._append_skills_prompt(
+        ProviderRequest(prompt="help"),
+        {"computer_use_runtime": runtime},
+        persona,
+        event,
+        context,
+    )
+    assert (snapshot.get("workspace-guide") is not None) == workspace_visible
+    if persona == {"skills": []}:
+        assert snapshot.skills == ()
+
+
+def test_skill_routes_survive_profile_save(tmp_path):
+    path = tmp_path / "profile.json"
+    routes = [{"skill_name": "restricted", "loop": "work"}]
+    path.write_text(json.dumps({"btw": {"skill_routes": routes}}), encoding="utf-8")
+    config = AstrBotConfig(
+        config_path=str(path), default_config={"btw": {"skill_routes": []}}
+    )
+    config.save_config()
+    assert (
+        json.loads(path.read_text(encoding="utf-8-sig"))["btw"]["skill_routes"]
+        == routes
+    )