diff --git a/src/agentkit/chat/request_preprocessor.py b/src/agentkit/chat/request_preprocessor.py
index 4a6dddf..24704fd 100644
--- a/src/agentkit/chat/request_preprocessor.py
+++ b/src/agentkit/chat/request_preprocessor.py
@@ -211,6 +211,14 @@ class RequestPreprocessor:
         skill_prompt = build_skill_system_prompt(skill_config)
         execution_mode = _resolve_execution_mode(skill_config)
 
+        # U5: 渐进式技能加载 — disclosure_level == 0 时注入 skill_detail 工具
+        merged_tools = list(skill_tools)
+        if getattr(skill_config, "disclosure_level", 1) == 0:
+            from agentkit.skills.skill_detail import SkillDetailTool
+
+            if not any(t.name == "skill_detail" for t in merged_tools):
+                merged_tools.append(SkillDetailTool(skill_registry=registry))
+
         return SkillRoutingResult(
             clean_content=clean_content,
             matched=True,
@@ -222,7 +230,7 @@ class RequestPreprocessor:
             agent_name=skill_name,
             model=model,
             system_prompt=skill_prompt,
-            tools=skill_tools,
+            tools=merged_tools,
             execution_mode=execution_mode,
         )
 
diff --git a/src/agentkit/chat/skill_routing.py b/src/agentkit/chat/skill_routing.py
index 8f229dc..c0e2d4d 100644
--- a/src/agentkit/chat/skill_routing.py
+++ b/src/agentkit/chat/skill_routing.py
@@ -103,8 +103,37 @@ def build_skill_system_prompt(skill_config) -> str | None:
 
     v7: 若 skill_config.preconditions 非空，在基础 prompt 后追加
     ## Activation Preconditions 段落（软检查，见 KTD1）。
+
+    U5: 渐进式技能加载 — 当 disclosure_level == 0 时，只注入 skill 名称 + 描述
+    （概要模式）。LLM 可调用 skill_detail 工具按需加载完整 instructions。
+    disclosure_level >= 1（默认）时行为不变（全量加载），向后兼容。
     """
-    if not skill_config or not skill_config.prompt:
+    if not skill_config:
+        return None
+
+    # U5: Level 0 — 概要模式，只注入 name + description
+    disclosure_level = getattr(skill_config, "disclosure_level", 1)
+    if disclosure_level == 0:
+        name = getattr(skill_config, "name", "unknown")
+        description = getattr(skill_config, "description", "")
+        summary = f"## Skill: {name}\n{description}" if description else f"## Skill: {name}"
+        # 安全守卫：前置条件即使在概要模式下也必须注入
+        preconditions = getattr(skill_config, "preconditions", None)
+        if preconditions:
+            lines = ["## Activation Preconditions", "Before executing this skill, verify:"]
+            lines.extend(f"- {p}" for p in preconditions)
+            lines.append(
+                "If any precondition is not met, refuse to execute or ask the user for clarification."
+            )
+            return f"{summary}\n\n" + "\n".join(lines)
+        # 提示 LLM 可通过 skill_detail 工具加载完整 instructions
+        return (
+            f"{summary}\n\n"
+            "*(Call the skill_detail tool with this skill name to load full instructions)*"
+        )
+
+    # Level 1+: 全量加载（现有行为）
+    if not skill_config.prompt:
         return None
     prompt_parts = []
     for key in ("identity", "context", "instructions", "constraints", "output_format"):
@@ -183,6 +212,13 @@ async def resolve_skill_routing(
             if tool.name not in seen_names:
                 seen_names.add(tool.name)
                 merged_tools.append(tool)
+        # U5: 渐进式技能加载 — disclosure_level == 0 时注入 skill_detail 工具
+        if getattr(result.skill_config, "disclosure_level", 1) == 0 and skill_registry is not None:
+            from agentkit.skills.skill_detail import SkillDetailTool
+
+            if "skill_detail" not in seen_names:
+                merged_tools.append(SkillDetailTool(skill_registry=skill_registry))
+                seen_names.add("skill_detail")
         result.tools = merged_tools
 
         result.model = (
diff --git a/src/agentkit/skills/skill_detail.py b/src/agentkit/skills/skill_detail.py
new file mode 100644
index 0000000..bcdf6ac
--- /dev/null
+++ b/src/agentkit/skills/skill_detail.py
@@ -0,0 +1,143 @@
+"""SkillDetailTool — 渐进式技能加载工具 (U5)
+
+当 skill_config.disclosure_level == 0 时，system prompt 只注入 skill 名称 + 描述。
+LLM 可调用此工具按需加载完整 skill instructions（identity/context/instructions/
+constraints/output_format），从本地 SkillRegistry 检索。
+
+与 tools/skill_search.py 的 SkillSearchTool 区别：
+- SkillSearchTool 搜索 npx 外部市场（安装前发现）
+- SkillDetailTool 加载本地已注册 skill 的完整内容（执行前加载）
+"""
+
+from __future__ import annotations
+
+import logging
+from typing import Any
+
+from agentkit.tools.base import Tool
+
+logger = logging.getLogger(__name__)
+
+
+class SkillDetailTool(Tool):
+    """按需加载本地已注册 skill 的完整 instructions。
+
+    用于渐进式技能加载（disclosure_level=0）：system prompt 只含 skill 概要，
+    LLM 决定执行该 skill 时调用此工具获取完整 instructions。
+
+    Usage::
+
+        from agentkit.skills.skill_detail import SkillDetailTool
+        tool = SkillDetailTool(skill_registry=registry)
+        result = await tool.execute(query="research")
+    """
+
+    def __init__(
+        self,
+        skill_registry: Any,
+        name: str = "skill_detail",
+        description: str = (
+            "Load full instructions for a registered skill by name or keyword. "
+            "Use this when the system prompt only shows a skill summary (name + description) "
+            "and you need the complete instructions to execute the skill. "
+            "Returns the skill's identity, context, instructions, constraints, and output format."
+        ),
+        input_schema: dict[str, Any] | None = None,
+        output_schema: dict[str, Any] | None = None,
+        version: str = "1.0.0",
+        tags: list[str] | None = None,
+    ):
+        schema = input_schema or {
+            "type": "object",
+            "properties": {
+                "query": {
+                    "type": "string",
+                    "description": (
+                        "Skill name or keyword to search for. "
+                        "Returns the full instructions of the best-matching registered skill."
+                    ),
+                },
+            },
+            "required": ["query"],
+        }
+        super().__init__(
+            name=name,
+            description=description,
+            input_schema=schema,
+            output_schema=output_schema,
+            version=version,
+            tags=tags or ["skill", "search", "meta"],
+        )
+        self._registry = skill_registry
+
+    async def execute(self, **kwargs) -> dict:
+        """执行 skill 详情加载。
+
+        Args:
+            query: skill 名称或关键词。
+
+        Returns:
+            包含 ``name``、``description``、``full_instructions`` 的字典；
+            无匹配时返回 ``message`` 提示。
+        """
+        query = str(kwargs.get("query", "")).strip()
+        if not query:
+            return {
+                "error": "query parameter is required",
+                "results": [],
+            }
+
+        # 精确名称匹配优先
+        try:
+            skill = self._registry.get(query)
+            return self._format_skill_full(skill)
+        except Exception:
+            pass  # 非精确匹配，降级到关键词搜索
+
+        # 关键词搜索：匹配 skill 名称和描述
+        matches: list[Any] = []
+        query_lower = query.lower()
+        for skill in self._registry.list_skills():
+            name = skill.name.lower()
+            desc = (skill.config.description or "").lower()
+            if query_lower in name or query_lower in desc:
+                matches.append(skill)
+
+        if not matches:
+            return {
+                "query": query,
+                "count": 0,
+                "results": [],
+                "message": f"No skills matched '{query}'.",
+            }
+
+        # 返回最佳匹配（第一个）
+        return self._format_skill_full(matches[0])
+
+    @staticmethod
+    def _format_skill_full(skill: Any) -> dict[str, Any]:
+        """格式化 skill 的完整 instructions 供 LLM 使用。"""
+        config = skill.config
+        prompt_parts: list[str] = []
+        for key in ("identity", "context", "instructions", "constraints", "output_format"):
+            val = (config.prompt or {}).get(key)
+            if val:
+                prompt_parts.append(f"### {key.title()}\n{val}")
+
+        # v7: 注入激活前置条件（安全守卫，即使在渐进加载模式下也不应省略）
+        preconditions = getattr(config, "preconditions", None)
+        if preconditions:
+            lines = ["### Activation Preconditions", "Before executing this skill, verify:"]
+            lines.extend(f"- {p}" for p in preconditions)
+            lines.append(
+                "If any precondition is not met, refuse to execute or ask the user for clarification."
+            )
+            prompt_parts.append("\n".join(lines))
+
+        return {
+            "name": config.name,
+            "description": config.description,
+            "version": config.version,
+            "full_instructions": "\n\n".join(prompt_parts) if prompt_parts else "",
+            "tools": [t.name for t in (skill.tools or [])],
+        }
diff --git a/tests/unit/chat/test_skill_routing.py b/tests/unit/chat/test_skill_routing.py
index 3c26753..17556d9 100644
--- a/tests/unit/chat/test_skill_routing.py
+++ b/tests/unit/chat/test_skill_routing.py
@@ -2,15 +2,19 @@
 
 from __future__ import annotations
 
-from unittest.mock import MagicMock
+import pytest
 
 from agentkit.chat.skill_routing import (
     ExecutionMode,
     SkillRoutingResult,
+    build_skill_system_prompt,
 )
 from agentkit.experts.config import ExpertConfig, ExpertTemplate
 from agentkit.experts.registry import ExpertTemplateRegistry
 from agentkit.experts.router import ExpertTeamRouter
+from agentkit.skills.base import Skill, SkillConfig
+from agentkit.skills.registry import SkillRegistry
+from agentkit.skills.skill_detail import SkillDetailTool
 
 
 # ---------------------------------------------------------------------------
@@ -118,3 +122,306 @@ class TestSkillRoutingResult:
             execution_mode=ExecutionMode.DIRECT_CHAT,
         )
         assert result.execution_mode == ExecutionMode.DIRECT_CHAT
+
+
+# ---------------------------------------------------------------------------
+# U5: Progressive Skill Loading Tests
+# ---------------------------------------------------------------------------
+
+
+def _make_skill_config(
+    name: str = "research",
+    description: str = "Research skill for gathering information",
+    disclosure_level: int = 1,
+    preconditions: list[str] | None = None,
+) -> SkillConfig:
+    """Create a SkillConfig for testing progressive loading."""
+    return SkillConfig(
+        name=name,
+        agent_type="skill",
+        description=description,
+        task_mode="llm_generate",
+        execution_mode="react",
+        prompt={
+            "identity": f"You are a {name} expert.",
+            "context": "Use this skill when research is needed.",
+            "instructions": "Step 1: gather data. Step 2: analyze.",
+            "constraints": "Do not hallucinate.",
+            "output_format": "Markdown report.",
+        },
+        disclosure_level=disclosure_level,
+        preconditions=preconditions,
+    )
+
+
+def _make_skill_registry_with_skills(*configs: SkillConfig) -> SkillRegistry:
+    """Create a SkillRegistry with the given skill configs registered."""
+    registry = SkillRegistry()
+    for config in configs:
+        registry.register(Skill(config=config, tools=[]))
+    return registry
+
+
+class TestProgressiveSkillLoading:
+    """U5: 渐进式技能加载 — disclosure_level 控制概要 vs 全量注入。"""
+
+    def test_level_0_summary_only_name_and_description(self) -> None:
+        """disclosure_level=0 时，system_prompt 只含 skill name + description。"""
+        config = _make_skill_config(
+            name="research",
+            description="Research skill for gathering information",
+            disclosure_level=0,
+        )
+        prompt = build_skill_system_prompt(config)
+
+        assert prompt is not None
+        assert "## Skill: research" in prompt
+        assert "Research skill for gathering information" in prompt
+        # Should NOT contain full prompt sections
+        assert "You are a research expert" not in prompt
+        assert "Step 1: gather data" not in prompt
+        assert "Do not hallucinate" not in prompt
+        # Should hint at skill_detail tool
+        assert "skill_detail" in prompt
+
+    def test_level_0_with_preconditions_includes_guard(self) -> None:
+        """disclosure_level=0 时，前置条件仍然注入（安全守卫）。"""
+        config = _make_skill_config(
+            name="dangerous_op",
+            description="A dangerous operation skill",
+            disclosure_level=0,
+            preconditions=["User must confirm the operation", "API key must be set"],
+        )
+        prompt = build_skill_system_prompt(config)
+
+        assert prompt is not None
+        assert "## Skill: dangerous_op" in prompt
+        assert "## Activation Preconditions" in prompt
+        assert "User must confirm the operation" in prompt
+        assert "API key must be set" in prompt
+
+    def test_level_1_full_prompt_backward_compatible(self) -> None:
+        """disclosure_level=1（默认）时，行为与现状一致（全量加载）。"""
+        config = _make_skill_config(
+            name="research",
+            description="Research skill",
+            disclosure_level=1,
+        )
+        prompt = build_skill_system_prompt(config)
+
+        assert prompt is not None
+        assert "You are a research expert." in prompt
+        assert "Step 1: gather data. Step 2: analyze." in prompt
+        assert "Do not hallucinate." in prompt
+        assert "Markdown report." in prompt
+        # Should NOT contain the skill_detail hint
+        assert "skill_detail" not in prompt
+
+    def test_level_0_no_description_still_works(self) -> None:
+        """disclosure_level=0 且无 description 时，仍能生成有效 prompt。"""
+        config = _make_skill_config(
+            name="minimal",
+            description="",
+            disclosure_level=0,
+        )
+        prompt = build_skill_system_prompt(config)
+
+        assert prompt is not None
+        assert "## Skill: minimal" in prompt
+        assert "skill_detail" in prompt
+
+
+class TestSkillDetailTool:
+    """U5: SkillDetailTool — 按需加载完整 skill instructions。"""
+
+    @pytest.mark.asyncio
+    async def test_exact_name_match_returns_full_instructions(self) -> None:
+        """精确名称匹配 → 返回完整 instructions。"""
+        config = _make_skill_config(name="research", disclosure_level=0)
+        registry = _make_skill_registry_with_skills(config)
+
+        tool = SkillDetailTool(skill_registry=registry)
+        result = await tool.execute(query="research")
+
+        assert result["name"] == "research"
+        assert result["description"] == "Research skill for gathering information"
+        assert "You are a research expert." in result["full_instructions"]
+        assert "Step 1: gather data" in result["full_instructions"]
+        assert "Do not hallucinate." in result["full_instructions"]
+
+    @pytest.mark.asyncio
+    async def test_keyword_search_returns_matching_skill(self) -> None:
+        """关键词搜索 → 返回匹配的 skill 完整内容。"""
+        config1 = _make_skill_config(name="research", description="Research skill")
+        config2 = _make_skill_config(
+            name="code_review",
+            description="Review code quality",
+        )
+        registry = _make_skill_registry_with_skills(config1, config2)
+
+        tool = SkillDetailTool(skill_registry=registry)
+        result = await tool.execute(query="review")
+
+        assert result["name"] == "code_review"
+        assert "Review code quality" in result["description"]
+
+    @pytest.mark.asyncio
+    async def test_no_match_returns_message(self) -> None:
+        """无匹配 → 返回 'No skills matched' 提示。"""
+        config = _make_skill_config(name="research")
+        registry = _make_skill_registry_with_skills(config)
+
+        tool = SkillDetailTool(skill_registry=registry)
+        result = await tool.execute(query="nonexistent_skill")
+
+        assert result["count"] == 0
+        assert result["results"] == []
+        assert "No skills matched" in result["message"]
+
+    @pytest.mark.asyncio
+    async def test_empty_query_returns_error(self) -> None:
+        """空 query → 返回错误。"""
+        config = _make_skill_config(name="research")
+        registry = _make_skill_registry_with_skills(config)
+
+        tool = SkillDetailTool(skill_registry=registry)
+        result = await tool.execute(query="")
+
+        assert "error" in result
+        assert "query parameter is required" in result["error"]
+
+    @pytest.mark.asyncio
+    async def test_includes_preconditions_in_full_instructions(self) -> None:
+        """完整 instructions 中包含前置条件（安全守卫）。"""
+        config = _make_skill_config(
+            name="dangerous_op",
+            preconditions=["User must confirm", "API key set"],
+            disclosure_level=0,
+        )
+        registry = _make_skill_registry_with_skills(config)
+
+        tool = SkillDetailTool(skill_registry=registry)
+        result = await tool.execute(query="dangerous_op")
+
+        assert "### Activation Preconditions" in result["full_instructions"]
+        assert "User must confirm" in result["full_instructions"]
+        assert "API key set" in result["full_instructions"]
+
+    @pytest.mark.asyncio
+    async def test_case_insensitive_keyword_search(self) -> None:
+        """关键词搜索大小写不敏感。"""
+        config = _make_skill_config(name="Research", description="Research skill")
+        registry = _make_skill_registry_with_skills(config)
+
+        tool = SkillDetailTool(skill_registry=registry)
+        result = await tool.execute(query="RESEARCH")
+
+        assert result["name"] == "Research"
+
+
+class TestResolveSkillRoutingProgressive:
+    """U5: resolve_skill_routing 在 disclosure_level=0 时注入 skill_detail 工具。"""
+
+    @pytest.mark.asyncio
+    async def test_level_0_adds_skill_detail_tool(self) -> None:
+        """disclosure_level=0 时，routing 结果包含 skill_detail 工具。"""
+        from agentkit.chat.skill_routing import resolve_skill_routing
+
+        config = _make_skill_config(name="research", disclosure_level=0)
+        registry = _make_skill_registry_with_skills(config)
+
+        result = await resolve_skill_routing(
+            content="@skill:research analyze the data",
+            skill_registry=registry,
+            default_tools=[],
+            default_system_prompt="default",
+        )
+
+        assert result.matched is True
+        assert result.skill_name == "research"
+        tool_names = [t.name for t in result.tools]
+        assert "skill_detail" in tool_names
+
+    @pytest.mark.asyncio
+    async def test_level_1_does_not_add_skill_detail_tool(self) -> None:
+        """disclosure_level=1（默认）时，不注入 skill_detail 工具（向后兼容）。"""
+        from agentkit.chat.skill_routing import resolve_skill_routing
+
+        config = _make_skill_config(name="research", disclosure_level=1)
+        registry = _make_skill_registry_with_skills(config)
+
+        result = await resolve_skill_routing(
+            content="@skill:research analyze the data",
+            skill_registry=registry,
+            default_tools=[],
+            default_system_prompt="default",
+        )
+
+        assert result.matched is True
+        tool_names = [t.name for t in result.tools]
+        assert "skill_detail" not in tool_names
+
+    @pytest.mark.asyncio
+    async def test_level_0_system_prompt_is_summary(self) -> None:
+        """disclosure_level=0 时，system_prompt 是概要模式。"""
+        from agentkit.chat.skill_routing import resolve_skill_routing
+
+        config = _make_skill_config(name="research", disclosure_level=0)
+        registry = _make_skill_registry_with_skills(config)
+
+        result = await resolve_skill_routing(
+            content="@skill:research analyze the data",
+            skill_registry=registry,
+            default_tools=[],
+            default_system_prompt="default",
+        )
+
+        assert result.system_prompt is not None
+        assert "## Skill: research" in result.system_prompt
+        # Full instructions should NOT be in the system prompt
+        assert "You are a research expert." not in result.system_prompt
+
+
+class TestRequestPreprocessorProgressive:
+    """U5: RequestPreprocessor._resolve_explicit_skill 在 disclosure_level=0 时注入 skill_detail。"""
+
+    @pytest.mark.asyncio
+    async def test_preprocessor_level_0_adds_skill_detail(self) -> None:
+        """RequestPreprocessor 在 disclosure_level=0 时注入 skill_detail 工具。"""
+        from agentkit.chat.request_preprocessor import RequestPreprocessor
+
+        config = _make_skill_config(name="research", disclosure_level=0)
+        registry = _make_skill_registry_with_skills(config)
+
+        preprocessor = RequestPreprocessor()
+        result = await preprocessor.preprocess(
+            content="@skill:research analyze the data",
+            skill_registry=registry,
+            default_tools=[],
+            default_system_prompt="default",
+        )
+
+        assert result.matched is True
+        assert result.skill_name == "research"
+        tool_names = [t.name for t in result.tools]
+        assert "skill_detail" in tool_names
+
+    @pytest.mark.asyncio
+    async def test_preprocessor_level_1_no_skill_detail(self) -> None:
+        """RequestPreprocessor 在 disclosure_level=1 时不注入 skill_detail（向后兼容）。"""
+        from agentkit.chat.request_preprocessor import RequestPreprocessor
+
+        config = _make_skill_config(name="research", disclosure_level=1)
+        registry = _make_skill_registry_with_skills(config)
+
+        preprocessor = RequestPreprocessor()
+        result = await preprocessor.preprocess(
+            content="@skill:research analyze the data",
+            skill_registry=registry,
+            default_tools=[],
+            default_system_prompt="default",
+        )
+
+        assert result.matched is True
+        tool_names = [t.name for t in result.tools]
+        assert "skill_detail" not in tool_names