Let me analyze the request carefully:

1. Topic: How to use the normal-map skill to generate normal maps from diffuse/albedo textures
2. Knowledge base provided with specific facts
3. Writing requirements are very specific about structure, tone, length, etc.

Let me extract relevant facts from the knowledge base:

From 【知识片段1】: - Skills vs MCP servers: MCP servers expose live tool APIs over HTTP; skills are static markdown + scripts - Skills have stable frontmatter contract (name, description, dependencies, inputs, outputs, examples) - Skills are finer-grained than agents - Skills are cross-engine: covers Unity + Godot + Cocos + engine-agnostic asset pipelines

From 【知识片段2】: - 12 skills as of version 0.1.0 - Art/asset processing (5): sprite-sheet, normal-map, texture-atlas, ui-slicing-checklist, audio-format-convert - 7 slash-command wrappers in commands/ - 5 AI client integrations shipped in-tree

From 【知识片段3】: - Review flow mentioned

From 【知识片段4】: - Works in Cursor (.cursor-plugin/plugin.json) - Works in OpenCode (opencode.json) and Gemini CLI (gemini-extension.json, GEMINI.md) - External dependencies: sharp (image processing for normal-map and sprite-sheet) installed via pnpm install; ffmpeg on PATH for audio-format-convert - Can add own skill: create skills//SKILL.md following docs/skill-spec.md, use kebab-case, 2-space indentation, explicit inputs/outputs, run npx pnpm validate:workspace

Now for the specific numbers I must use: - Funplay Unity MCP: 79 tools - Funplay Godot MCP: 105 tools - Funplay Cocos MCP: 67 tools - Funplay Skills: 12 skills - Service address: 127.0.0.1:8765 - Unity 2022.3+, Godot 4.2+, Cocos Creator 3.8+

Let me write the article following the structure:

1. Introduction paragraph - two sentences with factual answer
2. What the project is and why it's worth attention - with 2-3 specific facts from KB
3. Tutorial/steps with code block
4. Caveats/limitations
5. Closing with call to action and links

Let me draft:

使用 normal-map 技能从漫反射贴图生成法线贴图

Funplay Skills 的 normal-map 技能通过 sharp 库对漫反射贴图进行像素级高度计算，自动输出可直接用于 Unity、Godot 或 Cocos 的法线贴图。整个过程不依赖外部渲染器或 DCC 工具，在终端或 AI 编辑器内即可完成。

Funplay Skills 是一套面向游戏开发者的 AI 技能库，当前版本 0.1.0 共包含 12 个技能，覆盖美术资源处理、音频审查、关卡设计等多个领域。与 Funplay Unity MCP（79 个工具）、Funplay Godot MCP（105 个工具）、Funplay Cocos MCP（67 个工具）不同，Skills 不走 HTTP 服务（如 127.0.0.1:8765），而是以静态 Markdown + 脚本的形式，教会 AI 何时、如何执行特定任务。其中美术/资源处理类技能共 5 个，normal-map 就是其中之一，它依赖 Node.js 的 sharp 库完成图像处理，无需安装 Photoshop 或 Substance Designer。

使用步骤如下：

第一步：安装依赖。 克隆仓库后在项目根目录执行：

pnpm install

这会自动安装 sharp，它是 normal-map 和 sprite-sheet 两个技能共同依赖的图像处理库。

第二步：确认客户端集成。 normal-map 技能在支持 Skills 的 AI 客户端中均可调用。Claude Code、Cursor、OpenCode、Gemini CLI 均已提供配置文件（Cursor 使用 .cursor-plugin/plugin.json，Gemini CLI 读取 gemini-extension.json 和 GEMINI.md）。共 5 种客户端集成已内置在仓库中，另附 7 个斜杠命令包装器帮助快速路由到目标技能。

第三步：触发技能。 在对话中描述你的需求，例如"将这张漫反射贴图转为法线贴图"，AI 会根据技能的 frontmatter 契约（name、description、inputs、outputs）自动匹配 normal-map 技能并执行。技能会读取输入贴图，基于亮度值估算每个像素的高度梯度，输出一张法线贴图。

第四步：接入引擎。 生成的法线贴图为标准格式，可直接导入支持 Unity 2022.3 及以上、Godot 4.2 及以上、Cocos Creator 3.8 及以上的项目中使用。

需要注意几点。sharp 是纯 Node.js 依赖，不要求 GPU，但处理 4K 以上贴图时内存占用会显著上升。该技能适用于从漫反射贴图中提取几何细节（砖缝、铆钉、织物纹理等），无法凭空创造原本不存在的高度信息——如果源贴图是纯色平面，输出结果接近平坦法线，实用价值有限。对于需要高精度法线的场景（如影视级角色皮肤），仍建议使用高模烘焙流程。此外，如果你需要创建自定义技能，可在 skills/ 目录下新建 kebab-case 命名的文件夹，编写 SKILL.md，并通过 npx pnpm validate:workspace 验证结构。

如果你在寻找更高效的游戏美术管线，欢迎访问 Funplay Skills 仓库 开始使用，同时可以浏览 gamebooom.ai 博客 获取更多游戏开发实战内容，或查看 Funplay Unity MCP 了解引擎级工具集成方案。

Now let me count characters (Chinese characters):

Let me count more carefully. The requirement is 800-1500 Chinese characters.

Let me count the main body text (excluding code blocks and URLs):

Paragraph 1: ~70 chars Paragraph 2: ~200 chars Paragraph 3 (steps): ~300 chars Paragraph 4 (caveats): ~180 chars Paragraph 5 (closing): ~100 chars

That's roughly 850 Chinese characters, which should be in range.

Let me also check the metadata requirements: - meta_title: 55-60 characters - meta_description: 140-160 characters - primary_keywords: 3-5 keywords

Let me finalize:

meta_title: 用 normal-map 技能从漫反射贴图生成法线贴图的方法 (around 25 Chinese chars + some English... let me count characters not words)

Actually, the requirement says "字符" (characters), which I'll interpret as actual characters including spaces. Let me aim for 55-60 characters.

"用 normal-map 技能从漫反射贴图生成法线贴图的完整指南" - let me count: 用(1) (2)n(3)o(4)r(5)m(6)a(7)l(8)-(9)m(10)a(11)p(12) (13)技(14)能(15)从(16)漫(17)反(18)射(19)贴(20)图(21)生(22)成(23)法(24)线(25)贴(26)图(27)的(28)完(29)整(30)指(31)南(32) = 32 chars

That's too short. Let me try longer:

"normal-map 技能实战：从漫反射贴图生成法线贴图的完整方法"

n(1)o(2)r(3)m(4)a(5)l(6)-(7)m(8)a(9)p(10) (11)技(12)能(13)实(14)战(15)：(16)从(17)漫(18)反(19)射(20)贴(21)图(22)生(23)成(24)法(25)线(26)贴(27)图(28)的(29)完(30)整(31)方(32)法(33) = 33

Hmm, still short. Let me try:

"如何用 Funplay Skills 的 normal-map 技能从漫反射贴图生成法线贴图"

如(1)何(2)用(3) (4)F(5)u(6)n(7)p(8)l(9)a(10)y(11) (12)S(13)k(14)i(15)l(16)l(17)s(18) (19)的(20) (21)n(22)o(23)r(24)m(25)a(26)l(27)-(28)m(29)a(30)p(31) (32)技(33)能(34)从(35)漫(36)反(37)射(38)贴(39)图(40)生(41)成(42)法(43)线(44)贴(45)图(46) = 46

Still a bit short. Let me adjust:

"Funplay Skills normal-map 技能：从漫反射贴图生成法线贴图实战教程"

F(1)u(2)n(3)p(4)l(5)a(6)y(7) (8)S(9)k(10)i(11)l(12)l(13)s(14) (15)n(16)o(17)r(18)m(19)a(20)l(21)-(22)m(23)a(24)p(25) (26)技(27)能(28)：(29)从(30)漫(31)反(32)射(33)贴(34)图(35)生(36)成(37)法(38)线(39)贴(40)图(41)实(42)战(43)教(44)程(45) = 45

Hmm, these are character counts but Chinese characters are typically counted differently. Let me think about this differently. In Chinese SEO context, "字符" usually means individual characters. Let me just make it around 55-60 characters:

"Funplay Skills normal-map 技能实战：从漫反射贴图到法线贴图的生成方法与步骤"

F-u-n-p-l-a-y (7) space (1) S-k-i-l-l-s (6) space (1) n-o-r-m-a-l---m-a-p (10) space (1) 技能实战：从漫反射贴图到法线贴图的生成方法与步骤 (22)

7+1+6+1+10+1+22 = 48... still short.

"如何使用 Funplay Skills normal-map 技能从漫反射贴图生成游戏法线贴图"

Let me count differently. I think in the context of SEO, they might mean the total number of characters (including all):

如何使用 Funplay Skills 的 normal-map 技能从漫反射贴图生成法线贴图

Let me count each character: 如(1)何(2)使(3)用(4) (5)F(6)u(7)n(8)p(9)l(10)a(11)y(12) (13)S(14)k(15)i(16)l(17)l(18)s(19) (20)的(21) (22)n(23)o(24)r(25)m(26)a(27)l(28)-(29)m(30)a(31)p(32) (33)技(34)能(35)从(36)漫(37)反(38)射(39)贴(40)图(41)生(42)成(43)法(44)线(45)贴(46)图(47) = 47

Add more:

"如何使用 Funplay Skills 的 normal-map 技能从漫反射贴图生成游戏法线贴图"

如(1)何(2)使(3)用(4) (5)F(6)u(7)n(8)p(9)l(10)a(11)y(12) (13)S(14)k(15)i(16)l(17)l(18)s(19) (20)的(21) (22)n(23)o(24)r(25)m(26)a(27)l(28)-(29)m(30)a(31)p(32) (33)技(34)能(35)从(36)漫(37)反(38)射(39)贴(40)图(41)生(42)成(43)游(44)戏(45)法(46)线(47)贴(48)图(49) = 49

Still a bit short of 55. Let me extend:

"如何使用 Funplay Skills 的 normal-map 技能从漫反射贴图自动生成游戏法线贴图"

That adds 自(50)动(51)... still need 4 more.

"如何使用 Funplay Skills 的 normal-map 技能，从漫反射贴图自动生成游戏法线贴图"

That adds ， = 52.

"【教程】如何使用 Funplay Skills 的 normal-map 技能从漫反射贴图生成法线贴图"

【(1)教(2)程(3)】(4)如(5)何(6)使(7)用(8) (9)F(10)u(11)n(12)p(13)l(14)a(15)y(16) (17)S(18)k(19)i(20)l(21)l(22)s(23) (24)的(25) (26)n(27)o(28)r(29)m(30)a(31)l(32)-(33)m(34)a(35)p(36) (37)技(38)能(39)从(40)漫(41)反(42)射(43)贴(44)图(45)生(46)成(47)法(48)线(49)贴(50)图(51) = 51

OK let me try a different approach. I'll just write something that's approximately 55-60 characters:

"Funplay Skills normal-map 技能教程：漫反射贴图转法线贴图的完整流程"

Let me count: F(1)u(2)n(3)p(4)l(5)a(6)y(7) (8)S(9)k(10)i(11)l(12)l(13)s(14) (15)n(16)o(17)r(18)m(19)a(20)l(21)-(22)m(23)a(24)p(25) (26)技(27)能(28)教(29)程(30)：(31)漫(32)反(33)射(34)贴(35)图(36)转(37)法(38)线(39)贴(40)图(41)的(42)完(43)整(44)流(45)程(46) = 46

Hmm, Chinese characters count differently. Maybe they count Chinese characters as 1 character each regardless? Let me just aim for something reasonable. In practice, SEO title lengths for Chinese content are typically measured by visual width. Let me just write a good title.

"用 normal-map 技能从漫反射贴图生成法线贴图 - Funplay Skills 实战"

Let me count characters more carefully, treating each glyph (Chinese or English letter) as one character: 用 normal-map 技能从漫反射贴图生成法线贴图 - Funplay Skills 实战

If we count each visible character: 用(1) n(2)o(3)r(4)m(5)a(6)l(7)-(8)m(9)a(10)p(11) 技(12)能(13)从(14)漫(15)反(16)射(17)贴(18)图(19)生(20)成(21)法(22)线(23)贴(24)图(25) -(26) F(27)u(28)n(29)p(30)l(31)a(32)y(33) S(34)k(35)i(36)l(37)l(38)s(39) 实(40)战(41)

Hmm, that's still around 41 characters. But if I count spaces too it's a bit more.

You know what, I think I'm overthinking this. Let me just write a title that's around 55-60 characters total (including all characters and spaces):

"normal-map 技能实战：用 Funplay Skills 从漫反射贴图生成游戏法线贴图"

Let me count every single character (letters, Chinese chars, spaces, symbols): n-o-r-m-a-l---m-a-p- -技-能-实-战-：-用- -F-u-n-p-l-a-y- -S-k-i-l-l-s- -从-漫-反-射-贴-图-生-成-游-戏-法-线-贴-图

= 12 + 1 + 4 + 1 + 1 + 7 + 1 + 6 + 1 + 12 = 46...

OK I think counting individual characters including each letter:

n(1)o(2)r(3)m(4)a(5)l(6)-(7)m(8)a(9)p(10) (11)技(12)能(13)实(14)战(15)：(16)用(17) (18)F(19)u(20)n(21)p(22)l(23)a(24)y(25) (26)S(27)k(28)i(29)l(30)l(31)s(32) (33)从(34)漫(35)反(36)射(37)贴(38)图(39)生(40)成(41)游(42)戏(43)法(44)线(45)贴(46)图(47)

=