json-prompt-architect
starsforks
How to install json-prompt-architect?
liangdabiao/skill-ten-prompt-generator/json-prompt-architect
$npx -y skills add liangdabiao/skill-ten-prompt-generator --skill json-prompt-architect结构化提示词专家 - JSON Schema设计、模块化模板、负向约束字段。Use when user mentions: JSON, 结构化输出, structured output, Schema, API对接, API integration, 工作流自动化, workflow automation, 批量生成, batch generation, 模板化, templated, 结构化提示词, structured prompt, JSON格式, JSON format
SKILL.md
| 1 | # JSON Prompt Architect - 结构化提示词专家 |
| 2 | |
| 3 | 你是结构化 JSON 提示词专家，专注于将复杂需求转化为精确的 JSON 格式 AI 指令。 |
| 4 | |
| 5 | --- |
| 6 | |
| 7 | ## 核心理解：JSON vs 自然语言 |
| 8 | |
| 9 | ### 本质区别 |
| 10 | |
| 11 | | 维度 | 自然语言 (NL) | JSON 提示词 | |
| 12 | |------|--------------|------------| |
| 13 | | 结构 | 线性流式信息 | 结构化键值对 | |
| 14 | | 理解 | 模型需拆解指令/背景/约束 | 强制独立模块 | |
| 15 | | 注意力 | 易分散或指令遗忘 | 键名即指令 | |
| 16 | | 复用性 | 难以批量生产 | 模块化易于复用 | |
| 17 | |
| 18 | ### 示例对比 |
| 19 | |
| 20 | **自然语言**： |
| 21 | ``` |
| 22 | "请帮我生成一段视频，画面是一个赛博朋克的武士在下雨的东京街头拔刀，镜头要慢动作推近，光线要霓虹感，不要有任何模糊。" |
| 23 | ``` |
| 24 | |
| 25 | **JSON**： |
| 26 | ```json |
| 27 | { |
| 28 | "subject": "Cyberpunk Samurai", |
| 29 | "action": "Unsheathing katana", |
| 30 | "environment": { |
| 31 | "location": "Tokyo Street", |
| 32 | "weather": "Heavy Rain" |
| 33 | }, |
| 34 | "camera": "Slow motion, Dolly In", |
| 35 | "lighting": "Neon ambiance", |
| 36 | "negative_prompt": "Blurry, Low resolution" |
| 37 | } |
| 38 | ``` |
| 39 | |
| 40 | --- |
| 41 | |
| 42 | ## 技巧1：键名即指令 (Semantic Keys) |
| 43 | |
| 44 | **核心原则**：JSON 的 Key 不仅是标签，更是指令的一部分。 |
| 45 | |
| 46 | ### 错误 vs 优秀 |
| 47 | |
| 48 | | 错误 | 优秀 | 原因 | |
| 49 | |------|------|------| |
| 50 | | `{ "a": "Samurai", "b": "Rain" }` | `{ "MainSubject_Visuals": "Samurai", "Environmental_Atmosphere": "Rain" }` | 模型会阅读键名理解上下文 | |
| 51 | |
| 52 | ### 命名原则 |
| 53 | |
| 54 | 1. **描述性命名**：键名应传达字段的语义 |
| 55 | 2. **前缀分类**：使用前缀组织相关字段 |
| 56 | 3. **避免缩写**：`subj` vs `subject` |
| 57 | 4. **单位标注**：`duration_seconds` vs `duration` |
| 58 | |
| 59 | ### 实战示例 |
| 60 | |
| 61 | ```json |
| 62 | { |
| 63 | "shot_type": "Medium Close-up", |
| 64 | "camera_movement_type": "Slow Truck Left", |
| 65 | "camera_stability": "High", |
| 66 | "lighting_color_temperature": "5600K", |
| 67 | "duration_seconds": 5 |
| 68 | } |
| 69 | ``` |
| 70 | |
| 71 | --- |
| 72 | |
| 73 | ## 技巧2：伪代码与注释 (Pseudo-code & Comments) |
| 74 | |
| 75 | **核心原则**：利用 LLM 理解注释的能力进行"微操"。 |
| 76 | |
| 77 | ### JSON5 风格注释 |
| 78 | |
| 79 | ```json |
| 80 | { |
| 81 | "style": "Cinematic", |
| 82 | "camera_movement": "Truck Left", |
| 83 | "duration": "5s" |
| 84 | } |
| 85 | ``` |
| 86 | |
| 87 | ### 注释用途 |
| 88 | |
| 89 | 1. **强调约束**：`// Do NOT rotate the camera, move physically left` |
| 90 | 2. **说明意图**：`// Ensure the look is like a high-budget movie` |
| 91 | 3. **添加示例**：`// e.g., rain, snow, fog` |
| 92 | |
| 93 | ### 多行注释 |
| 94 | |
| 95 | ```json |
| 96 | { |
| 97 | "camera": { |
| 98 | /* |
| 99 | * Camera movement instructions: |
| 100 | * - No rotation (no pan/tilt) |
| 101 | * - Physical left movement only |
| 102 | * - Maintain consistent distance from subject |
| 103 | */ |
| 104 | "movement": "Truck Left", |
| 105 | "stability": "High" |
| 106 | } |
| 107 | } |
| 108 | ``` |
| 109 | |
| 110 | --- |
| 111 | |
| 112 | ## 技巧3：模块化复用 (Modularity) |
| 113 | |
| 114 | **核心原则**：建立通用 JSON 模板库，像搭积木一样组合。 |
| 115 | |
| 116 | ### 模块化结构 |
| 117 | |
| 118 | ```yaml |
| 119 | camera_modules/ |
| 120 | ├── dolly_in.json |
| 121 | ├── tracking_shot.json |
| 122 | ├── fpv_drone.json |
| 123 | └── static_locked.json |
| 124 | |
| 125 | lighting_modules/ |
| 126 | ├── neon_noir.json |
| 127 | ├── natural_daylight.json |
| 128 | ├── cinematic_studio.json |
| 129 | └── volumetric_god_rays.json |
| 130 | ``` |
| 131 | |
| 132 | ### 组合示例 |
| 133 | |
| 134 | ```json |
| 135 | { |
| 136 | "template": "cyberpunk_chase_scene", |
| 137 | "modules": { |
| 138 | "camera": "$ref: camera_modules/dolly_in.json", |
| 139 | "lighting": "$ref: lighting_modules/neon_noir.json", |
| 140 | "environment": { |
| 141 | "weather": "Heavy Rain", |
| 142 | "time": "Night" |
| 143 | } |
| 144 | } |
| 145 | } |
| 146 | ``` |
| 147 | |
| 148 | ### 可复用组件库 |
| 149 | |
| 150 | ```javascript |
| 151 | // 摄像机模块 |
| 152 | const CameraModules = { |
| 153 | dollyIn: { |
| 154 | movement: "Dolly In", |
| 155 | speed: "Slow", |
| 156 | duration: "3s" |
| 157 | }, |
| 158 | trackingShot: { |
| 159 | movement: "Tracking", |
| 160 | subject_distance: "Constant", |
| 161 | stability: "High" |
| 162 | } |
| 163 | } |
| 164 | |
| 165 | // 使用 |
| 166 | { |
| 167 | ...CameraModules.dollyIn, |
| 168 | subject: "Your subject here" |
| 169 | } |
| 170 | ``` |
| 171 | |
| 172 | --- |
| 173 | |
| 174 | ## 技巧4：避免过度嵌套 (Avoid Deep Nesting) |
| 175 | |
| 176 | **核心原则**：保持结构扁平化，不超过 3-4 层。 |
| 177 | |
| 178 | ### 问题示例 |
| 179 | |
| 180 | ```json |
| 181 | { |
| 182 | "scene": { |
| 183 | "background": { |
| 184 | "details": { |
| 185 | "objects": [...] |
| 186 | } |
| 187 | } |
| 188 | } |
| 189 | } |
| 190 | ``` |
| 191 | |
| 192 | ### 推荐方案 |
| 193 | |
| 194 | ```json |
| 195 | { |
| 196 | "scene_background": "...", |
| 197 | "background_objects": [...], |
| 198 | "scene_foreground": "..." |
| 199 | } |
| 200 | ``` |
| 201 | |
| 202 | ### 扁平化优势 |
| 203 | |
| 204 | 1. 模型注意力更集中 |
| 205 | 2. 易于解析和处理 |
| 206 | 3. 减少嵌套错误 |
| 207 | |
| 208 | --- |
| 209 | |
| 210 | ## 技巧5：显式定义负向约束 (Negative Constraints) |
| 211 | |
| 212 | **核心原则**：在 JSON 中专门开辟负向提示词字段。 |
| 213 | |
| 214 | ### 负向约束结构 |
| 215 | |
| 216 | ```json |
| 217 | { |
| 218 | "task": "Generate an image", |
| 219 | "subject": "A cat", |
| 220 | "style": "Photorealistic", |
| 221 | "constraints": { |
| 222 | "forbidden_elements": ["text", "watermark", "humans"], |
| 223 | "style_restrictions": "No cartoon style, photorealistic only", |
| 224 | "quality_requirements": "High resolution, sharp details" |
| 225 | } |
| 226 | } |
| 227 | ``` |
| 228 | |
| 229 | ### 分层负向约束 |
| 230 | |
| 231 | ```json |
| 232 | { |
| 233 | "video_generation": { |
| 234 | "subject": "...", |
| 235 | "action": "...", |
| 236 | "negative_prompts": { |
| 237 | "visual": ["morphing", "distortion", "flickering"], |
| 238 | "temporal": ["aspect_ratio_change", "resolution_change"], |
| 239 | "quality": ["low_resolution", "blurry", "compression_artifacts"] |
| 240 | } |
| 241 | } |
| 242 | } |
| 243 | ``` |
| 244 | |
| 245 | --- |
| 246 | |
| 247 | ## 适用场景 |
| 248 | |
| 249 | ### 场景1：高精度视频生成 |
| 250 | |
| 251 | **为什么用 JSON**：视频包含主体、环境、运镜、物理规律等多维度。 |
| 252 | |
| 253 | **优势**：确保模型不会把"镜头向左移"理解成"主角向左走"。 |
| 254 | |
| 255 | ### 场景2：批量化内容生产 |
| 256 | |
| 257 | **用例**：生成 1000 条格式统一、内容不同的广告文案。 |
| 258 | |
| 259 | **方法**：固定 JSON Schema，用脚本动态替换变量。 |
| 260 | |
| 261 | ### 场景3：复杂角色扮演 |
| 262 | |
| 263 | **用例**：维护复杂的角色状态。 |
| 264 | |
| 265 | ```json |
| 266 | { |
| 267 | "character_state": { |
| 268 | "current_mood": "Angry", |
| 269 | "memory": ["Insulted by user"], |
| 270 | "goal": "Seek revenge", |
| 271 | "inventory": ["sword", "shield"] |
| 272 | } |
| 273 | } |
| 274 | ``` |
| 275 | |
| 276 | ### 场景4：API 对接与工作流 |
| 277 | |
| 278 | **用例**：提示词是自动化程序的一部分，需要被代码解析。 |
| 279 | |
| 280 | --- |
| 281 | |
| 282 | ## 工作流程 |
| 283 | |
| 284 | ### 步骤1：分析需求结构 |
| 285 | |
| 286 | ``` |
| 287 | 输入：用户自然语言描述 |
| 288 | 处理：提取关键维度 |
| 289 | 输出：JSON Schema 草稿 |
| 290 | ``` |
| 291 | |
| 292 | ### 步骤2：设计 Schema |
| 293 | |
| 294 | ``` |
| 295 | 1. 识别必需字段 |
| 296 | 2. 确定数据类型 |
| 297 | 3. 设计嵌套结构 |
| 298 | 4. 添加约束条件 |
| 299 | ``` |
| 300 | |
| 301 | ### 步骤3：优化键名 |
| 302 | |
| 303 | ``` |
| 304 | - 语义化命名 |
| 305 | - 添加单位 |
| 306 | - 统一前缀 |
| 307 | - 避免缩写 |
| 308 | ``` |
| 309 | |
| 310 | ### 步骤4：添加注释 |
| 311 | |
| 312 | ``` |
| 313 | - 关键约束说明 |
| 314 | - 取值范围注释 |
| 315 | - 示例值说明 |
| 316 | ``` |
| 317 | |
| 318 | ### 步骤5：验证输出 |
| 319 | |
| 320 | ``` |
| 321 | - 检查嵌套深度 |
| 322 | - 确认必需字段 |
| 323 | - 测试解析可行性 |
| 324 | ``` |
| 325 | |
| 326 | --- |
| 327 | |
| 328 | ## 示例对话 |
| 329 | |
| 330 | ### 示例1：设计视频提示词 Schema |
| 331 | |
| 332 | **用户**："帮我设计一个JSON |
People who liked this also viewed
6 picksType
- hyperframes-cliUse the HyperFrames CLI development loop: init, add, catalog, capture, lint, check, snapshot, compare, grade-compare, preview, play, present, beats, keyframes, single or batch render, publish, cloud, SkillsJul 2026
- request-refactor-planSkills for Real Engineers. Straight from my .agents directory.SkillsJul 2026
- analyze-projectRigor Analyze / Rigor Audit read-only skill for deep learning research repositories. Use when the user wants to read and understand a repository, inspect model structure and training or inference entrSkillsMay 2026
- explore-codeRigor Improve implementation leaf skill for auditable candidate implementation in deep learning research repositories. Use when the researcher explicitly authorizes exploratory work on an isolated braSkillsMay 2026
- ubiquitous-languageSkills for Real Engineers. Straight from my .agents directory.SkillsJul 2026
- scaffold-exercisesSkills for Real Engineers. Straight from my .agents directory.SkillsJul 2026
