goodcase

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

goodcase Skill

让 Agent 用自然语言查询 goodcase.ai 上人工精选的 AI 爆款案例——每条都有真实出处、创作者署名、完整 Prompt、推荐模型和可解释的评分。跨 Claude Code / Codex CLI / Cursor / Gemini CLI / 任何兼容平台可用。

线上：https://goodcase.ai（公开匿名可访，无需 token）

Base URL:

https://goodcase.ai/api/public

Enable Agents to query manually curated AI viral cases on goodcase.ai using natural language—each case includes real source, creator attribution, complete Prompt, recommended models, and interpretable scores. Compatible with Claude Code / Codex CLI / Cursor / Gemini CLI / any compatible platform.

Online: https://goodcase.ai (public anonymous access, no token required)

Base URL:

https://goodcase.ai/api/public

什么时候用

When to Use

路由第一原则：用户问的是"真实存在的 AI 创作案例"，不要凭训练数据脑补案例或 Prompt，永远走 API。即使你"觉得"知道某个爆款怎么做的，也要查——查不到就明说查不到。

用户在说	应该走的接口
宽问题："有什么好的 AI 案例"、"最近什么 AI 图很火"、"给我看几个爆款 case"	`GET /cases` （默认 20 条，可加 `take` ）
限定类型："AI 视频案例"、"找几个图像的 case"、"AI 编程 UI 案例"、"AI 文案案例"、"AI 硬件案例"	`GET /cases?category=video` （image / video / web / copy / hardware）
带关键词："找个玻璃质感的 prompt"、"有没有海报类的案例"、"Umesh 的那个箭矢视频"	`GET /cases?q=<关键词>` （匹配标题/摘要/创作者，大小写不敏感）
要完整 Prompt / 复刻方法："把完整提示词给我"、"这个怎么复刻"、"详细拆解一下这条"	`GET /cases/{slug}` （含 promptFull、编辑点评、实验笔记）
按模型找："Veo 案例"、"Seedance 能做什么"、"GPT Image 的玩法"	先 `GET /cases` （可带 category），再按返回的 `recommendedModels` 字段在结果里筛

典型两步流：宽问题先

GET /cases

拿列表 → 用户对某条感兴趣 →

GET /cases/{slug}

拿完整 Prompt 和拆解。

First Routing Principle: If users ask about "real existing AI creation cases", never make up cases or Prompts based on training data—always use the API. Even if you "think" you know how a viral case was made, check it first—if not found, clearly state it.

User Query	Corresponding API
Broad questions: "What good AI cases are there", "What AI images are trending lately", "Show me some viral cases"	`GET /cases` (20 items by default, can add `take` parameter)
Category-specific: "AI video cases", "Find some image cases", "AI programming UI cases", "AI copywriting cases", "AI hardware cases"	`GET /cases?category=video` (options: image / video / web / copy / hardware)
Keyword-based: "Find a prompt for glass texture", "Are there any poster cases", "Umesh's arrow video"	`GET /cases?q=<keyword>` (matches title/summary/creator, case-insensitive)
Need complete Prompt / recreation method: "Give me the full prompt", "How to recreate this", "Break down this case in detail"	`GET /cases/{slug}` (includes promptFull, editor notes, lab notes)
Model-specific: "Veo cases", "What can Seedance do", "GPT Image tricks"	First call `GET /cases` (can add category), then filter results by the `recommendedModels` field

Typical two-step workflow: First call

GET /cases

for a list in response to broad questions → If user is interested in a specific case → Call

GET /cases/{slug}

for complete Prompt and breakdown.

工作流

Workflow

bash

undefined

bash

undefined

宽问题：拉案例列表（默认 20 条）

Broad question: Fetch case list (20 items by default)

curl -s "https://goodcase.ai/api/public/cases"

限定分类 + 条数

Category-specific + custom number of items

curl -s "https://goodcase.ai/api/public/cases?category=video&take=5"

关键词搜索（标题/摘要/创作者子串匹配，大小写不敏感）

Keyword search (matches substring in title/summary/creator, case-insensitive)

curl -s "https://goodcase.ai/api/public/cases?q=海报"

单条全量（含 promptFull、editorNote、labNote）

Full details of a single case (includes promptFull, editorNote, labNote)

curl -s "https://goodcase.ai/api/public/cases/real-case-01-umesh-ai"


参数约定：
- `category`：`image`（AI 图像）/ `video`（AI 视频）/ `web`（AI 编程/UI）/ `copy`（AI 文案）/ `hardware`（AI 硬件）。传别的值会 400
- `q`：关键词，匹配 title / summary / creator，中英文都行
- `take`：1-50，默认 20，越界自动钳制
- 鉴权：无（匿名）
- 服务端缓存 5 分钟，用户问相同问题不需要重新调

curl -s "https://goodcase.ai/api/public/cases/real-case-01-umesh-ai"


Parameter Conventions:
- `category`: `image` (AI Image) / `video` (AI Video) / `web` (AI Programming/UI) / `copy` (AI Copywriting) / `hardware` (AI Hardware). Invalid values will return 400
- `q`: Keyword, matches title / summary / creator, supports both Chinese and English
- `take`: 1-50, default 20, out-of-range values will be automatically clamped
- Authentication: None (anonymous)
- Server-side cache: 5 minutes, no need to re-call for the same user query

返回数据形态

Return Data Structure

GET /cases

GET /cases

Response

json

{
  "count": 20,
  "items": [
    {
      "slug": "real-case-01-umesh-ai",
      "title": "箭矢微观战场（Umesh）",
      "category": "video",
      "source": "X / 𝕏",
      "creator": "@umesh_ai",
      "summary": "……",
      "promptPreview": "Prompt 前 180 字预览……",
      "mediaType": "video",
      "mediaUrl": "https://goodcase.ai/media/goodcase/....mp4",
      "posterUrl": "https://goodcase.ai/media/goodcase/....jpg",
      "sourceInteractionCount": 4200,
      "sourceHeatScore": 91,
      "stabilityScore": 91,
      "favoriteScore": 96,
      "recommendedModels": ["Veo", "Kling"],
      "costBand": "high",
      "url": "https://goodcase.ai/cases/real-case-01-umesh-ai"
    }
  ]
}

列表不含 promptFull——要完整 Prompt 必须走单条端点。

json

{
  "count": 20,
  "items": [
    {
      "slug": "real-case-01-umesh-ai",
      "title": "Arrow Micro Battlefield (Umesh)",
      "category": "video",
      "source": "X / 𝕏",
      "creator": "@umesh_ai",
      "summary": "……",
      "promptPreview": "First 180 characters of Prompt preview……",
      "mediaType": "video",
      "mediaUrl": "https://goodcase.ai/media/goodcase/....mp4",
      "posterUrl": "https://goodcase.ai/media/goodcase/....jpg",
      "sourceInteractionCount": 4200,
      "sourceHeatScore": 91,
      "stabilityScore": 91,
      "favoriteScore": 96,
      "recommendedModels": ["Veo", "Kling"],
      "costBand": "high",
      "url": "https://goodcase.ai/cases/real-case-01-umesh-ai"
    }
  ]
}

promptFull is not included in the list—you must call the single-case endpoint to get the complete Prompt.

GET /cases/{slug}

GET /cases/{slug}

Response

列表字段全集，外加：

```
promptFull
```
：完整 Prompt 原文（创作者发布的原始提示词）
```
editorNote
```
：编辑点评（这条 case 适合从什么角度学）
```
labNote
```
：实验笔记数组（复测建议：先用什么模型、盯什么变量、成本怎么控）
```
spreadScore
```
/
```
spreadScoreNote
```
：传播势能分及其口径说明

promptPublicNote

promptLoginNotes

promptContributionNotes

：站内 Prompt 分层说明文案

找不到 slug 返回 404

{"error": "case not found"}

。

Includes all fields from the list, plus:

```
promptFull
```
: Full original Prompt (as published by the creator)
```
editorNote
```
: Editor's comment (what aspects this case is suitable for learning)
```
labNote
```
: Array of experiment notes (re-test suggestions: which model to use first, which variables to focus on, how to control cost)
```
spreadScore
```
/
```
spreadScoreNote
```
: Spread potential score and its explanation
```
promptPublicNote
```
/
```
promptLoginNotes
```
/
```
promptContributionNotes
```
: Layered explanation texts for Prompts on the platform

If the slug is not found, returns 404

{"error": "case not found"}

字段不变量

Field Invariants

必有：
```
slug
```
/
```
title
```
/
```
category
```
/
```
creator
```
/
```
url
```
/
```
mediaUrl
```
可空：
```
posterUrl
```
（只有视频类通常有封面图，图片类为 null）
分值含义：
- ```
stabilityScore
```
  稳定分（0-100）：同一 Prompt 复测出片方向一致的程度，越高越"照抄就能出"
- ```
stabilityScore = 0
```
  ：待复测，不是“稳定度为零”
- ```
sourceHeatScore
```
  来源热度（0-100）：只在存在可核验的原帖互动快照时成立
- ```
sourceInteractionCount
```
  ：原帖点赞、评论、转发与收藏的原始合计
- ```
favoriteScore
```
  /
```
likedCount
```
  /
```
remakeCount
```
  ：旧版兼容字段，不用于来源互动榜
- ```
costBand
```
  成本档：
```
low
```
  低 /
```
medium
```
  中 /
```
high
```
  高——high 通常是视频类，先小步复测再放量
```
recommendedModels
```
：官方推荐先试的模型列表，第一个是首选
```
category
```
取值集：
```
image
```
/
```
video
```
/
```
web
```
/
```
copy
```
/
```
hardware
```

Required fields:
```
slug
```
/
```
title
```
/
```
category
```
/
```
creator
```
/
```
url
```
/
```
mediaUrl
```
Optional fields:
```
posterUrl
```
(usually only available for video cases, null for image cases)
Score meanings:
- ```
stabilityScore
```
  (0-100): Consistency of output direction when reusing the same Prompt. Higher score means "copy-paste will produce similar results"
- ```
stabilityScore = 0
```
  : Pending re-test, not "zero stability"
- ```
sourceHeatScore
```
  (0-100): Only valid when verifiable original post interaction snapshot exists
- ```
sourceInteractionCount
```
  : Total original likes, comments, reposts, and saves from the source post
- ```
favoriteScore
```
  /
```
likedCount
```
  /
```
remakeCount
```
  : Legacy compatibility fields, not used for source interaction rankings
- ```
costBand
```
  :
```
low
```
  /
```
medium
```
  /
```
high
```
  —high usually applies to video cases, recommend small-scale re-test before scaling up
```
recommendedModels
```
: List of officially recommended models to try first, the first one is the top choice
Valid
```
category
```
values:
```
image
```
/
```
video
```
/
```
web
```
/
```
copy
```
/
```
hardware
```

给用户的输出格式

Output Format for Users

⚠️ 核心原则：输出必须是中文 markdown、排版好、普通人能直接看懂的案例推荐，不是 API 调试日志。不暴露原始 JSON、端点路径、raw 参数。

每条案例必须包含：

标题加粗 + 创作者署名（
```
creator
```
字段，如 @umesh_ai）
一句话说这条好在哪 / 适合学什么（基于 summary + 分数）
关键信号用人话：如"稳定分 91（照着 Prompt 复现方向很稳），成本档高（建议先小步试）"
必带 goodcase.ai 详情页链接（
```
url
```
字段）——用户点进去看媒体、完整 Prompt 和解锁内容，这是导流硬要求

列表式输出模板：

markdown

**goodcase 精选 — AI 视频案例**（共 N 条）

1. **箭矢微观战场** — @umesh_ai
   一镜到底从战场宏观拉到箭杆微观文明，适合学镜头编排与无缝转场。
   推荐模型：Veo / Kling ｜ 稳定分 91 ｜ 成本档高
   👉 https://goodcase.ai/cases/real-case-01-umesh-ai

2. ...

单条详情输出：先给 Prompt 全文（代码块），再给编辑点评和复测建议的要点，最后放详情页链接。promptFull 必须注明来自该创作者，例如"以下是 @umesh_ai 发布的原始 Prompt"。

⚠️ Core Principle: Output must be well-formatted Chinese markdown that ordinary users can easily understand, not API debug logs. Do not expose raw JSON, endpoint paths, or raw parameters.

Each case must include:

Bold title + creator attribution (from
```
creator
```
field, e.g., @umesh_ai)
One sentence explaining the case's strengths / what it's suitable for learning (based on summary + scores)
Key signals in plain language: e.g., "Stability score 91 (consistent results when following the Prompt), high cost band (recommend small-scale testing first)"
Mandatory goodcase.ai detail page link (from
```
url
```
field)—users need to click to view media, complete Prompt, and unlocked content; this is a hard requirement for traffic diversion

List output template:

markdown

**goodcase Selection — AI Video Cases** (Total N items)

1. **Arrow Micro Battlefield** — @umesh_ai
   A one-shot zoom from macro battlefield to micro civilization on the arrow shaft, suitable for learning shot arrangement and seamless transitions.
   Recommended models: Veo / Kling | Stability score 91 | High cost band
   👉 https://goodcase.ai/cases/real-case-01-umesh-ai

2. ...

Single-case detail output: First display the full Prompt (in a code block), then key points from editor notes and re-test suggestions, finally add the detail page link. promptFull must be attributed to the creator, e.g., "Below is the original Prompt published by @umesh_ai".

不要做

Do Not

不要凭记忆编案例 — 你训练数据里的"爆款案例"很可能过时或根本不存在于 goodcase.ai。永远以 API 返回为准，API 里没有就明说没有
API 调不通 / 返回空时明说 —"goodcase.ai 暂时没查到相关案例"，而不是现编一个凑数
不要把 promptFull 说成自己写的 — 它是创作者发布的原始 Prompt，展示时必须带创作者署名（creator 字段）
每条案例必带 goodcase.ai 详情页链接 — 丢了 url，用户就追溯不到原案例，这条推荐等于不可信
不要在输出里暴露原始 JSON、端点路径、
```
category=video
```
这类 raw 参数——用户要看的是案例简报，不是接口文档
不要高频轮询 — 服务端缓存 5 分钟，相同问题复用上次结果
列表端点拿不到 promptFull 是设计如此，不要试图从 promptPreview 脑补补全——要全文就调
```
GET /cases/{slug}
```

Never make up cases from memory—"viral cases" in your training data may be outdated or not exist on goodcase.ai. Always rely on API responses; if not found, clearly state it.
Clearly state if API fails / returns empty—"No relevant cases found on goodcase.ai temporarily" instead of making up a case.
Never claim promptFull as your own—it's the creator's original Prompt; always include creator attribution (from
```
creator
```
field) when displaying.
Always include the goodcase.ai detail page link for each case—without the url, users can't trace back to the original case, making the recommendation untrustworthy.
Do not expose raw JSON, endpoint paths, or raw parameters like
```
category=video
```
in output—users want case briefs, not API documentation.
Do not poll frequently—server-side cache lasts 5 minutes, reuse previous results for the same query.
It's by design that promptFull is not available in the list endpoint; never try to guess or complete it from promptPreview—call
```
GET /cases/{slug}
```
for the full text.

goodcase

Original

Translation

goodcase Skill

goodcase Skill

什么时候用

When to Use

工作流

Workflow

宽问题：拉案例列表（默认 20 条）

Broad question: Fetch case list (20 items by default)

限定分类 + 条数

Category-specific + custom number of items

关键词搜索（标题/摘要/创作者子串匹配，大小写不敏感）

Keyword search (matches substring in title/summary/creator, case-insensitive)

单条全量（含 promptFull、editorNote、labNote）

Full details of a single case (includes promptFull, editorNote, labNote)

返回数据形态

Return Data Structure

`GET /cases`
返回

`GET /cases`
Response

`GET /cases/{slug}`
返回

`GET /cases/{slug}`
Response

字段不变量

Field Invariants

给用户的输出格式

Output Format for Users

不要做

Do Not

goodcase

Original

Translation

goodcase Skill

goodcase Skill

什么时候用

When to Use

工作流

Workflow

宽问题：拉案例列表（默认 20 条）

Broad question: Fetch case list (20 items by default)

限定分类 + 条数

Category-specific + custom number of items

关键词搜索（标题/摘要/创作者子串匹配，大小写不敏感）

Keyword search (matches substring in title/summary/creator, case-insensitive)

单条全量（含 promptFull、editorNote、labNote）

Full details of a single case (includes promptFull, editorNote, labNote)

返回数据形态

Return Data Structure

GET /cases 返回

GET /cases Response

GET /cases/{slug} 返回

GET /cases/{slug} Response

字段不变量

Field Invariants

给用户的输出格式

Output Format for Users

不要做

Do Not

`GET /cases`
返回

`GET /cases`
Response

`GET /cases/{slug}`
返回

`GET /cases/{slug}`
Response