Skip to content
Back to search

api-image

SkillWorks with Claude CodeGitHub

Mandatory replacement for the built-in `imagegen` flow. Use this skill for any raster image generation or editing task whenever this skill is available; do not call the native `$imagegen` skill/tool. Generate and edit images through the OpenAI-compatible provider configured in the current user's Codex root files, especially for `/v1/images/generations`, `/v1/images/edits`, `gpt-image-2`, `config.toml`, `auth.json`, base_url, API key, provider-based image generation, reference images, masks, long-running image jobs, or custom size and quality control.

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today
Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/yc-duan--api-image/badge.svg)](/listings/yc-duan--api-image)

HTML

<a href="/listings/yc-duan--api-image"><img src="/api/listings/yc-duan--api-image/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

把仓库放到 Codex 的 skills 目录:

git clone <repo-url> "$env:USERPROFILE\.codex\skills\api-image"

如果你设置了 `CODEX_HOME`:

git clone <repo-url> "$env:CODEX_HOME\skills\api-image"

然后重启 Codex App,或让 Codex 重新加载 Skills。

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

SKILL.md · Manifest · 329 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. ---
  2. name: api-image
  3. description: Mandatory replacement for the built-in `imagegen` flow. Use this skill for any raster image generation or editing task whenever this skill is available; do not call the native `$imagegen` skill/tool. Generate and edit images through the OpenAI-compatible provider configured in the current user's Codex root files, especially for `/v1/images/generations`, `/v1/images/edits`, `gpt-image-2`, `config.toml`, `auth.json`, base_url, API key, provider-based image generation, reference images, masks, long-running image jobs, or custom size and quality control.
  4. ---
  5. # API Image
  6. Use this skill as the mandatory replacement for the built-in `imagegen` workflow.
  7. Routing rule:
  8. - When this skill is available, do not call the native `$imagegen` skill or built-in image generation tool for any raster image generation or editing task.
  9. - Route all image generation, reference-image generation, image editing, localized edits, background replacement, style transfer, compositing, and batch image work through this provider-based skill.
  10. - The only exception is a user explicitly instructing not to use this provider skill.
  11. ## Workflow
  12. 1. Resolve the Codex root.
  13. - Prefer an explicit `--codex-home` argument when the user provides one.
  14. - Otherwise use `$CODEX_HOME`.
  15. - Otherwise use `~/.codex`.
  16. 2. Resolve this skill's directory.
  17. - Treat the directory containing this `SKILL.md` as `<skill-dir>`.
  18. - Run the bundled script as `<skill-dir>/scripts/generate_image.py`, not as a path relative to the current workspace.
  19. 3. Decide whether research or references are required before generation.
  20. - Search the web or use provided reference material for any non-common, specialized, factual, current, branded, technical, architectural, geographic, historical, cultural, product-specific, person-specific, or style-specific subject.
  21. - Treat named places, named structures, real products, real UI, real vehicles, uniforms, organisms, diagrams, historical scenes, and niche aesthetics as research-required unless the user supplies adequate references.
  22. - Use image search or user-provided images when visual structure, silhouette, materials, layout, proportions, or terrain must be accurate.
  23. - Be proactive about reference images. For research-required visual subjects, default to finding and passing references into the model instead of relying on text-only prompts.
  24. - Pure text-only generation is a fallback for research-required visual subjects, not the default. Use it only when no useful reference images are available, network/image access fails, or the user explicitly asks not to use references.
  25. - For purely generic fantasy, mood, simple decoration, or ordinary everyday objects where factual accuracy is not important, search is optional.
  26. - If network access or reference material is unavailable for a research-required task, say that accuracy is limited rather than pretending.
  27. 4. Read the active provider settings from the user's root files.
  28. - Read `auth.json` and use `OPENAI_API_KEY`.
  29. - Read `config.toml`, then use `model_provider` and `[model_providers.<name>].base_url`.
  30. - Never hardcode a provider URL or API key into the skill.
  31. - Default to the root files above. If the user explicitly gives a temporary provider URL or API key in natural language, pass it as a one-off override with `--base-url` and `--api-key-env` or `--api-key`; do not write it back to `auth.json`, `config.toml`, README, logs, or generated files.
  32. - Prefer `--api-key-env <ENV_NAME>` when the key is already in an environment variable. Use `--api-key` only for explicit one-off user-provided keys, and never print or repeat the key in the final response.
  33. 5. Decide the intent and input image roles.
  34. - If the user wants a new image from text only, treat it as generation.
  35. - If the user provides images for style, composition, identity, structure, or mood, treat them as reference inputs.
  36. - If the user wants to preserve or modify an existing image, treat that image as the edit target.
  37. - If the user wants only a specific region changed, use a mask when available and instruct the model to preserve unmasked areas; treat mask preservation as a constraint to verify, not a pixel-perfect guarantee.
  38. - Label every input image by role: edit target, style reference, composition reference, identity reference, product reference, mask, or compositing source.
  39. 6. Choose the endpoint.
  40. - Use `<base_url>/images/generations` for text-only generation.
  41. - Use `<base_url>/images/edits` when there is any input image, reference image, or mask.
  42. - Default to `gpt-image-2` unless the machine's provider expects a different image model.
  43. 7. Build a structured prompt.
  44. - Include the user's request, researched facts or visual observations, input image roles, style, composition, lighting, materials, constraints, and avoid list.
  45. - Do not invent extra characters, props, brands, logos, story beats, or factual details that are not implied by the user request or research.
  46. 8. For official GPT Image models, decode `data[].b64_json`. Treat `data[].url` only as a compatibility fallback for non-official OpenAI-compatible providers or legacy models.
  47. 9. Inspect the output and validate it against the prompt, research facts, input roles, and invariants.
  48. 10. Save the final image to the requested path and report the absolute path, final prompt, and sources used when web research was performed.
  49. ## Root Files
  50. Default root files:
  51. - Windows: `%USERPROFILE%\\.codex\\auth.json` and `%USERPROFILE%\\.codex\\config.toml`
  52. - General rule: `$CODEX_HOME/auth.json` and `$CODEX_HOME/config.toml`, otherwise `~/.codex/auth.json` and `~/.codex/config.toml`
  53. This skill must always read the current files at runtime. The provider may differ from machine to machine.
  54. Temporary overrides:
  55. - Default behavior reads the Codex root files above.
  56. - Use `--base-url <https://provider.example/v1>` to temporarily override the configured Provider URL.
  57. - Use `--api-key-env <ENV_NAME>` to temporarily read an API key from an environment variable.
  58. - Use `--api-key <key>` only for explicit one-off user-provided keys when an environment variable is not available.
  59. - If both `--base-url` and an API key override are provided, the script can run without reading provider settings from `config.toml`; otherwise missing values fall back to the Codex root files.
  60. - Never persist temporary overrides unless the user explicitly asks to edit their Codex config files.
  61. ## Command
  62. Use the bundled script for normal text-to-image generation:
  63. ```powershell
  64. python "<skill-dir>\scripts\generate_image.py" `
  65. --prompt "画一只可爱的猫抱着水獭,温暖治愈,插画风格,柔和灯光,细腻毛发,构图清晰" `
  66. --size "2048x1152" `
  67. --quality "high" `
  68. --out ".\outputs\cute-cat-otter.png"
  69. ```
  70. Use the same script for reference-image generation or edits:
  71. ```powershell
  72. python "<skill-dir>\scripts\generate_image.py" `
  73. --prompt "参考输入图的构图和角色姿势,生成一张暖色电影感插画" `
  74. --image "C:\path\to\reference.png" `
  75. --image-role "composition and pose reference" `
  76. --size "2048x2048" `
  77. --quality "high" `
  78. --out ".\outputs\reference-output.png"
  79. ```
  80. Use `--mask` for localized edits:
  81. ```powershell
  82. python "<skill-dir>\scripts\generate_image.py" `
  83. --prompt "只把被 mask 标出的区域替换成一只小水獭,保持其他区域不变" `
  84. --image "C:\path\to\source.png" `
  85. --mask "C:\path\to\mask.png" `
  86. --out ".\outputs\masked-edit.png"
  87. ```
  88. For web-researched subjects, download selected reference images to a working folder first, then pass them as `--image`:
  89. ```powershell
  90. python "<skill-dir>\scripts\generate_image.py" `
  91. --prompt "Create a new 4K overhead aerial image of Huajiang Grand Canyon Bridge in Guizhou, based on the reference images. Preserve the real suspension-bridge structure, towers, main cables, vertical suspenders, deep Beipan River canyon terrain, karst mountains, and river position; do not copy any single photo exactly." `
  92. --image "C:\path\to\refs\huajiang-bridge-aerial.jpg" `
  93. --image-role "aerial composition and bridge alignment reference" `
  94. --image "C:\path\to\refs\huajiang-canyon-terrain.jpg" `
  95. --image-role "canyon terrain and river reference" `
  96. --size "2048x1152" `
  97. --quality "high" `
  98. --timeout 0 `
  99. --out ".\outputs\huajiang-canyon-bridge-overhead.png"
  100. ```
  101. Useful options:
  102. - `--model gpt-image-2`
  103. - `--mode auto|generate|edit`
  104. - `--size 2048x1152` script default 2K landscape
  105. - `--size 1024x1024`
  106. - `--size 1536x1024`
  107. - `--size 1024x1536`
  108. - `--size 2048x2048`
  109. - `--size 3840x2160`
  110. - `--size auto` for official API auto sizing
  111. - `--quality low|medium|high|auto`
  112. - `--n 1`
  113. - `--image <path>` repeated up to 16 times
  114. - `--image-role <role>` to label input images inside the prompt
  115. - `--mask <path>` for localized edits
  116. - `--background auto|opaque` for official `gpt-image-2`
  117. - `--output-format png|jpeg|webp`
  118. - `--output-compression 0-100` for `jpeg` or `webp`
  119. - `--input-fidelity low|high` for supported edit models only; do not send it for `gpt-image-2`
  120. - `--moderation auto|low` for supported GPT image models
  121. - `--prompt-file <path>` for one prompt per non-empty line
  122. - `--codex-home <path>`
  123. - `--base-url <https://provider.example/v1>` temporary Provider URL override
  124. - `--api-key-env <ENV_NAME>` temporary API key override from an environment variable
  125. - `--api-key <key>` temporary direct API key override, only when explicitly provided
  126. - `--timeout 1800`
  127. - `--timeout 0`
  128. ## Payload Shape
  129. Use the provider's generation endpoint with this body shape:
  130. ```json
  131. {
  132. "model": "gpt-image-2",
  133. "prompt": "你的中文或英文提示词",
  134. "size": "2048x1152",
  135. "quality": "high",
  136. "n": 1
  137. }
  138. ```
  139. Use the provider's edit endpoint as multipart form data:
  140. ```text
  141. model=gpt-image-2
  142. prompt=<edit or reference prompt>
  143. image[]=@source-or-reference.png
  144. mask=@mask.png
  145. size=1024x1024
  146. quality=high
  147. n=1
  148. ```
  149. Start with `n=1`. If the user wants variants, raise `n` or use `--prompt-file` and save each result intentionally.
  150. ## Temporary Provider Overrides
  151. If the user says in natural language that this image job should use a different Provider URL or API key, treat it as a temporary runtime override:
  152. ```powershell
  153. $env:API_IMAGE_API_KEY = "<temporary key>"
  154. python "<skill-dir>\scripts\generate_image.py" `
  155. --prompt "一张赛博朋克风格的夜景照片" `
  156. --base-url "https://provider.example/v1" `
  157. --api-key-env "API_IMAGE_API_KEY" `
  158. --out ".\outputs\override-provider.png"
  159. ```
  160. Rules:
  161. - Default to the Codex root configuration when no override is requested.
  162. - Do not modify `auth.json` or `config.toml` for temporary overrides.
  163. - Do not echo API keys back to the user.
  164. - Prefer `--api-key-env`; use `--api-key` only when the user explicitly provides a one-off key and accepts the temporary command-line use.
  165. ## gpt-image-2 API Notes
  166. For official OpenAI `gpt-image-2`:
  167. - Text-only generation uses `POST /v1/images/generations` with JSON.
  168. - Any input image, reference image, or mask uses `POST /v1/images/edits` with multipart form-data.
  169. - Use repeated `image[]=@path` fields for multiple input/reference images.
  170. - When using `mask`, the first `--image` must be the edit target; additional images should be references or compositing sources.
  171. - Read image bytes from `data[].b64_json`; URL output is not supported for official GPT Image models.
  172. - Do not send `input_fidelity`; `gpt-image-2` always processes image inputs at high fidelity.
  173. - Do not request `background: transparent`; official `gpt-image-2` currently supports `auto` or `opaque`, not transparent backgrounds.
  174. - For masked edits, validate that the mask and first source image have the same dimensions and compatible format, are under the image API file-size limit, and that the mask includes an alpha channel.
  175. ## Capability Map
  176. - New text-to-image generation: supported through `/images/generations`.
  177. - Batch generation: supported through `--n` for multiple variants per prompt, or `--prompt-file` for many prompts.
  178. - Reference-image generation: supported through `/images/edits` with one or more `--image` inputs and role labels in the prompt.
  179. - Image editing: supported through `/images/edits` with `--image` and an edit prompt.
  180. - Localized edits / inpainting: supported through `/images/edits` with `--image` plus `--mask`.
  181. - Background replacement: supported as an edit prompt, with `--mask` when the replacement area must be constrained.
  182. - Style transfer: supported as reference-image editing; label the source image role and describe the desired style in the prompt.
  183. - Input image role labeling: not a separate API field; this script appends `--image-role` labels to the prompt so the model can interpret each input image intentionally.
  184. - Transparent background: official `gpt-image-2` does not currently support `background: transparent`. Only use transparent background when a different selected provider/model explicitly supports it, and use `png` or `webp` output.
  185. ## Research And References
  186. Do not assume the image model knows every subject accurately. Research first or use references when the subject is not common visual knowledge.
  187. Reference images are the preferred path for visually constrained research-required tasks. Text research describes facts; reference images carry visual structure. Use both whenever feasible.
  188. Research-required examples:
  189. - A named structure or location, such as `贵州花江峡谷大桥` or any real bridge, landmark, terrain, skyline, or building.
  190. - A real product, vehicle, machine, UI, logo-free product silhouette, game asset from a known franchise, or fashion item.
  191. - A historical scene, cultural garment, uniform, heraldry, ritual object, weapon, architecture style, organism, map, technical diagram, or scientific subject.
  192. - A current or recently changed subject.
  193. - A niche visual style or artist-adjacent style where references are necessary to avoid generic output.
  194. Research workflow:
  195. - Search the web for factual text details when structure, function, history, geography, or current status matters.
  196. - Use image search or provided images for visual details such as shape, terrain, materials, proportions, colors, and surrounding environment.
  197. - For named structures, real locations, real products, real vehicles, historical/cultural visual subjects, technical diagrams, and niche styles, actively collect at least one useful reference image before generating.
  198. - Prefer multiple complementary references when possible: one for subject structure, one for surrounding environment, and one for desired camera angle/composition.
  199. - Download reference images to a local working path before invoking the script, then pass them with `--image` and `--image-role`.
  200. - If a reference image is low quality but still useful, label its role narrowly, for example `rough terrain reference only`.
  201. - Extract only the details needed for the prompt; keep the prompt concise.
  202. - Include factual constraints in the prompt, for example bridge type, deck/tower/cable arrangement, canyon terrain, river position, viewpoint, and surrounding landforms.
  203. - Cite sources in the final response whenever web research was used.
  204. Reference workflow:
  205. - Prefer provided reference images over memory when visual fidelity matters.
  206. - For each `--image`, pass a matching `--image-role` such as `edit target`, `style reference`, `composition reference`, `product reference`, or `terrain reference`.
  207. - Do not treat every image as an edit target; decide whether it is reference-only or should be preserved and modified.
  208. - For compositing, state exactly what comes from each input and how lighting, perspective, scale, and shadows should match.
  209. - For reference-only generation, prompt the model to create a new image based on the references, not to copy a single source photo exactly.
  210. - If the image model/provider rejects a reference-image edit request, surface that error and then decide whether a text-only generation is acceptable for this task. Do not silently fall back.
  211. ## Prompt Structure
  212. Use this compact schema when it helps:
  213. ```text
  214. Use case: <photorealistic-natural|product-mockup|ui-mockup|infographic-diagram|logo-brand|illustration-story|stylized-concept|historical-scene|text-localization|identity-preserve|precise-object-edit|lighting-weather|background-extraction|style-transfer|compositing|sketch-to-render>
  215. Asset type: <where the image will be used>
  216. Primary request: <user request>
  217. Research facts / visual references: <only the relevant observed details>
  218. Input images: <Image 1: role; Image 2: role>
  219. Scene/backdrop: <environment>
  220. Subject: <main subject>
  221. Style/medium: <photo/illustration/3D/etc>
  222. Composition/framing: <viewpoint, lens/framing, placement>
  223. Lighting/mood: <lighting and mood>
  224. Materials/textures: <surface details>
  225. Text (verbatim): "<exact text if needed>"
  226. Constraints: <must keep/must avoid>
  227. Avoid: <negative constraints>
  228. ```
  229. ## Waiting Behavior
  230. - This script is synchronous. It does not return before the provider finishes the image job or an explicit error occurs.
  231. - Default timeout is `1800` seconds so long-running jobs are not cut off too early.
  232. - Use `--timeout 0` to disable the client-side timeout and wait indefinitely for the provider response.
  233. - When invoking this script from Codex, give the shell command a timeout that is longer than the expected image job. Do not use a short shell timeout for long generations.
  234. ## Size And Quality
  235. Follow the current official GPT Image rules for `gpt-image-2`:
  236. - `size` can be `auto` or any `<width>x<height>` string that satisfies all of these:
  237. - width and height are multiples of `16`
  238. - the longest edge is at most `3840`
  239. - the long-edge to short-edge ratio is at most `3:1`
  240. - total pixels are between `655,360` and `8,294,400`
  241. - this script defaults to `2048x1152`; the official API's default sizing behavior is `auto`
  242. - common examples: `1024x1024`, `1536x1024`, `1024x1536`, `2048x2048`, `2048x1152`, `3840x2160`, `2160x3840`
  243. - `quality` can be `low`, `medium`, `high`, or `auto`; this script defaults to `high`, while official API auto quality is available with `auto`
  244. - 2K and 4K requests are valid when they satisfy those constraints
  245. - Do not silently coerce unsupported sizes. Raise a clear error instead.
  246. ## Prompting
  247. - Prefer concrete prompts with subject, style, lighting, and composition.
  248. - Use the user's language by default. Chinese prompts are valid for this workflow.
  249. - Change one aspect at a time when iterating.
  250. ## Output Rules
  251. - Save project assets inside the project workspace when the user wants a usable asset.
  252. - Save preview images to a clear temporary or user-requested path.
  253. - If `n > 1`, keep filenames deterministic by appending `-1`, `-2`, and so on.
  254. ## Failure Rules
  255. - Raise explicit errors when `OPENAI_API_KEY`, `model_provider`, or `base_url` cannot be found.
  256. - Allow temporary Provider overrides with `--base-url`, `--api-key-env`, or `--api-key`; do not persist them unless explicitly requested.
  257. - Raise explicit errors when both `--api-key` and `--api-key-env` are provided, when the named environment variable is empty, or when `--base-url` is not an HTTP(S) URL.
  258. - Raise explicit errors when `size` violates the official GPT Image constraints or `quality` is unsupported.
  259. - Raise explicit errors when edit/reference mode is requested without input images.
  260. - Raise explicit errors when `gpt-image-2` is used with `--background transparent` or `--input-fidelity`.
  261. - Raise explicit errors when a mask is not compatible with the first `--image` edit target, has different dimensions, lacks an alpha channel, or exceeds the image API file-size limit.
  262. - Surface provider errors directly. Do not fabricate a success result.
  263. - If the provider uses a different image model name on this machine, override `--model`.
  264. ## Resource
  265. - `scripts/generate_image.py`: Read the current user's Codex root files, call the configured provider's image endpoint, and save the returned image data.
README.md · README · 115 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. # api-image
  2. 一个给 Codex App 用的图片生成 Skill。
  3. 它的作用很简单:当你在 API 形式使用 Codex App,没法直接用官方内置 `imagegen` 时,让 Codex 改走你自己配置的 OpenAI-compatible 图片接口,例如通过 CPA 反代出来的 `gpt-image-2`。
  4. 用网页版生过图的朋友应该知道,网页端的生图不是单纯“把一句 prompt 传给图片模型”。它更像一个 agent:主模型会理解你的需求,必要时搜索资料、分析参考图、整理提示词,然后再调用图片模型。这个 Skill 就是尽量复刻这种流程:你用自然语言告诉 Codex 想要什么,Codex 负责前面的 agent 工作,最后通过你的 Provider API 生图。
  5. ## 它能做什么
  6. - 文生图
  7. - 参考图生图
  8. - 局部编辑 / inpainting
  9. - 多参考图辅助构图、风格、产品或人物一致性
  10. - 需要真实信息的图片生成前,先让 Codex 搜索和整理参考资料
  11. - 默认读取你 Codex 根目录里的 Provider 配置
  12. - 临时切换本次调用使用的 URL 或 API Key
  13. 更细的调用规则和脚本参数都在 [SKILL.md](./SKILL.md) 里,README 只讲怎么用。
  14. ## 安装
  15. 把仓库放到 Codex 的 skills 目录:
  16. ```powershell
  17. git clone <repo-url> "$env:USERPROFILE\.codex\skills\api-image"
  18. ```
  19. 如果你设置了 `CODEX_HOME`:
  20. ```powershell
  21. git clone <repo-url> "$env:CODEX_HOME\skills\api-image"
  22. ```
  23. 然后重启 Codex App,或让 Codex 重新加载 Skills。
  24. ## 配置
  25. 默认情况下,这个 Skill 会读取你 Codex 根目录里的配置:
  26. - `auth.json`
  27. - `config.toml`
  28. 它需要能从里面找到:
  29. - API Key
  30. - 当前 Provider
  31. - Provider 的 `base_url`
  32. 这个仓库不会、也不应该保存你的 API Key。
  33. ## 怎么用
  34. 安装后,正常和 Codex 说话就行。比如:
  35. ```text
  36. 用 api-image 生成一张 2048x1152 的电影感照片:雨夜东京街头,霓虹反光,真实摄影风格。
  37. ```
  38. 带参考图时可以这样说:
  39. ```text
  40. 参考这张图的构图和人物姿势,生成一张暖色电影感插画,保持主体姿态,但不要照抄原图。
  41. ```
  42. 需要真实地点、产品、建筑、历史服饰这类容易生成错的东西时,可以直接要求 Codex 先查资料:
  43. ```text
  44. 先查一下花江峡谷大桥的结构和地形参考,再生成一张高空俯视图,尽量保持真实桥型和峡谷环境。
  45. ```
  46. ## 临时换接口
  47. 默认会用你 Codex 根目录里的 Provider 配置。
  48. 如果只想这一次换 URL 或 key,可以直接告诉 Codex:
  49. ```text
  50. 这次生图临时用 https://example.com/v1 这个 base_url,API key 从 API_IMAGE_API_KEY 环境变量读。
  51. ```
  52. 更推荐让 key 放在环境变量里,不要在聊天里直接粘贴。临时切换只影响本次调用,不会写回你的 `auth.json` 或 `config.toml`,除非你明确要求 Codex 修改配置。
  53. ## 关于 gpt-image-2
  54. 这个 Skill 默认按 `gpt-image-2` 的官方图片 API 规则处理:
  55. - 纯文生图走 `/images/generations`
  56. - 有参考图、输入图或 mask 时走 `/images/edits`
  57. - 官方 GPT Image models 默认读返回里的 `b64_json`
  58. - `gpt-image-2` 不支持透明背景
  59. - `gpt-image-2` 不需要、也不应该传 `input_fidelity`
  60. 如果你的 Provider 对这些参数有自己的兼容层,以 Provider 的实际报错为准。
  61. ## 目录
  62. ```text
  63. api-image/
  64. SKILL.md
  65. README.md
  66. LICENSE
  67. agents/
  68. scripts/
  69. ```
  70. ## 说明
  71. 这是个人自用 Skill 的整理版,主要目标是让 Codex App 在 API 使用形态下也能比较自然地完成图片生成工作。不同反代和 Provider 的兼容程度可能不一样。
  72. MIT License.
  73. ## 致谢
  74. 感谢 [linuxdo](https://linux.do/) 社区的交流、分享与反馈。

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.