Guide · Agent skill
How to use the story-cover agent skill: novel cover generator
Updated · written with AI from the project's own documentation and checked against our catalogue data
story-cover is an agent skill from the oh-story-claudecode project that builds an image prompt for a novel cover and, when asked, generates the cover image with the title and pen name rendered on it. This guide is for authors and designers who have never used an agent skill before. By the end you will know what the skill does, how to install it, how to ask for a cover in plain language, and what it needs to run safely.
Novel Cover Generatorzenstory-ai/oh-story-claudecode · MIT Download the skill Quick start
- 1Download and unzip. Download the story-cover ZIP from this page and unzip it into ~/.claude/skills/ or your project's .claude/skills/ folder.
- 2Or upload it in the Claude apps. Open Settings, go to Capabilities, then Skills, and upload the story-cover ZIP.
- 3Set up the image route. In Codex CLI, the built-in ImageGen tool needs no key. For the API fallback, set GPT_IMAGE_API_KEY.
- 4Start a cover request. Ask in plain language, such as 帮我做个封面, or use /story-cover or /封面.
- 5Confirm the book details. Let the agent read your project files, or give it the title, pen name and target platform when asked.
- 6Choose a composition. Pick one of the two or three composition options the agent offers. It generates only that one unless you ask for all.
- 7Export the upload size. If your platform has a fixed upload size, the crop step makes a copy at that size and keeps the original.
What is an agent skill, and what does story-cover do?
An agent skill is a folder of instructions and optional scripts that an AI coding agent loads when your request matches the skill. The skill tells the agent how to carry out one type of task. story-cover is the skill for novel covers. It is licensed under MIT.
When it runs, the agent first looks for your book. It reads the title, genre, target platform and core premise from your project files, or asks you for the missing details. It then picks a cover style that fits the genre and writes a detailed image prompt. If you want an image, it generates a cover at 3:4 or 2:3 with the title and author name in Chinese, and can crop the result to a platform's upload size.
- Builds a cover prompt from the book title, pen name and genre. If you only want the prompt, it stops there and generates no image.
- Chooses a style from a built-in library of genre and platform looks.
- Generates the image with the image tool in Codex CLI when available, or with an API fallback.
- Exports a copy at a platform's exact upload size, such as 600×800 for Fanqie, when one is set.
Install the story-cover skill
Download the story-cover ZIP from this page and unzip it. For a personal setup, place the folder in ~/.claude/skills/. To limit it to one book project, place it in that project's .claude/skills/ folder.
You can also upload the skill in the Claude apps: open Settings, go to Capabilities, then Skills, and upload the ZIP there.
- Personal install: ~/.claude/skills/story-cover
- Project install: your project's .claude/skills/story-cover
- Claude apps: Settings → Capabilities → Skills, then upload
Prepare the requirements
The skill needs an image generation route. Inside Codex CLI, it uses the built-in ImageGen tool first and needs no API key. If that tool is not available, it falls back to the OpenAI image API, which requires a GPT_IMAGE_API_KEY. The source notes that this key can point to OpenAI or a compatible proxy.
Some tools are needed for scripts. The skill lists curl, jq and base64. For the optional crop step it uses ImageMagick (magick or convert), or macOS sips if ImageMagick is missing. Prompt-only mode needs none of these image tools, keys or output folders.
- Image route: Codex CLI built-in ImageGen, or GPT_IMAGE_API_KEY for the API fallback
- Command-line tools: curl, jq, base64
- Optional crop: ImageMagick (magick or convert), or macOS sips
- Prompt-only mode: no image tool, key or output folder needed
Trigger the skill with natural requests
The skill responds to its slash commands and to plain requests. The source lists /story-cover and /封面, plus phrases such as 帮我做个封面, 生成封面图, 封面提示词 and 封面设计. Ask for a prompt only if you do not want an image yet.
If the agent cannot find your book, it asks you once for the title, pen name and target platform. It will not invent a pen name or leave the field blank.
- /story-cover to start the skill directly
- /封面 as the Chinese shortcut
- 帮我做个封面 for a full cover with an image
- 封面提示词 when you only want the prompt text
Work through the cover process
The skill collects the book details, decides the genre, builds the prompt, then generates. It offers two or three composition options and generates only the one you choose, unless you ask for all of them. Each prompt has three layers: text, style and scene. Text uses the author's original title and pen name.
After generation, the skill checks text legibility, genre match, composition, platform fit and whether the title and pen name stay visible after cropping. If something fails, it can change the composition, colour, font or platform style and regenerate one image.
- Save location: BOOK_DIR/封面/ when a book folder exists, otherwise ./covers/{书名}/封面/
- Earlier versions are kept, and new files are numbered as 封面_vN.png
- A .prompt.txt file is saved next to each image
Safety notes: scripts and network access
This skill runs scripts and uses the network. The API fallback sends your prompt and any reference image to the API endpoint set by GPT_IMAGE_BASE_URL, which defaults to https://api.openai.com/v1. Only set a proxy URL if you trust that provider. The skill will not silently switch to the API if the built-in tool fails. It reports the error first.
The skill also scans your project folder for book directories. It skips folders that start with a dot, node_modules, 拆文库/ and 对标/. If you do not want your book files read, keep the skill out of that project or run prompt-only mode with the details typed in.
- Network: yes, the API fallback calls the configured image endpoint
- Scripts: yes, the crop step runs shell commands on your image file
- File access: it reads project files to find the title, genre and platform
Tips for better covers
Give the skill a clear genre and platform. The genre sets the style and the platform sets the upload ratio. For Fanqie, the skill uses 3:4 at 600×800. Other platforms use 2:3 unless you give the exact pixel size.
Treat the output as a first draft. AI images can have text errors or copyright issues, so check the title, the pen name and the platform rules before you publish.
- Name the platform before you ask, so the ratio is right from the start
- Ask for a close-up, a minimal design or a bold title if the first result is too busy
- Keep the .prompt.txt file so you can rerun or tweak the same idea later
What people use it for
- Draft a first cover for a Chinese web novel before commissioning a designer.
- Get a prompt-only set of cover concepts to refine by hand in an image editor.
- Create a 600×800 Fanqie upload copy from a 3:4 original.
- Compare several composition options for the same genre before picking one.
Frequently asked questions
How do I install the story-cover agent skill?
Download the ZIP from this page and unzip it into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder. In the Claude apps, go to Settings, then Capabilities, then Skills, and upload the ZIP.
What do I need to run story-cover?
In Codex CLI, the built-in ImageGen tool is preferred and needs no API key. The API fallback needs GPT_IMAGE_API_KEY. The skill also lists curl, jq and base64, and uses ImageMagick or macOS sips for the optional crop step.
Is story-cover free to use?
The skill is MIT licensed. The API fallback uses an OpenAI or compatible image API key, which may be billed by that provider. The source does not state pricing, so check your provider's terms.
Can story-cover make a cover without generating an image?
Yes. Ask for a cover prompt only, and the skill builds and delivers a full prompt without calling any image tool, key or output folder. Ask for an image when you want one.
Why does the title text look wrong or get cut off?
The skill relies on the image model to render Chinese text, so check the title and pen name in each result. Crop to the platform size and check that both remain visible. If they do not, ask for a different composition or a bolder title style.
How does story-cover compare with a designer or other cover tools?
story-cover is a fast way to produce a first draft and starting concepts. The source says AI output may need manual cleanup, and final commercial covers that need exact typography or licensed art still need a designer or manual editing.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
story-cover suits Chinese web-novel authors who want a quick first-draft cover, and writers using Codex CLI or Claude Code who already keep a book folder. Choose a different route if you need a final commercial cover with exact typography, licensed artwork, or if you do not want the skill reading your book files.
Download the skill