beautiful-feishu-whiteboard: 35 colour palettes for Feishu whiteboards, without a layout engine

It supplies palette and mood, and the agent still has to compose the layout

The repository is an agent skill, meant for Claude Code and other agents that read a SKILL.md file. It is explicitly not an automatic layout chart tool. Your agent decides what boxes, arrows and groupings the diagram contains. The templates then hand it a tasteful colour palette and mood, along with the hard rendering rules of the whiteboard medium.

That division of labour has a direct consequence. If you ask for a diagram and your agent produces a confusing structure, a different style will not repair it, because the style only controls how the structure looks. The quality of the arrangement comes from the agent and your prompt.

The result is described as a real, editable Feishu whiteboard embedded in a doc, not a screenshot. Anyone with access can move a node afterwards, which is the whole reason to go through the whiteboard renderer instead of exporting a picture. The project's own tagline calls the output gorgeous, but the useful claim is the editable part.

Native shapes only and no opacity: the Feishu SVG renderer sets the limits

The skill is built on knowledge of what the Feishu SVG whiteboard renderer can and cannot do, verified on actual boards and written down in RULES.md. The headline items are native shapes only, no opacity, and a text colour export quirk, with more in the file itself.

Two of these matter when you think about style. Without opacity, a palette cannot lean on translucent overlays or soft tints created by transparency, so each colour in a style has to be a solid value that you pick on purpose. And native shapes only means that anything the renderer cannot draw as one of its own shapes is off the table, however nice it would look in an ordinary SVG editor.

The text colour export quirk is the third trap. Colour you set on text may not come out the way you set it when the board is exported, which is why rules live in a separate file the agent reads rather than in the agent's memory of generic SVG. Treat RULES.md as the real specification and the palettes as skins on top of it.

Thirty-five styles in three tiers, all rendered with the same LLM training content

The gallery shows every style rendering identical content, the three stages of LLM training, so you can compare palettes directly. The 35 styles are grouped from restrained to bold, and each name links to a folder under templates/.

Restrained has nine: Avocado Press, Grove, Jade Lens, Long Table, Macchiato, Monochrome, Papier Bleu, Reading Room and Salmon Stamp.

Balanced has fifteen: Apricot Arc, Berry Pop, Bold Poster, Checker Bloom, Cobalt Bloom, Coral, Cut Bloom, Editorial Forest, Lime Slab, Linen Cut, Pin & Paper, Raw Grid, Riptide Cobalt, Soft Editorial and Violet Marker.

Bold has eleven: BlockFrame, Burst Panel, Confetti Wedge, Court Press, Crayon Stack, Grove Block, Mint Brut, Neo-Grid Bold, Riso Brut, Specimen Bold and Stencil & Tablet.

Because the sample content is always the same, the gallery shows you colour and mood but not how a style copes with your own content. A palette that looks calm on a three-stage pipeline may feel different on a dense architecture diagram, and you only find out by trying it. The repository also has CATALOG.md at its top level.

Three install routes, and only the npx route distinguishes project from global scope

The simplest route is to tell your agent to install the beautiful-feishu-whiteboard skill from github.com/zarazhangrui/beautiful-feishu-whiteboard. If you prefer to run the installer yourself, the npx skills command has a project level form and a global form:

# project level (./.claude/skills or your agent's skills dir)
npx skills add zarazhangrui/beautiful-feishu-whiteboard

# or globally (available in every project)
npx skills add zarazhangrui/beautiful-feishu-whiteboard -g

The -g flag is the only difference. Without it the skill lands in the project's skills directory, so it travels with that repository. With it, the skill is available in every project on your machine.

The third route is manual, cloning straight into your agent's skills folder:

git clone https://github.com/zarazhangrui/beautiful-feishu-whiteboard \
  ~/.claude/skills/beautiful-feishu-whiteboard

The manual clone path is written for Claude Code's ~/.claude/skills directory. If you use another agent, put the clone wherever that agent looks for skills, since the destination in the example belongs to Claude Code.

Writing boards needs lark-cli, Node.js 20 and an authenticated Feishu or Lark account

The requirements are firm. You need Node.js 20 or newer, and a Feishu or Lark account, because boards are written into your own tenant rather than rendered somewhere neutral. You also need lark-cli, which is the npm package @larksuite/cli, installed and authenticated:

npm install -g @larksuite/cli
lark-cli config init     # first run: scan the QR code
lark-cli auth login      # authorize your Feishu/Lark account

The first-run step involves scanning a QR code with the Feishu app, so it needs a person with a phone at hand. It is awkward to set up on a headless build server where nobody can scan anything.

The consequence of the tenant requirement is that this skill cannot produce a whiteboard for someone who has no Feishu or Lark workspace. Output goes into your account, so you should expect to see new content there and make sure the account you authorize is the one you want your agent writing into. The requirements list in the repository continues past these three items, so read the rest of it before your first run.

Editable output means a human can take over after the agent finishes

The reason this project insists on the SVG whiteboard renderer is editability. A screenshot of a styled diagram cannot be fixed by a colleague, but a real whiteboard inside a Feishu doc can. If the agent misplaces a label, you fix it in the doc rather than re-prompting.

The hard rules in RULES.md exist for that same reason. A board that avoids unsupported features can be opened and changed in Feishu without surprises, whereas one built from arbitrary SVG could lose elements on the way in.

The maintainer also publishes a Feishu doc version of the introduction, with usage scenario showcases, and a Chinese README at README.zh.md. Both are good places to check examples of finished boards before you spend time authorising lark-cli.

One limit to keep in mind: the skill gives the agent colours and rules, not a guarantee that the agent follows them. If the board comes out with the wrong look, check that the agent actually loaded SKILL.md and pointed at the template folder for the style you chose.

No releases and a June 2026 last push: pin the repository yourself

The repository has no GitHub releases, so there is no version number to pin and no changelog tied to a tag. The default branch is main, the license is MIT, and the repository is not archived. The last push was on 2026-06-11.

Because there are no tags, installing today and installing in three months can give you different files. Both the npx route and the manual git clone follow whatever main contains at that moment. If a particular set of templates works for you, copy the folder into your own repository or record the commit hash you cloned.

The top-level layout is small: assets/, scripts/ and templates/ directories, plus CATALOG.md, LICENSE, README.md, README.zh.md, RULES.md and SKILL.md. The primary language shows as Shell, which reflects the contents of scripts/ rather than any application code, since the product itself is the collection of styles and rules. Open scripts/ and read it before running anything it contains, as you would with any skill that executes commands on your machine.

Editorial conclusion

Use this skill if you already work with a coding agent and a Feishu or Lark account and want whiteboards that stay editable and look designed. Skip it if you expect automatic layout, if you cannot install and authenticate lark-cli, or if you need a tagged release to pin, since the repository has none.

Frequently asked questions

Official sources

1. Issues
2. License: MIT
3. README
4. zarazhangrui/beautiful-feishu-whiteboard on GitHub