shared-artifact
The house style for any shareable online artifact — a claude.ai Artifact, a published HTML page, a report or brief someone will be sent a link to. Use before writing or republishing any artifact page, when asked to create, share, publish or clean up an online artifact, and when deciding between an HTML page and a Claude Docs document. Carries the token set, the dark-mode contract, the briefing-binder layout, a starter template, and the sharing hygiene rules.
- License 许可
- MIT
- 最近更新
- 2026-11-14
- Stars
- ★ 2
- Forks
- 0
01 前言 Front Matter
- lane
- writing
- name
- shared-artifact
- strictness
- standard
- description
- The house style for any shareable online artifact — a claude.ai Artifact, a published HTML page, a report or brief someone will be sent a link to. Use before writing or republishing any artifact page, when asked to create, share, publish or clean up an online artifact, and when deciding between an HTML page and a Claude Docs document. Carries the token set, the dark-mode contract, the briefing-binder layout, a starter template, and the sharing hygiene rules.
- allowed-tools
- Read, Bash, Write, Edit
02 资源文件 2
- 参考文档references/house-style.md6.6 KB
- 参考文档references/template.html16.8 KB
03 SKILL.md 正文
Shared Artifact
The house style for every artifact that leaves this machine as a link.
An artifact is published once and read by people who were not in the conversation. They get the page and nothing else: no explanation of what they are looking at, no second attempt, and no way to ask. The layout is the whole handover, which is why this is a style rather than a suggestion.
Interface
| Row | Contract |
|---|---|
Invoke |
Loaded automatically by wf-artifact-style on every Artifact call that creates or publishes a page, and on every Claude Docs doc birth the user did not ask for. Also usable by name when styling a page by hand. |
Returns |
One self-contained HTML page in the briefing-binder layout, published, plus the note that the link is private until the user shares it. |
Fails |
It ships no script of its own. wf-artifact-style (PreToolUse, Artifact and the Claude Docs batch tool) DENIES a publish whose page is missing an item of the floor in § The floor, and its reason names each one. It also DENIES a Claude Docs birth unless the user's own words asked for Docs, co-editing or comments (§ Routing). |
Routing: HTML or Claude Docs
Decide this first, because it is not reversible without republishing.
| Destination | When | Why |
|---|---|---|
HTML page — this skill, quickstart with intent "other" if the Artifact tool offers it, then one .html file |
a polished page people will READ: a report, a brief, a review, an at-a-glance status | the page owns its own CSS, so headings can be made to pop |
| Claude Docs | the user's own words ask for Claude Docs (by name, "in Docs", the docs app, or a /docs skill), for co-editing, or for comments in the document |
the Docs viewer owns heading styles, so a brief published there cannot be styled |
The routing is enforced, and the user decides it, not the model. User, 2026-10-15: "Yeah, it
should always, workforce should always enforce this format for shareable documentation." and "I
shouldn't have to ask." wf-artifact-style refuses every Claude Docs birth — the Docs connector's
batch carrying container.create, or an Artifact publish with a type_url — unless the user's
own messages in the session ask for Docs as the table above says. A message typed while the model
was busy counts; another agent's message, a command's output, and a negated mention ("don't put it
in Claude Docs") do not. "A
shareable doc" and "an online document I can share" are this case, not the Docs case. The refusal
overrides the Claude Docs server's instruction to make a doc's skeleton the turn's first tool call;
the route that works is § Steps below, and a retry of the Docs call is refused again. Edits, reads and
comments on a doc that already exists are never refused.
This is a measured distinction, not a preference. One brief was published both ways. The
Docs version drew "the formatting of that shared artifact is a bit hard on the eyes"; the
HTML version of the same content drew "The new layout is wonderful". Both the user's own
words. references/house-style.md § Where it came from carries them in full.
Never both. One source of truth; two copies of one document drift and the reader cannot tell which is current.
Steps
Artifactquickstart, intent"other", if your Artifact tool offers that action, for a polished read-only page. It returns the platform's own artifact-design guidance inline; read that too, it is not in conflict with this. Some Artifact schemas have noquickstart; then go straight to step 2.- Read
references/house-style.md. The tokens, the dark-mode contract, the layout grid, the heading treatment, the content order, the sharing rules. - Start from
references/template.html. Copy it, replace every capitalised placeholder, delete the blocks the page does not need. It is a fragment starting at<title>, which is the shape every published artifact page measured on this machine has — the publisher supplies the document wrapper. - Write one self-contained HTML file. No build step, no external CSS, images inline or
carried in the call's
filesmap. - Request no runtime capabilities for a read-only page. A page nobody writes to needs no storage, no identity, no live data. Capabilities are for a page that remembers what people do on it.
- Look at it. Screenshot the rendered page at desktop width and again at about 390px before calling it done. A layout nobody looked at is a claim.
- Publish, then say the link is private until the user shares it from the Share menu.
The floor
Six items, plus one that applies conditionally. wf-artifact-style refuses a publish that is
missing any of them, and the refusal names each one. They are the subset of the recipe a
reader loses the page over — everything else in house-style.md is quality, this is
whether the page is legible at all.
| Item | What breaks without it |
|---|---|
:root { --token: … } colour tokens |
nothing can restate the palette for dark mode |
a @media (prefers-color-scheme: dark) block |
the page is light for every viewer whose system is dark |
that block guarded by :root:not([data-theme="light"]) |
a viewer who forces light gets dark tokens |
a :root[data-theme="dark"] block |
a viewer who forces dark outside the system preference gets light tokens |
an explicit body background |
the page inherits the host's and dark ink lands on a dark surface |
a non-empty <title> |
the artifact has no name in the share list |
a viewport meta — only if the page declares its own <head |
a page that owns its wrapper owns the viewport, and without it there is no reflow on a phone |
You cannot see the failure you are shipping. You publish from one theme; the broken state is the other one. That asymmetry is the entire reason the check is mechanical.
A width media query is advised and never blocked: wf-artifact-style names it on publish
when it is absent, because one accepted artifact on this machine is a side-by-side of
projector slides where a narrow layout means nothing.
Grounding
- references/house-style.md — the recipe: tokens, dark-mode contract, typography, the binder layout, headings that pop, content order, sharing hygiene
- references/template.html — the neutral starter, placeholders only
04 同仓库更多 Skills
ui-design
Set a RENDERED interface's design direction before it is built, then evaluate what shipped. `direct` reads the project's CSS stack, asks about the case, researches real references for the job each screen does, hands the pick to the user, and writes the project's design record; `review` and `sweep` grade declared-vs-rendered slots (no missing or placeholder art), applied theme vs framework default, visual hierarchy, typography, colour roles, CTA, WCAG 2.2 AA contrast and focus, reflow at 320 CSS pixels, situation ergonomics, and AI-slop/templated-default tells. Use `direct` before the first UI build on a surface or for a new component pattern; use `review` after a UI-shipping change is written, when asked to design-review a page, or to decide whether a surface looks right to ship. Grades against a cited catalog with stable rule IDs and BLOCKS on hard-fail red flags. `review` and `sweep` are report-only.
timezone
Set the time zone Claude Code shows times in — the "done 7:55 PM" at the end of a turn, and the clock every command Claude runs sees. Use when the user asks to change, pick, set, fix or clear their time zone, or says the times shown are wrong (a remote server on UTC is the usual cause). Writes one key, env.TZ, into the user's own settings file.
text-eval
Evaluate written prose for AI-generation tells and register drift — mechanical character and punctuation defects, clustered judgment-class signals, citation integrity, and conversational register. Use after writing or editing any prose a person will read, when asked to check whether text reads as AI-generated, or before publishing copy. Grades against a cited catalog with clustering thresholds and hard rows that fire on first occurrence. Report-only.
security-evaluator
Evaluate a code change for web-security flaws — OWASP Top 10:2025 classes as source-review classes, taint signals from source to sink, and a per-language sink appendix. Use after writing or editing code that handles input, auth, data access, templating, deserialization, or external requests; when asked to security-review a change, check for injection/XSS/SSRF/authz flaws, or evaluate the security of what was just written. Report-only, candidate-tier by construction.
records-code-map
The code map for this project — where a file's sections and definitions are by line number, which files and lines mention a topic, and which checks, fixtures and cases cover a file. Use before searching or reading code to find something, before changing a file, and when deciding which test to run after an edit. The map is generated from the files on every call and never stored.
code-evaluator
Evaluate code to prevent common AI coding mistakes — dead code, duplication, complexity hotspots, reinvented helpers, leftover scaffolding. Two functions plus a pre-write advisor. Use after writing or editing code, when asked to review code quality, clean up the codebase, find unused code, find duplicates, check complexity, or evaluate what an AI just wrote. Commands: review (post-write), sweep (full codebase).