Artifact Library 0.13.1

Changed

• Agents repeat what they read back. The skill, the agent guide and docs/agents.md tell agents to write out every question with the chosen answer and note, and every comment, mark and verdict, in their own reply before acting on it. Output of artifact feedback, artifact wait and MCP tools is often hidden from the user, and a permission check on the agent's next step only sees the conversation.
• The skill's MCP list names every tool (recent, versions, wait_for_feedback, lint).
• README: every feature through 0.13 (diagrams, annotations, Instant sync, the built-in blocks, Clean and Bold), with new screenshots in the default Clean style. docs/video/screenshots.py makes its own demo content, so they can be retaken on any fresh server.
• Docs: current image tags, link references for every release in this changelog, data-wait in the API capabilities, exit code 4 in the CLI reference.

Fixed

• artifact lint read diagrams as prose. Mermaid draws node labels as paragraphs inside the SVG, so a flowchart could raise paragraph-run, and a diagram's toolbar icons counted as extra diagrams. Text drawn inside an SVG is now part of the drawing, and an SVG inside a diagram, another SVG or a button counts once.
• Copies of references no longer call themselves a reference. The "Reference" label next to the page name (code review, code walkthrough, implementation review, results and evidence, test report, visual regression) and the "Example content" tags of the plan deck and decision record are reference-only, so they are hidden on pages made from them.
• Reference titles and descriptions no longer name old versions or the old Quiet theme.
• npm run build:seed leaves out __pycache__ and .DS_Store at any depth.

Install

# CLI (Linux, macOS)
curl -fsSL https://raw.githubusercontent.com/plugincompany/artifact-library/main/install.sh | sh
# CLI (Windows PowerShell)
irm https://raw.githubusercontent.com/plugincompany/artifact-library/main/install.ps1 | iex
# Server
docker run -d -p 3000:3000 -v artifact-data:/data ghcr.io/plugincompany/artifact-library:0.13.1

Verify downloads with SHA256SUMS. The -slim image (ghcr.io/plugincompany/artifact-library:0.13.1-slim) has no server-side previews.

Full Changelog: v0.13.0...v0.13.1

Artifact Library 0.13.0

Added

• Diagrams ([data-diagram], pre.mermaid, Kit.diagram(el, spec)): Mermaid text drawn on the page: flowchart, sequence, state, ER, class, gantt, timeline, mindmap, sankey, architecture, quadrant, xy chart, git graph, block and more, with the ELK layout for big graphs (data-layout="elk"). Coloured from kit tokens in Clean and Bold, light and dark, and redrawn when readers switch. Readers zoom (buttons, Ctrl + wheel, keys), drag to pan, open it full screen, view and copy the source and download the SVG. A syntax error shows in place with its line and logs a console error, so artifact lint reports it. Kit.diagram.render(text) returns an SVG for custom layouts.
• Syntax colouring with Shiki: on the artifact server, pre[data-lang] and [data-diff] are upgraded to VS Code's grammars (230+ languages) after the built-in colouring has drawn them, coloured with kit tokens (new --syn-t, --syn-o, --syn-p, --syn-r, --syn-tag). data-numbers (with data-start) adds line numbers and data-lines="4-6,9" marks lines, with either colouring. Kit.highlightAsync(text, lang).
• Image annotations (kit/kit-data.js): readers mark issues on screenshots. Every image opened in the kit's viewer has an Annotate button (or A): drag a box or tap a point, write what is wrong. img[data-annotate] also shows the marks on the page with an Annotate button. Saved as one thread per image, annotations.<id>; artifact feedback prints each mark with its number and region in percent and pixels, with replies under it. Agents answer with artifact data append … --item '{"text":"…","replyTo":"<mark id>"}', shown under the mark on open pages within about 10 s.
• Deleting items: DELETE /api/v1/artifacts/:slug/data/:key/items/:id, artifact data remove-item, MCP data_remove_item and Kit.data.removeItem(key, id) remove one thread item and the items replying to it. Readers use it to delete their own marks on a screenshot (Delete, then confirm).
• Shared browser libraries at /_artifact/libs/: Mermaid + ELK, Shiki and d3, bundled by npm run build:libs (scripts/build-libs.mjs) into libs/dist/ and shipped in the Docker image (LIBS_DIR). They are served once instead of being copied into every artifact's kit/, and kit.js loads only what a page uses; entry files are revalidated so upgrades reach published pages, chunks are immutable. Kit.libs.load("d3") gives d3 for one-off visuals and Kit.libs.color("--accent") a kit colour as hex. Without them (opened from disk, another host, <html data-kit-libs="off">) diagrams show their source and code keeps the built-in colouring. Licences in /_artifact/libs/LICENSES.txt.
• The style guide, the agent guide, the skill and the docs show when to use a diagram, the code options and annotations.

Changed

• The kit's viewer takes extensions (Kit.lightbox.extend), used by the annotations.
• artifact lint counts diagrams as visuals.

Install

# CLI (Linux, macOS)
curl -fsSL https://raw.githubusercontent.com/plugincompany/artifact-library/main/install.sh | sh
# CLI (Windows PowerShell)
irm https://raw.githubusercontent.com/plugincompany/artifact-library/main/install.ps1 | iex
# Server
docker run -d -p 3000:3000 -v artifact-data:/data ghcr.io/plugincompany/artifact-library:0.13.0

Verify downloads with SHA256SUMS. The -slim image (ghcr.io/plugincompany/artifact-library:0.13.0-slim) has no server-side previews.

Full Changelog: v0.12.0...v0.13.0

Artifact Library 0.12.0

Added

• Steps ([data-steps], data-steps-src="PLAN.todo", Kit.steps(el, spec)): a shared checklist as a kit block, for a runbook (phases of steps with copyable commands, the output to expect and "if this fails") or a to-do list the agent hands to the reader (owner such as You or Agent, due, and a link to where to do it). Everyone sees the same ticks with who ticked and when, and a note per step; done steps fold to one line; a sticky progress bar has "Jump to next step" and "Start a new run", which archives the run. side puts the ticks in a side menu and history lists archived runs. It lives in kit/kit-data.js, so it works in any page that loads it: standalone, or as one chapter of a plan or report. title, key, heading, bar, runs and labels adapt it; an array is taken as the steps; it returns next() and fires kit:steps ({done, total, run}). Documented with a live example in the style guide.
• Notes from the agent on a step, shown under it like on a question: updates: [{text, author, at, tone}] in the spec when the page is republished, or live: artifact data append <slug> comments.s-<id> --item '{"text":"…"}' --author "Claude Code" appears on open pages within about 10 s.
• Progress is saved in the format ref-runbook always used, so agents read it as before: state.checklist {runId, title, startedAt, startedBy, steps: {<id>: {done, by, at, note, noteBy, noteAt}}}, and archived runs as state.checklist.<runId> (plus archivedAt, archivedBy). key saves under another name, for a second list on one page.

Changed

• ref-runbook is built on the steps block. Its app.js keeps only the header, the facts, the seeding note and the rollback, so fixes to the checklist reach every page instead of each copy of the renderer. Artifacts already made from the old runbook keep their own copy and keep working.
• The skill, the agent guide and the docs tell agents to use [data-steps] for to-dos for the reader and procedures to run together, instead of copying a renderer.

Fixed

• install.sh replaces the binary atomically: it writes the new file next to the old one and renames it, instead of moving it from /tmp, which is a copy when /tmp is another filesystem. An agent that starts artifact during an upgrade gets the old or the new binary, never a half-written one.

Install

# CLI (Linux, macOS)
curl -fsSL https://raw.githubusercontent.com/plugincompany/artifact-library/main/install.sh | sh
# CLI (Windows PowerShell)
irm https://raw.githubusercontent.com/plugincompany/artifact-library/main/install.ps1 | iex
# Server
docker run -d -p 3000:3000 -v artifact-data:/data ghcr.io/plugincompany/artifact-library:0.12.0

Verify downloads with SHA256SUMS. The -slim image (ghcr.io/plugincompany/artifact-library:0.12.0-slim) has no server-side previews.

Full Changelog: v0.11.0...v0.12.0

Artifact Library 0.11.0

Added

• Resizable side menu. On wide screens, drag the right edge of a kit page's side menu (or focus the handle and use the arrow keys, Shift for bigger steps); double-click or Home resets it. The width is remembered for every kit page on the server and is limited to 200–560 px.
• Reference-only chrome is marked data-ref-only: the Collection button, "Example content" tags and the "Reference artifact · source …" footer of every reference. The kit shows them on the seeded references (ref-*, agent-guide, reference-collection, artifact-cli) and hides them on pages made from one, so a runbook or plan copied from a reference no longer links readers to the collection or calls itself a reference.

Changed

• The skill, the reference docs and the blank template tell agents to remove reference chrome and write their own footer (who prepared it, for whom, when, from what). The template no longer has a Collection button or a "Reference" label.

Fixed

• Scrolling up a long page could crawl. When the side menu followed the section being read, it used scrollIntoView, which also scrolled the page a few pixels (the page's scroll padding for the sticky bar counts against menu links near the top). Each nudge moved the reading line to the previous section and nudged again, so scrolling up a long runbook fought the reader for seconds. The menu now scrolls itself only; a 31-step runbook went from 19 page nudges per scroll up to none.

Install

# CLI (Linux, macOS)
curl -fsSL https://raw.githubusercontent.com/plugincompany/artifact-library/main/install.sh | sh
# CLI (Windows PowerShell)
irm https://raw.githubusercontent.com/plugincompany/artifact-library/main/install.ps1 | iex
# Server
docker run -d -p 3000:3000 -v artifact-data:/data ghcr.io/plugincompany/artifact-library:0.11.0

Verify downloads with SHA256SUMS. The -slim image (ghcr.io/plugincompany/artifact-library:0.11.0-slim) has no server-side previews.

Full Changelog: v0.10.0...v0.11.0

Artifact Library 0.10.0

Added

• Questions ([data-questions], data-questions-src, Kit.questions(el, spec)): several decisions on one page as a kit block. Each question has option cards with a summary, pros and cons, effort, a recommended tag and an optional picture (image, code, HTML or a sketch), what happens for each choice and a note; above them a progress bar with "Accept all recommended", beside them a jump list (from four questions), below them a summary table and "I'm done". It lives in kit/kit-data.js, so it works in any page that loads it: standalone, or as one chapter of a plan, review or report. heading, nav, toolbar, summary, finish and labels adapt it; it returns acceptAll(), finish() and answers() and fires kit:answers. Documented with a live example in the style guide.
• Notes from the agent on a question, shown under its context: updates: [{text, author, at, tone}] (or a plain advice string) in the spec when the page is republished, with agent as the default author, or live without a new version: artifact data append <slug> comments.q-<id> --item '{"text":"…"}' --author "Claude Code" appears on open pages within about 10 s. Newest last; older than the latest two fold away, resolved ones are hidden. Agents use it to answer a reader's note or explain a changed recommendation.
• Answers are saved in the format the questionnaire reference always used, so agents and artifact feedback read them as before: answer.<id> {questionId, question, choice, choiceLabel, recommended, recommendedLabel, followsRecommendation, note} and state.submitted {at, author, answered, total, unanswered, followsRecommendations}.

Changed

• ref-decision-questionnaire is built on the questions block. Its app.js keeps only the hero, the progress ring, the accept-all button and its own sketches, so fixes to the questions reach every page instead of each copy of the renderer. Artifacts already made from the old questionnaire keep their own copy and keep working.
• ref-code-review: "Decisions for you" uses the questions block (Built / Agreed / Overruled). Its answer.<decisionId> now has the standard shape above instead of {decision, choice, label, keeps, note}; followsRecommendation means the reviewer agrees with what was built.
• The skill, the agent guide and the docs tell agents to use [data-questions] whenever they need decisions from the reader, instead of copying a renderer.

Install

# CLI (Linux, macOS)
curl -fsSL https://raw.githubusercontent.com/plugincompany/artifact-library/main/install.sh | sh
# CLI (Windows PowerShell)
irm https://raw.githubusercontent.com/plugincompany/artifact-library/main/install.ps1 | iex
# Server
docker run -d -p 3000:3000 -v artifact-data:/data ghcr.io/plugincompany/artifact-library:0.10.0

Verify downloads with SHA256SUMS. The -slim image (ghcr.io/plugincompany/artifact-library:0.10.0-slim) has no server-side previews.

Full Changelog: v0.9.0...v0.10.0

Artifact Library 0.9.0

Added

• Lazy images: <img data-src="…" width height alt> loads when it comes near the screen (data-srcset/data-sizes too), with a placeholder while loading and a labelled "Image not found" box with its path when the file is missing. Printing loads every image first (Kit.images.loadAll()); img[loading=lazy] gets the same placeholder and fade.
• Full-size viewer: img[data-lightbox] opens an image full size; images with the same data-lightbox="<group>" page through together, with arrow keys, swipe and Esc. Scripts open it with Kit.lightbox.open(items, index).
• Galleries ([data-gallery], Kit.gallery(el, spec)) in six layouts: grid, masonry, strip, sidebar, annotated and compare. Write <figure>s with an image and a caption, or give JSON (or data-gallery-src pointing at a global) with titles, captions, tags and notes: pins or boxes placed in percent of the image.
• Code diffs ([data-diff], Kit.diff(el, spec)) from before/after code, a git diff patch or hunks, with word-level change marks, folded unchanged runs, notes and highlighted lines. Readers switch between Unified, Split and Result; the choice holds for every diff on the server, and phones get unified. range cuts a patch down to the lines that matter.
• Syntax colouring for pre[data-lang="ts"] and other languages (js, py, sh, css, html, json, sql, …), and Kit.highlight(text, lang) for your own markup. All four blocks are documented with live examples in the style guide.
• New reference ref-code-review: a change explained for a reviewer: why it exists, screenshots of the result in a gallery with notes, each change with its reasoning and short hand-picked diffs, alternatives and tests, decisions the reviewer confirms or overrules, and a verdict per change and overall, all saved for the agent. Unlike ref-code-walkthrough, which renders a whole diff by chapter, every snippet is chosen and trimmed by hand.
• artifact lint warns with eager-images when more than 12 images load with the page, and counts galleries and diffs as visuals.

Changed

• Clean is the default style. Every kit page opens in Clean; put data-kit-style="bold" on <html> to open a page in Bold. Readers still switch with the Bold / Clean button, and a style they picked before still wins. The references link kit/themes/clean/clean.css in their pages so Clean is there on the first paint. Artifacts already published keep the kit they were published with, so they don't change.
• publish.sh: KIT_THEME=bold ./publish.sh <dir> publishes a reference that opens in Bold; KIT_THEME=clean is no longer needed (it does nothing); other theme names still link kit/themes/<name>/<name>.css last.
• artifact lint loads every lazy image before checking, so a missing image is reported as missing-file wherever it sits on the page.

Install

# CLI (Linux, macOS)
curl -fsSL https://raw.githubusercontent.com/plugincompany/artifact-library/main/install.sh | sh
# CLI (Windows PowerShell)
irm https://raw.githubusercontent.com/plugincompany/artifact-library/main/install.ps1 | iex
# Server
docker run -d -p 3000:3000 -v artifact-data:/data ghcr.io/plugincompany/artifact-library:0.9.0

Verify downloads with SHA256SUMS. The -slim image (ghcr.io/plugincompany/artifact-library:0.9.0-slim) has no server-side previews.

Full Changelog: v0.8.0...v0.9.0

Artifact Library 0.8.0

Added

• artifact wait <slug>: waits until readers send an answer, comment or verdict, then prints it with a cursor for the next wait; agents run it in the background (in Claude Code, Bash with run_in_background) and hear back as soon as people respond, instead of the user having to tell them. changes by agents never end a wait; exit code 4 on --timeout (default 24h); network errors and server restarts are retried.
• Readers decide when answers reach the agent. The toolbar on every artifact has an Instant sync to agent switch (⋯ menu). It is on by default; it is one setting per browser for every artifact on the server. With it off, saves are held and the toolbar shows "N changes not sent" with a Send to agent button (state.sent). artifact wait honours it: it returns on a sent save, a Send or a page's "I'm done" (state.submitted), with everything saved since the start. --until submitted waits only for an explicit send, --until any for every save. artifact feedback marks held changes "not sent yet".
• MCP tool wait_for_feedback: the same wait in steps of up to 50 s, returning a cursor to continue from.
• Data documents saved by a reader holding changes carry held: true. The data list (GET /api/v1/artifacts/:slug/data) takes role=human|agent and wait=<seconds> (long-poll, at most 25 s, with since); servers advertise it as the data-wait capability. The CLI polls older servers every 5 s instead.
• The skill and the docs tell agents to start a wait after sharing a link that asks for answers.

Changed

• Licensed under the MIT License (was Apache-2.0).

Install

# CLI (Linux, macOS)
curl -fsSL https://raw.githubusercontent.com/plugincompany/artifact-library/main/install.sh | sh
# CLI (Windows PowerShell)
irm https://raw.githubusercontent.com/plugincompany/artifact-library/main/install.ps1 | iex
# Server
docker run -d -p 3000:3000 -v artifact-data:/data ghcr.io/plugincompany/artifact-library:0.8.0

Verify downloads with SHA256SUMS. The -slim image (ghcr.io/plugincompany/artifact-library:0.8.0-slim) has no server-side previews.

Full Changelog: v0.7.2...v0.8.0

Artifact Library 0.7.2

Changed

• Chapter menus list main chapters only. Sections nested inside a chapter (questions inside "Decisions", findings inside "Findings") are items of that chapter: the top bar leaves them out, and a side menu shows them as a sub-menu that folds open while that chapter is being read. data-toc-skip leaves any section out. artifact lint counts chapters the same way, so it no longer asks for every question in the menu.
• ref-decision-questionnaire wraps its questions in one "Decisions" chapter (#decisions), with its own question list for jumping between them.

Install

# CLI (Linux, macOS)
curl -fsSL https://raw.githubusercontent.com/plugincompany/artifact-library/main/install.sh | sh
# CLI (Windows PowerShell)
irm https://raw.githubusercontent.com/plugincompany/artifact-library/main/install.ps1 | iex
# Server
docker run -d -p 3000:3000 -v artifact-data:/data ghcr.io/plugincompany/artifact-library:0.7.2

Verify downloads with SHA256SUMS. The -slim image (ghcr.io/plugincompany/artifact-library:0.7.2-slim) has no server-side previews.

Full Changelog: v0.7.1...v0.7.2

Artifact Library 0.7.1

Changed

• The chapter menu stays on screen. The skill, the writing guide, the agent guide and the docs now make the sticky top bar (header.topbar > nav.topnav[data-toc]) the default place for the chapter menu, with a sticky side nav for long documents, and say never to put it in the page body.
• artifact lint reports a new error, nav-not-sticky, when the chapters are linked only from a menu that scrolls away; a sticky menu that links a few chapters but not most is a chapters-not-in-nav warning.

Fixed

• A nav.topnav[data-toc] placed in the page body is moved into the page's top bar by the kit.
• Top bars on wide screens: a long chapter menu no longer squeezes the brand or the controls. When the chapter links do not fit, they fold into the Contents dropdown (as on phones); menus of links to other pages fade at the edge instead of cutting a word off.

Install

# CLI (Linux, macOS)
curl -fsSL https://raw.githubusercontent.com/plugincompany/artifact-library/main/install.sh | sh
# CLI (Windows PowerShell)
irm https://raw.githubusercontent.com/plugincompany/artifact-library/main/install.ps1 | iex
# Server
docker run -d -p 3000:3000 -v artifact-data:/data ghcr.io/plugincompany/artifact-library:0.7.1

Verify downloads with SHA256SUMS. The -slim image (ghcr.io/plugincompany/artifact-library:0.7.1-slim) has no server-side previews.

Full Changelog: v0.7.0...v0.7.1

Artifact Library 0.7.0

Added

• artifact lint <slug>: the server renders a published artifact in its browser at 1440 and 390 px and reports concrete problems: script errors, missing files, sideways scrolling on phones, a missing title, and no global chapter menu (errors, exit code 3); walls of text, chapters without a visual, long runs of paragraphs, headings without ids, small text, images without alt text and a missing description (warnings). --screenshots <dir> saves the rendered pages so agents without a browser can look at them. Also POST /api/v1/artifacts/:slug/lint and an MCP lint tool; servers without previews answer LINT_UNAVAILABLE.
• Kit blocks for explaining: an automatic chapter menu ([data-toc]: top links, a side list, or a Contents sheet on phones, with scroll-spy), flow diagrams ([data-flow], Kit.flow), comparison tables with a recommended row (table.compare, tr.rec), risk tables ([data-risks], Kit.risks), before/after (.ba), "what changes / what stays the same" (.delta) and file lists under steps (.steps .files). All documented with live examples in the style guide.
• Writing guidance for agents: the skill opens with a ten-point bar every artifact meets (name the reader, conclusion first, a global chapter menu for three or more chapters, plain language and short paragraphs, a visual for every comparison, flow and set of numbers, measured evidence, decisions answered on the page, and a lint pass before sharing). A new references/writing.md explains each rule with before/after examples; the agent guide has two new chapters.

Install

# CLI (Linux, macOS)
curl -fsSL https://raw.githubusercontent.com/plugincompany/artifact-library/main/install.sh | sh
# CLI (Windows PowerShell)
irm https://raw.githubusercontent.com/plugincompany/artifact-library/main/install.ps1 | iex
# Server
docker run -d -p 3000:3000 -v artifact-data:/data ghcr.io/plugincompany/artifact-library:0.7.0

Verify downloads with SHA256SUMS. The -slim image (ghcr.io/plugincompany/artifact-library:0.7.0-slim) has no server-side previews.

Full Changelog: v0.6.0...v0.7.0