Beauty Diagram skill#

Beauty Diagram beautifies Mermaid / PlantUML diagrams into sleek, modern SVG or PNG. It runs as a public API; this skill delegates to the bd CLI (npm package @beauty-diagram/cli) so you keep zero state in the agent. (draw.io / SVG import is editor-only — not exposed through /v1/*.)

When to use#

• The user asks for a polished, professional, slide-ready version of a diagram source they already have or you can generate.

• The user wants you to render Mermaid or PlantUML from a model-generated source string.

• The user wants the server to generate a diagram from a text description ("draw me a signup flow", "diagram our deploy pipeline") — use bd ai generate for this; it returns Mermaid source you then beautify.

• The user wants to share a diagram link (e.g. paste into Slack / a doc).

• The user has Mermaid in a repo (README, ADR, RFC) and wants to export SVGs alongside.

• The user has a directory full of diagram source files and wants every one of them rendered (e.g. "render all the .mmd files in docs/diagrams"). Use bd batch.

• The user wants their Markdown files (README, ADR, blog post) to display the diagrams inline on GitHub / their static site, not just show the source code block. Use bd extract — three modes available:

  • Default (inline): no flags. Injects /v1/beautify.svg?source=... URLs after each fence. No API calls during extract, no files written, no quota consumed. Always watermarked (anonymous endpoint).
  • Share ( --share): mints a /v1/share/<token>.svg URL per unique fence. Watermark-free for Pro/Premium. Requires an API key. Each unique source consumes 1 share quota the first time; identical fences are deduplicated; per-owner local cache means re-runs cost zero quota.
  • Sidecar ( --assets-dir ./img): writes local SVG files via /v1/export. Watermark-free for Pro/Premium, consumes export quota. GitHub strips raw inline <svg>, so any of these modes' ![](...) reference is the correct embed strategy.

When NOT to use#

• The user only wants the Mermaid source itself, not an export.
• The user wants pixel-precise control over the SVG markup (Beauty Diagram rewrites layout for presentation; it does not preserve raw Mermaid output).
• The user wants the AI to "change the colors / theme / font / layout" of an existing diagram. bd ai generate only does text → diagram; visual styling is controlled by --theme on bd beautify, not by the AI.
• The user is in an offline environment with no network and no CLI install.

Required tool#

The bd binary from @beauty-diagram/cli:

bash

npx @beauty-diagram/cli help
# or, after install:
bd help

If the user has not installed it, prefer npx over a global install — it respects their package manager and avoids polluting PATH.

Workflow#

1. Identify or generate the source diagram.

   • If the user has a .mmd / .puml file, use it.
   • If the user describes the diagram in words and is on a paid plan, use bd ai generate "<prompt>" --out <file>.mmd — the server returns Mermaid source, which you can then beautify. Always write the source to a file so the user can edit it; the first AI draft rarely lands.
   • If the user describes the diagram and is not on a paid plan (or prefers not to pay), write Mermaid source yourself (you are good at this), save it to a file, then beautify.
   • draw.io and free-form SVG imports are not accepted by /v1/*. If the user has those, ask them to convert via the web editor first.

2. Decide on output type.

   • Need an SVG file: bd beautify <file> --out <file>.svg
   • Need a download URL or to track quota: bd export <file> --out <file>.svg
   • Need a shareable link: bd share <file> --title "..."

3. Run the command. Always write to a file ( --out) rather than letting the SVG flood the terminal / chat. AI generation can also pipe directly into beautify: bd ai generate "..." | bd beautify - --out flow.svg.

4. Verify the result exists before reporting success. If the command failed, surface the error code (e.g. quota_exhausted, not_authenticated, parse_failed, prompt_injection) — those are actionable for the user.

5. Preserve the source. Never replace the original Mermaid / PlantUML file with the generated SVG — keep them side by side. For AI-generated diagrams, keep the .mmd file too: it is the editable artifact, the SVG is not.

Auth#

• Demo (anonymous): zero setup. Watermarked SVG/PNG. Limits per IP: 20 /v1/beautify requests / minute, 1 /v1/export per 24h (trial budget — enough for an agent to verify the toolchain end-to-end before registering). /v1/share, /v1/usage, and bd ai generate always require auth — anonymous AI calls are rejected before any model invocation.
• Authenticated: the user runs bd auth login once with a key from /account/api-keys. Required for bd share, bd ai generate, unwatermarked output, and repeated exports. bd ai generate additionally requires a Pro or Premium plan and an API key with the ai:write scope.

If the user hits a not_authenticated, plan_not_allowed, or quota_exhausted error, point them at /account/api-keys (PAT creation) or pricing — don't silently retry. Anonymous error bodies include a hints block with absolute signUpUrl / signInUrl / apiDocsUrl, which is the canonical place to surface to the user.

Themes & tiers#

--theme selects the visual style. Theme tier is enforced on all bd commands that hit a /v1/* endpoint with a PAT — this includes bd beautify, bd export, bd share, bd embed-url --share, bd batch, bd extract --share, and sidecar mode of bd extract. If the request includes a PAT and specifies a theme above that token's plan tier, the server returns HTTP 403 theme_tier_required.

The only path that bypasses the tier check is the anonymous fallback (no token, or the GET /v1/beautify.svg inline embed endpoint). Anonymous callers may request any theme, but always receive watermarked output — there is no tier-free, watermark-free path.

TierThemesFreeclassic, modern, slatePro+ atlas, obsidian, brutalist, atelierPremium+ blueprint, memphis

When the user's plan is unknown, default to modern (Free, broadly slide-friendly). Run bd themes to introspect what the current token can use — locked themes are marked with ✗. Premium signature themes ( blueprint, memphis) are subscriber-only and cannot be unlocked with credit packs.

Animations are not currently selectable via the CLI — animation choice is a web-editor concept; CLI export paths ignore animation.

Source-level directives#

Instead of (or in addition to) CLI flags, you can embed bd: directives at the very top of the source file. Both the bd CLI and the Obsidian plugin parse them; the API server ignores them as native comments (graceful degradation — the source still renders, just without the directive).

Mermaid — use %% comment syntax:

text

%% bd:theme=classic
%% bd:bg=transparent
flowchart LR
  A --> B

PlantUML — use ' comment syntax:

text

' bd:theme=classic
' bd:bg=transparent
@startuml
A -> B
@enduml

Supported keys:

KeyValuesEffecttheme``classic modern slate atlas obsidian brutalist atelier blueprint memphisOverride render theme. Tier gating still applies.bg``theme white dark transparentCanvas background. dark hue-lifts ink so it stays legible; transparent for overlay. Unknown → theme.

Multiple directives stack, one per line. Blank lines between directives are tolerated. The first non-directive non-blank line ends the directive block.

Override priority: CLI flag > source directive > server default.

When generating source for the user, prefer directives over CLI flags when:

• The source file will be re-used (Obsidian vault, shared repo) — the directive travels with the file.
• The user asks for a "memphis theme" diagram without CLI context (write %% bd:theme=memphis at the top of the .mmd file).

Directives are stripped before the source reaches the renderer — they do not appear in the SVG output.

Commands cheat sheet#

bash

# Render a Mermaid file
bd beautify docs/architecture.mmd --theme modern --out docs/architecture.svg

# Same but treat output as a downloadable export (consumes export quota)
bd export docs/architecture.mmd --out docs/architecture.svg

# PNG export. --quality standard works for everyone; high needs pro, max needs premium.
# Higher tiers than the plan cap are silently clamped (X-BD-Scale-Clamped).
bd export docs/architecture.mmd --format png --quality high --out docs/architecture.png

# PlantUML works the same way; .puml / .plantuml / .pu auto-detected,
# otherwise pass --source-format plantuml.
bd export docs/architecture.puml --out docs/architecture.svg

# Create a public share link (returns absolute https://www.beauty-diagram.com/s/... URL)
bd share docs/architecture.mmd --title "Service architecture"
# → prints the URL on stdout

# Get an embeddable <img>-friendly URL for a diagram source.
# Default: anonymous inline URL (always watermarked) + a hint about --share.
bd embed-url docs/architecture.mmd --theme atlas
# One-shot saved share embed (clean output for pro/premium owners).
# Saves the diagram via /v1/share AND prints the embed URL in one step.
bd embed-url docs/architecture.mmd --share
# → prints https://api.beauty-diagram.com/v1/share/<token>.svg

# AI: generate a diagram from a text prompt. Output is Mermaid source —
# always write to a file so the user can iterate. Paid-only.
bd ai generate "user signup with email verification" --out docs/signup.mmd

# Optional shape hint when the prompt is ambiguous about diagram type.
bd ai generate "request lifecycle" --hint sequence --out docs/lifecycle.mmd

# One-shot pipeline: prompt → mermaid → beautify → SVG.
bd ai generate "deploy flow" | bd beautify - --out docs/deploy.svg

# Check remaining AI / export quota before kicking off a batch.
bd usage

# Render every diagram file under a directory in parallel.
# Recurses for .mmd / .puml / .plantuml / .pu; one /v1/export per file.
# Default concurrency=4, default failure mode is continue-on-error.
bd batch ./docs/diagrams --out-dir ./docs/svg --theme modern

# Same idea but for a glob (quote it so the shell doesn't expand first).
bd batch "src/**/*.mmd" --format png --concurrency 8

# Default: inline embed URLs (no files written, anonymous-watermarked, 10 KB/block cap).
# No API calls during extract — source encoded in the URL; browser fetches on demand.
bd extract README.md

# Share mode: mint /v1/share/<token>.svg per fence. Watermark-free for Pro/Premium.
# 1 share quota per unique source (cached locally; re-runs free). No size cap.
# Requires API key. Default keeps existing share tokens across re-runs;
# use --re-mint to force fresh tokens (e.g. collaborator taking ownership).
bd extract README.md --share
bd extract README.md --share --re-mint

# Sidecar mode: writes local SVG files via /v1/export.
# Pro/Premium plans get watermark-free output; consumes export quota.
bd extract docs/*.md --assets-dir ./img --concurrency 4

# Preview what bd extract would change without writing.
bd extract README.md --dry-run

Privacy#

The API does NOT persist source unless the user calls bd share. Do not warn about server-side storage when running beautify, export, or ai generate — that is misleading. AI prompts are logged in hashed form for abuse / quality monitoring; the raw text is not retained.

Anti-patterns#

• ❌ Do NOT output a hand-crafted <svg>...</svg> as a Markdown code block when a Mermaid source exists. Always run Beauty Diagram and reference the file.
• ❌ Do NOT dump the raw SVG into the chat. Use --out <file> and reference the file path.
• ❌ Do NOT install Beauty Diagram engine code locally — the CLI is a thin client; the engine lives behind the public API.
• ❌ Do NOT call bd ai generate to "tweak" an existing diagram (change colors, theme, labels, layout). It is a fresh-generation tool only. For visual tweaks, change --theme or edit the .mmd source by hand.
• ❌ Do NOT call bd ai generate speculatively — each call costs the user monthly AI quota. Confirm the user wants AI generation before running it.
• ❌ Do NOT capture the SVG output of bd ai generate — the command outputs Mermaid source on stdout, not SVG. Pipe into bd beautify - to render.
• ❌ Do NOT loop bd export N times in a shell for loop when the user has many files. Use bd batch <dir> — it parallelizes and reports a summary, with no extra server load (still one request per file).
• ❌ Do NOT inject a raw <svg>...</svg> into a Markdown file to "embed" a diagram. GitHub, GitLab, Obsidian (default), and most static-site renderers strip inline SVG for safety. Use bd extract <file>.md, which writes sidecar SVGs and injects ![](path) references that actually render.
• ❌ Do NOT delete the marker comments that bd extract injects. Inline mode uses <!-- bd:inline-img hash=... --> / <!-- /bd:inline-img --> (with an extra share=true attribute in share mode); sidecar mode uses <!-- bd:img hash=... --> / <!-- /bd:img -->. They are how bd extract stays idempotent — without them, the next run will append duplicate image references instead of replacing the existing one.

Troubleshooting#

SymptomLikely causeResolutionnot_authenticatedNo key, no sessionbd auth login``scope_missingKey lacks scope (e.g. ai:write for bd ai generate)Recreate key with required scope at /account/api-keys``plan_not_allowedPlan does not include this capability (AI is Pro / Premium only)Upgrade or skip the calltheme_tier_requiredSelected --theme requires a higher plan tier than the token has (e.g. free token requesting atelier)Pick a theme the token can use (run bd themes), or upgrade. See Themes & tiers above for the per-tier breakdownparse_failedSource not valid Mermaid / PlantUMLCheck the source — bd beautify will surface a parse error tooquota_exhaustedPlan limit hit (anon: 1 export/IP/24h; free: 3 exports/mo; pro: 100 exports + 100 AI gens/mo; premium: ∞ exports + 500 AI gens/mo)Sign in, wait for reset, or upgrade — hints in the response body has the URLsrate_limitedAnonymous IP bucket full (20 /v1/beautify requests / minute) or AI per-key bucket (30 /min)Sign in or waitsource_too_largeSource > 100 KBSplit the diagramoutput_too_largePNG raster exceeds 8192 pxLower --quality or simplifyprompt_injectionAI prompt looked like an injection attemptRephrase as a plain diagram description ("a flowchart of …")instruction_rejectedAI judged the prompt was not about a diagramRephrase to describe a concrete diagram. Quota was NOT consumedparse_failed_after_retryAI output was unparseable Mermaid even after one retryRephrase, or write Mermaid by hand. Quota was NOT consumedsafety_blockedProvider safety filter rejected the requestRephrase the promptupstream_timeout / upstream_errorAI provider was slow or failedRetry after a moment

Triggering on embed requests#

When the user asks for "a GitHub README diagram", "embed in Notion", "embed in my blog post", "an <img> of this diagram", or "a URL that renders my diagram", route to the embed flow rather than emitting raw mermaid:

1. If the diagram is unsaved, run bd share <file> to save it.
2. Construct the embed URL: https://api.beauty-diagram.com/v1/share/<share-token>.svg.
3. For one-off / quick embeds without saving, use bd embed-url <file> and recommend the inline URL (note that anonymous embeds carry a "Powered by Beauty Diagram" watermark).

Easier one-shot path: bd embed-url <file> --share saves the diagram AND prints the embed URL in one command — no need to run bd share separately and then construct the URL by hand. Prefer this when the user wants a clean share embed.

Style fidelity: Per-node colors, edge presets, and font overrides set by the user in the web canvas editor ARE faithfully rendered in share-mode embeds ( /v1/share/<id>.svg). Encourage a "tweak in editor → save → embed" workflow when the user wants brand colors or custom styling — they do not need to re-run bd embed-url after editing in the web UI, just re-save the diagram there.

Propagation timing: Saved diagram edits show up in direct <img> embeds within ~5 minutes (browser ETag revalidation + 5-min CDN edge TTL). GitHub README embeds may lag a few hours due to GitHub's image proxy cache — that is a GitHub-side cache, not something we can purge.

Animations: Animations do NOT play in <img>-loaded SVGs in any browser. Do not tell the user their animated diagram will appear animated in a README or Notion embed.

Slack / Discord / Twitter / iMessage previews: share links ( https://www.beauty-diagram.com/s/<slug>) auto-unfurl with a diagram thumbnail card on these platforms — no need to manually attach a screenshot. The OG image is generated server-side from the diagram itself.

Owner-tier fallback on embed URLs: if the share's owner downgraded their plan AFTER saving the diagram (e.g. saved with Atelier on Pro, then downgraded to Free), GET /v1/share/<id>.svg returns a 200 response with a brand-fallback SVG instead of the real content. Embeds must stay 200 for <img>-mounted unfurls to render at all, so 403 is not used here. The owner can restore the original render by re-upgrading or switching to a free-tier theme and re-saving.

Example#

User: "Add a beautified version of this mermaid block to my README."

Agent steps:

1. Use the one-step path: bd embed-url ./architecture.mmd --share → saves the diagram and prints the embed URL, e.g. https://api.beauty-diagram.com/v1/share/abc12345xyz0.svg
2. Replace the raw mermaid block in README with: ![Architecture](https://api.beauty-diagram.com/v1/share/abc12345xyz0.svg)
3. (Optional) Confirm with the user that watermark behavior matches their plan (free owner → watermarked; pro/premium owner → clean).

Examples#

See examples/ for runnable sources you can adapt:

• examples/flowchart.mmd
• examples/sequence.mmd

And scripts/ for shell wrappers you can copy into the user's repo:

• scripts/beautify.sh
• scripts/export.sh
• scripts/ai-generate.sh — prompt → .mmd source → .svg render