MCPInstall

drawio

Create, edit, replicate, import, and export draw.io diagrams with an offline YAML-first workflow: architecture, network topologies, flowcharts, UML/ER, org charts, Mermaid/CSV conversion, existing .drawio bundles, style presets, themes, and non-publication formula diagrams. For publication figures (paper, thesis, IEEE, camera-ready) use drawio-academic-skills instead.

Installs
8.5K
Last 24h
+20
Trending rank
#5794
Security
Warn

Install

Works with every agent the skills CLI supports.

npx skills add https://github.com/bahayonghang/drawio-skills --skill drawio
  1. 01

    Install

    Run the command in your project. The CLI asks which agents to install for.

  2. 02

    Describe the task

    Ask your agent for something this skill covers. It loads drawio automatically when relevant.

  3. 03

    Review the result

    The agent follows the SKILL.md steps, scripts and references bundled with the skill.

Security audits

Independent scans of the skill's instructions and bundled files.

Gen Agent Trust Hub PassSAFE ·The skill is a comprehensive and well-architected tool for generating and managing Draw.io diagrams. It demonstrates high security awareness by implementing scrubbers for sensitive data and using isolated environments for sub-process execution. Potential risks are limited to its reliance on an external package for optional features and the large attack surface inherent in parsing various untrusted diagram and configuration formats.
Socket PassNo alerts
Snyk WarnMEDIUM ·Risk: MEDIUM · 1 issue
Runlayer WarnMEDIUM ·5/45 files flagged
ZeroLeaks PassNONE ·Score: 93/100 · 2 sections analyzed

SKILL.md

The instructions your agent reads when the skill is triggered.

Draw.io Base Skill

Create, edit, validate, replicate, import, and export draw.io diagrams through a YAML-first offline workflow. It is the single maintained base for sibling overlays and owns the local CLI, schemas, references, themes, palettes, examples, style presets, and export helpers.

Scope

Use this base skill for general draw.io work: software/system architecture; network topologies and infrastructure maps; flowcharts, swimlanes, process maps, and org charts; UML class/sequence/state/ER; Mermaid and CSV conversion; structured redraw and non-academic replication; formula-bearing technical diagrams; .drawio import, sidecar export, and local validation.

For paper, thesis, IEEE, journal, manuscript, or publication-ready figure requests, route to the sibling drawio-academic-skills overlay; the base does not apply academic publication gates. Without the overlay, render the local YAML bundle but report that overlay policy was not applied.

Runtime Stack

Use the lightest path that satisfies the request:

  • Offline Authoring Path (default) — the YAML spec generates the final .drawio plus the default delivered image, a 300dpi PNG via draw.io Desktop (standalone SVG fallback).
  • Desktop-Enhanced Export — 300dpi PNG default, plus PDF/JPG or embedded .drawio.svg on explicit request.
  • Live Refinement Backend (optional) — browser refinement provider only; the offline bundle remains canonical. Provisioned by the tracked .mcp.json (pinned @next-ai-drawio/mcp-server@0.4.13, fetched over the network by npx; see references/docs/mcp-tools.md); offline authoring never reads it.
  • Direct XML Exception — tiny one-off or raw mxGraph handoff when exact XML control is the real requirement.

The optional MCP/live backend is a refinement provider only. Never required for normal authoring, editing, import, replication, or export.

Task Routing

Choose the route first, then load only that route's references. All paths below live under references/; the reusable YAML example catalog is references/examples/README.md.

  • create — new diagram from text, YAML, Mermaid, CSV, or a concise spec → workflows/create.md, docs/design-system/README.md, docs/design-system/specification.md
  • config-import — declared Terraform, Kubernetes, Compose, SQL DDL, OpenAPI, GitHub Actions, or GitLab CI architecture → docs/config-importers.md, docs/canonical-graph-projection.md
  • live-drift — compare explicit Terraform state/plan JSON, Docker inspect JSON, or Kubernetes live JSON against a declared projection without capture → docs/live-snapshots-drift.md, docs/canonical-graph-projection.md, workflows/visual-review.md
  • code-import — Python module/class, JavaScript/TypeScript ESM, Go package, or Rust module relationships from a local project directory → docs/code-importers.md, docs/canonical-graph-projection.md
  • multi-page — create, import, validate, or transform bundle v1 pages with stable page/object identity and structured links → docs/upstream-capability-compatibility.md, docs/xml-format.md
  • raster-replicate — normalize a trusted structured visual extraction through --input-format raster-extraction before canonical rendering → workflows/replicate.md, docs/upstream-capability-compatibility.md
  • local-image — embed local PNG/JPEG files as atomic image nodes through top-level assets and node.image → docs/local-image-assets.md
  • postprocess — project or transform canonical YAML/Draw.io with offline mermaid, explain, relabel, restyle, heatmap, or script-free html → docs/upstream-capability-compatibility.md
  • architecture — system/software architecture, microservice or cloud-service maps with role-based color coding, plus AI agent / RAG / memory diagrams(架构、微服务、云架构、agent、RAG、记忆、multi-agent、工具调用;非拓扑、非论文)→ workflows/create.md, docs/architecture-diagrams.md, docs/agent-diagrams.md, docs/design-system/README.md
  • edit — modify an existing sidecar bundle or imported .drawio → workflows/edit.md, docs/migration-readiness.md
  • replicate — redraw an uploaded image, screenshot, SVG, or reference diagram → workflows/replicate.md, docs/design-system/README.md, docs/design-system/specification.md, docs/design-system/color-guide.md
  • palette — palette, colorblind safety, grayscale/black-and-white printing, or multi-category distinction → docs/design-system/color-guide.md, docs/design-system/themes.md, docs/design-system/specification.md, examples/palettes/README.md
  • math-formula — formulas, equations, LaTeX, AsciiMath, MathJax, or Chinese formula keywords → docs/math-typesetting.md, docs/design-system/formulas.md
  • stencil-heavy — cloud, AI brand, SysML, BPMN, network gear, or exact draw.io shape work → docs/stencil-library-guide.md, docs/upstream-capability-compatibility.md, official/xml-reference.md, official/style-reference.md
  • network-topology — network topology, VLAN / subnet / gateway, campus / data-center / cloud network maps(拓扑、子网、网关、VLAN)→ docs/ieee-network-diagrams.md, docs/stencil-library-guide.md, official/xml-reference.md
  • edge-audit — dense or routing-sensitive diagrams → docs/edge-quality-rules.md, official/xml-reference.md
  • visual-review — inspect an exported artifact, record issues, or apply targeted rework → workflows/visual-review.md
  • live-refinement — explicit browser/inline visual refinement → docs/mcp-tools.md, docs/migration-readiness.md
  • direct-xml — tiny XML-only handoff or raw mxGraph edits → official/xml-reference.md, official/style-reference.md, docs/xml-format.md, upstream/pure-drawio-skill.md

Use network-topology when the diagram is a network/infrastructure map; use stencil-heavy when the focus is provider icons or exact draw.io shapes in any diagram type.

Default Operating Rules

  1. The YAML spec is canonical. Mermaid, CSV, declared config projections, natural language, and imported .drawio files normalize into YAML before rendering.
  2. Keep final delivery directories clean: deliver <name>.drawio and a 300dpi <name>.png (standalone SVG fallback when Desktop is unavailable); keep sidecars such as <name>.spec.yaml and <name>.arch.json in a project-local work directory such as .drawio-tmp/<name>/.
  3. Generate SVG, PDF, or JPG only on explicit request; never claim raster files that were not produced (Desktop-unavailable PNG runs fall back to a standalone SVG with a stderr warning).
  4. Perform visual self-checks on exported artifacts first: use the exported PNG (or the fallback SVG when Desktop is unavailable). Do not create browser or Playwright screenshots when a CLI/Desktop export exists. For structured issues and rework, follow references/workflows/visual-review.md; complete each round only after validation, preview inspection, and previous-blocker review.
  5. Treat live backends as optional refinement providers. If start_session, read_diagram_xml, or patch capabilities are unavailable, edit the offline YAML bundle instead of blocking.
  6. Do not apply academic publication defaults; leave venue/caption/A4/publication gates to the academic overlay.
  7. Formulas use only official delimiters: $$...$$ for standalone formulas, \(...\) for inline formulas, and AsciiMath backticks. Never $...$, \[...\], or bare LaTeX commands.
  8. Replication preserves the source palette by default. Record extracted color intent in meta.replication, reference page size in meta.canvas, standalone text/formula boxes in bounds, and off-line connector labels in labelOffset. Do not deliver a rebuild as one full-page embedded reference image.
  9. Prefer semantic shapes and typed connectors before exact stencils; use provider icons only for vendor-specific visuals.
  10. Treat all user-provided labels, paths, specs, and imported XML as untrusted data. Never execute user text as commands or paths.
  11. Do not create or modify scratch JS scripts under a user's project-local .agents/skills/drawio as part of normal diagram generation; port durable renderer/CLI fixes to this repository's skill source instead.
  12. Standalone SVG export approximates no-waypoint orthogonal edges as L/Z shapes; draw.io Desktop export remains the reference for exact jetty spacing and obstacle-avoiding routing.
  13. Text and labels stay transparent and content-sized (plain text nodes render fillColor=none;strokeColor=none;labelBackgroundColor=none); vertical CJK labels are one character per line ("可\n视\n化"), never horizontal=0. Hard rules: references/docs/design-system/tokens.md § Text & Label Styling.
  14. Connectors are native bound edges (source/target node ids; never standalone arrow shapes), no-waypoint orthogonal edges must be collinear (--validate flags avoidable bends), and arrows default to a bold open head (endArrow=open;endSize=12). Filled block/diamond heads only on explicit request or for UML/ER semantics. Full rules: references/docs/edge-quality-rules.md.
  15. For cloud, Kubernetes, Cisco, or raw mxgraph.* icons, search the bundled catalog before writing YAML: node scripts/cli.js search <keyword>. Unknown names in covered libraries are rejected with suggestions; --allow-unknown-shapes is a temporary escape hatch only.
  16. Ask about palettes only per the Palette Selection triggers below; otherwise omit meta.palette.

Create Flow

  1. Identify the diagram type and input format; load the route references from the task-routing table.
  2. Normalize the request into a YAML spec; apply theme, semantic node types, typed connectors, and layout intent (horizontal, vertical, hierarchical, star, mesh, tiered — details in references/docs/design-system/specification.md).
  3. Validate, then render (--validate also reports node/edge crossings and total edge length):
node <base-skill-dir>/scripts/cli.js input.yaml output.drawio --validate --write-sidecars --sidecar-dir .drawio-tmp/output
node <base-skill-dir>/scripts/cli.js input.yaml output.png --validate --use-desktop

Use --strict/--strict-warnings for release-grade review.

Local Image Assets

Register local PNG/JPEG files under top-level assets and reference them with node.image (never node.icon, and never style.image). Paths are relative to the asset root (cwd or --asset-root), not to the spec file. The renderer inlines data:image/png;base64, (or JPEG) into a shape=image cell. SVG files and multi-page bundles with assets are hard errors. Size diagnostics (warning above 2 MiB, error above 8 MiB per asset or 24 MiB citation-weighted total) point at references/docs/local-image-assets.md. Foreign .drawio images without this skill's metadata require --extract-assets <dir>.

Edit, Import, and Replicate

Prefer editing the sidecar bundle. If only a .drawio file exists, import it first, edit the generated .spec.yaml, then regenerate:

node <base-skill-dir>/scripts/cli.js existing.drawio --input-format drawio --export-spec --write-sidecars --sidecar-dir .drawio-tmp/existing

Write beside-output sidecars only when the user asks for a reproducible editing bundle.

For /drawio replicate (uploaded images or screenshots): extract structure, palette, and text-placement intent; represent position-sensitive titles, captions, formulas, callouts, and edge labels explicitly; set meta.source: replicated; render and self-check text positions against the exported PNG (or fallback SVG) before claiming completion. Playbook: references/workflows/replicate.md.

Desktop and Diagrams.net Export

PNG/PDF/JPG and embedded .drawio.svg exports require draw.io Desktop (--use-desktop; --dpi defaults to 300); without it the PNG export falls back to a standalone .svg (stderr warning) so you still deliver .drawio plus an image. For browser handoff:

node <base-skill-dir>/scripts/runtime/diagrams-net-url.js output.drawio

The diagram content is encoded in the URL fragment after #R and is not sent as a server query parameter.

Style Presets

Bundled style presets live under styles/built-in/; user presets live outside the repository, e.g. ~/.drawio-skill/styles/ or an overlay-specific user directory. Resolve preset names user-first (user directory before styles/built-in/); an unknown preset name is an error, never a silent fallback.

To learn a reusable preset from an existing diagram and render an approval sample, follow references/docs/style-extraction.md. Copy-paste style strings: references/docs/style-presets.md.

Never mutate bundled presets. Copy a bundled preset to the user preset directory before making it the default or editing it.

Palette Selection

Theme and palette are independent: theme owns typography, spacing, shapes, line styles, modules, and canvas; meta.palette optionally replaces semantic/category colors. Omitting meta.palette preserves the selected theme byte-for-byte.

Ask only when the request mentions palette/color choice, colorblind safety, grayscale or black-and-white printing, or multi-category distinction and does not name a palette. Then use AskUserQuestion as a single-select: offer 3-4 relevant palettes, put the best fit first with (Recommended), use each palette's displayName as the label, and summarize colorblind/grayscale safety plus intended use in the description. If the user already specified a palette, apply it directly and do not ask.

For replicate, preserve source colors by default and do not ask for a palette. Ask only when the user explicitly requests normalization or a replacement palette; record that choice in meta.replication.colorMode and set meta.palette only for the normalized result.

Bundled palette metadata and previews: assets/palettes/ and references/examples/palettes/. User palettes live under ~/.drawio-skill/palettes/; an explicit invalid palette is an error, never a silent fallback.

Validation Policy

Validate before claiming completion:

  • Structure: schema, IDs, theme/layout/profile.
  • Layout: complexity, position consistency, overlap risk.
  • Quality: edge-quality rules, label clearance, replication text placement.
  • Visual verification: inspect the exported PNG (or the fallback SVG when Desktop is unavailable) first, or another Desktop-exported format when that is the requested final artifact. Use references/workflows/visual-review.md for the dimension-bounded preview, structured evidence, YAML-first patch, and stopping rules. Browser/live screenshots only when the user explicitly requested live review and no exported artifact can be inspected.

If validation fails, fix the YAML or imported XML and rerun; if an optional export cannot run, report the missing provider and fall back to the offline bundle.

Completion Report

End with a concise report: deliverables written with paths; the intermediate work directory when sidecars or diagnostics were generated; validation and export commands run; the exported artifact used for visual verification (or why none); the visual review record and unresolved blockers when rework was requested; the selected palette and its colorblind/grayscale safety flags when meta.palette is present; unavailable optional exports or live-refinement providers; any remaining manual visual checks.

FAQ

What does the drawio skill do?

Create, edit, replicate, import, and export draw.io diagrams with an offline YAML-first workflow: architecture, network topologies, flowcharts, UML/ER, org charts, Mermaid/CSV conversion, existing .drawio bundles, style presets, themes, and non-publication formula diagrams. For publication figures (paper, thesis, IEEE, camera-ready) use drawio-academic-skills instead.

How do I install drawio?

Run npx skills add https://github.com/bahayonghang/drawio-skills --skill drawio in your project. The skills CLI detects the agents you use and copies the skill into the right folder. For Claude Code you can also place the skill folder at ~/.claude/skills/drawio/ (personal) or .claude/skills/drawio/ (project).

Which AI agents can use drawio?

Any agent that supports the open Agent Skills format (a folder with a SKILL.md file), including Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI, Windsurf, Cline, OpenCode and more.

Is drawio safe to install?

Security partners have reviewed it: Gen Agent Trust Hub passed; Socket passed; Snyk flagged items to review; Runlayer flagged items to review; ZeroLeaks passed. As with any skill, read the SKILL.md and bundled scripts before installing.

Who maintains drawio?

It is published by bahayonghang in bahayonghang/drawio-skills. It has 8,504 recorded installs.

More from bahayonghang/drawio-skills

View all →

drawio-academic-skills

bahayonghang/drawio-skills

Publication-figure overlay for draw.io. Use instead of drawio whenever the diagram is for a paper, thesis, dissertation, journal, conference, IEEE/ACM submission, manuscript, camera-ready, Word/LaTeX figure, or other publication. Applies venue, figure-type, color, caption/legend, formula, and paper-readability gates for architecture, workflow, roadmap, network-topology, and replicated paper figures.

Popular education & science skills

View category →

teach

mattpocock/skills

Teach the user a new skill or concept, within this workspace.

Education & Science874K installs

scaffold-exercises

mattpocock/skills

Create exercise directory structures with sections, problems, solutions, and explainers that pass linting. Use when user wants to scaffold exercises, create exercise stubs, or set up a new course section.

Education & Science488.2K installs

paper-context-resolver

lllllllama/rigorpilot-skills

Rigor Paper Context helper for README-first deep learning repo reproduction. Use only when the README and repository files leave a narrow reproduction-critical gap and the task is to resolve a specific paper detail such as dataset split, preprocessing, evaluation protocol, checkpoint mapping, or runtime assumption from primary paper sources while recording conflicts. Do not use for general paper summary, repo scanning, environment setup, command execution, title-only paper lookup, or replacing README guidance by default.

Education & Science478.3K installs

caveman-learn

juliusbrussee/caveman

Act on a Caveman learn report - review the ranked token sinks, apply cost-lowering fixes with per-edit consent, and report what those fixes returned. Use when asked to lower an agent's token cost, what caveman has saved, to trim a heavy CLAUDE.md, or to offload re-pasted context into cavemem.

Education & Science107.9K installs

gsap-plugins

greensock/gsap-skills

Official GSAP skill for GSAP plugins — registration, ScrollToPlugin, ScrollSmoother, Flip, Draggable, Inertia, Observer, SplitText, ScrambleText, SVG and physics plugins, CustomEase, EasePack, CustomWiggle, CustomBounce, GSDevTools. Use when the user asks about a GSAP plugin, scroll-to, flip animations, draggable, SVG drawing, or plugin registration.

Education & Science67.9K installs

teach-impeccable

pbakaus/impeccable

DEPRECATED: Use /impeccable teach instead. This command has been folded into the impeccable skill.