Drawio Skills - Setup Guide¶

Source: bahayonghang/drawio-skills via skills.sh - 9.6K combined installs across 2 indexed listings; first documented November 4, 2026 (morning sweep) GitHub: 299 stars, MIT licence, last pushed 2026-09-08, 2 SKILL.md directories under skills/ Category: Diagrams for agents: draw.io XML authoring, layout rules and export workflows for architecture and flow documents Quality Tier: 🟢 Green: both indexed skills pass Gen Agent Trust Hub, Socket and Snyk, the repository ships an MIT LICENSE at the root and the source is plain YAML plus a local CLI

This is a two-part draw.io package built around one opinion: the diagram should be generated offline from YAML, not drawn through a browser session. The README calls the default path YAML/CLI -> .drawio + sidecars, with draw.io Desktop as an optional step for PNG, PDF, JPG and embedded SVG export.

The two skills split cleanly. skills/drawio is the base skill and owns the execution primitives: the CLI, schemas, references, themes, palettes, style presets, export helpers and a diagrams.net URL fallback. skills/drawio-academic-skills is a publication overlay that owns venue policy and paper assets, and it calls the base skill's CLI rather than shipping a second runtime.

Installation¶

The standard route is the skills CLI:

npx skills add bahayonghang/drawio-skills

Restart the client afterwards so it reloads the skill list.

For a manual install, clone the repo and copy the base skill into the client's skill directory:

git clone https://github.com/bahayonghang/drawio-skills.git
cp -r drawio-skills/skills/drawio ~/.codex/skills/

If you need the publication overlay, copy skills/drawio-academic-skills next to drawio in the same directory, because the overlay resolves its sibling ../drawio for execution. The README lists the common locations as ~/Library/Application Support/Claude/skills/ and ~/.config/Claude/skills/ for Claude, and ~/.codex/skills/ for Codex.

Live browser refinement is optional and off by default. Only add @next-ai-drawio/mcp-server when you want the base skill to iterate in a browser:

{
  "mcpServers": {
    "drawio": {
      "command": "npx",
      "args": ["--yes", "@next-ai-drawio/mcp-server@latest"]
    }
  }
}

What It Provides¶

Two listings carry the whole install base, and both clear the threshold.

Skill	Installs	Use For
drawio	8,619	YAML-first creation, editing, replication and import of draw.io diagrams: architectures, network topologies, flowcharts, UML and ER, org charts, Mermaid and CSV conversion, existing .drawio bundles, 7 themes and 15 palettes
drawio-academic-skills	1,009	Publication figures for papers, theses and IEEE or ACM submissions: venue and audience preflight, figure typing into architecture, roadmap or workflow, caption and legend checks, formula fidelity and A4, Word or LaTeX expectations

No indexed listings fall below the threshold; all 2 appear above.

Why This Matters for Hermes Agents¶

Diagram output is where agent-written documentation usually falls apart: a Mermaid diagram that renders in one viewer and breaks in another, or an architecture sketch with no stable identity, so every regeneration is a fresh guess.

This package keeps a canonical bundle instead. A diagram ships as .drawio plus .spec.yaml plus .arch.json, and the CLI can validate the YAML before it renders, so an agent edits a spec rather than hand-tuning XML.

The importers are the part that changes what an agent can do with a repository it is already reading. Terraform, Kubernetes manifests, Compose files, SQL DDL, OpenAPI specs, GitHub Actions and GitLab CI can all be turned into a diagram offline, with no provider CLI, no Graphviz and no network.

For publication work the overlay adds a policy layer that a general agent will not supply on its own: which figure type fits which venue, and what a camera-ready caption has to carry.

Usage¶

You say	What happens
"Create a horizontal login flow with six nodes"	drawio generates the YAML spec, renders the .drawio file and keeps the sidecars aligned
"Diagram our network with a firewall, a core switch and two app servers"	drawio uses semantic device types and fills link labels from interface, IP, VLAN and bandwidth metadata
"Turn these Terraform files into an architecture diagram"	the config-import route normalizes the Terraform input and renders it offline
"Map the module relationships in this Python repo"	the code-import route reads the local directory and renders module and class relationships
"Import our existing diagram and give me an editable spec"	the CLI converts the .drawio file into a YAML bundle with --input-format drawio --export-spec
"Export this as a PNG for the deck"	the CLI renders through draw.io Desktop with --use-desktop
"Make an IEEE-style workflow figure for the manuscript"	drawio-academic-skills applies venue gates and delivers the bundle plus standalone SVG

Verification¶

Fetch the base skill file and confirm the frontmatter and route list are present:

curl -fsSL https://raw.githubusercontent.com/bahayonghang/drawio-skills/main/skills/drawio/SKILL.md | head -20

Then render one of the shipped example specs, which the README uses as its representative network case:

node skills/drawio/scripts/cli.js skills/drawio/references/examples/vendor-device-mapping.yaml output.drawio --validate --write-sidecars

A .drawio file plus matching .spec.yaml and .arch.json next to it confirms the CLI, the schemas and the renderer all work. If the overlay was installed, repeat the fetch against skills/drawio-academic-skills/SKILL.md and check that ../drawio exists in the same parent directory.

Security¶

skills.sh verdicts for sampled skills (verified November 4, 2026) - re-check the security pages on skills.sh before production use:

Skill	Gen Agent Trust Hub	Socket	Snyk
drawio	Pass	Pass	Pass
drawio-academic-skills	Pass	Pass	Pass

Integration with CorpusIQ¶

CorpusIQ is the connector layer that lets AI assistants read the systems a business already runs on - accounting, payments, CRM, email, files and analytics - and answer with the source named. It is read-only by design: it retrieves data and never writes to, or acts in, a connected system, which is what makes it safe to point at live accounts.

This page is part of the CorpusIQ directory, the reference we maintain for the MCP server and agent-skill ecosystem. Run alongside CorpusIQ, a third-party server like this one can be combined with a company's own business data in a single answer.

See what CorpusIQ connects

Limitations¶

Snapshot data, verified November 4, 2026: 9,628 combined installs across 2 indexed listings; 299 GitHub stars; MIT licence; last pushed 2026-09-08. Counts drift over time.
PNG, PDF and JPG export depends on draw.io Desktop being installed. Without it the CLI reports those outputs as unavailable rather than failing silently.
The academic overlay is not standalone: it requires the sibling ../drawio directory, so a partial copy breaks publication workflows.
The README directs faithful replication of existing draw.io diagrams at a separate project on its dev branch rather than at this package, so treat the replicate route here as structural redraw rather than pixel matching.
Six postprocess operations ship (mermaid, explain, relabel, restyle, heatmap, html); the README names runbook, animated SVG, tube and sequence layout, compression, buildup, PPTX, timelapse and PR diff as deferred.
Multi-page bundles and local PNG or JPEG assets are v1 features with limits: SVG assets and multi-page assets blocks are out of scope.

Pro Tips¶

Keep the three-file bundle together. Regenerating from the spec gets the same diagram back; editing the .drawio by hand and dropping the spec is how the identity drift starts.
Run the validate step before every export. It catches structure and layout problems while they are still cheap to fix in YAML.
Prefer the offline importers for anything already declared in the repo. Reading Terraform or a Compose file is more reliable than asking the agent to describe the architecture from memory.
Install the overlay only if you write papers. For product docs the base skill carries everything, and the overlay adds policy you will not use.
Skip the MCP server until you need browser refinement. The offline path is the documented default and keeps the bundle canonical.

← Skills Catalog | Skills Marketplace → Powered by CorpusIQ