yassimba/loom/mermaid-skill
Claude Code and agent skill by yassimba/loom
- community
Checking that the SKILL.md file is still in the repository.
Generate Mermaid diagrams as code (.mmd) and export to PNG/SVG/PDF with mmdc. Use when the user wants a diagram — flowchart, sequence, class, ER, state, gantt, git graph, architecture, timeline, mindmap — proactively when explaining a system with 3+ components, a runtime flow, a schema, or a lifecycle, and when another skill (e.g. show-me) needs a diagram authored or exported to an image.
Install
npx vskill@latest install yassimba/loom/mermaid-skillSKILL.md
Mermaid Diagrams
Author .mmd text files and compile them to images with mmdc. Mermaid lays everything out automatically — no coordinates — and the text source lives in git and embeds in Markdown.
Route elsewhere when the request needs: pixel-precise placement, branded icons, or heavy styling → drawio; a hand-drawn aesthetic → excalidraw/tldraw; a freeform whiteboard → tldraw; strict conventional UML notation → plantuml.
Setup
mmdc renders through Puppeteer and needs a headless Chrome:
`Could not find Chrome` is a toolchain failure, never a diagram error. mmdc --version passes without Chrome, and a perfectly valid .mmd still fails to compile. Install the browser above or set PUPPETEER_EXECUTABLE_PATH to a system Chrome; leave the .mmd alone.
Workflow
- Author — pick the type from the shape of the content:
| The content is… | Type | |
|---|---|---|
| steps, branches, dependencies | flowchart | |
| who calls whom, in what order | sequenceDiagram | |
| tables and relations / types and inheritance | erDiagram / classDiagram | |
| statuses one thing moves through | stateDiagram-v2 | |
| deployed services and their wiring | architecture-beta | |
| schedule / branch history / proportions / idea tree | gantt / gitGraph / pie / mindmap | |
| events over time / effort-impact grid / plotted data | timeline / quadrantChart / xychart-beta |
One diagram answers one question — past ~20 nodes, split by concern into separate .mmds instead of cramming. Syntax references, consult when unsure: FLOWCHART · SEQUENCE · CLASS-ER · ARCHITECTURE (architecture-beta — newer syntax, read before first use) · OTHER-TYPES (state, gantt, gitGraph, pie, mindmap, journey, timeline, quadrant, xychart, C4). Done when every node, participant, and label names a real thing from the content being diagrammed — no placeholders.
- Compile — the compile is the syntax check; a parse error is a bug in the
.mmd, fix and re-compile until it renders:
``bash mmdc -i diagram.mmd -o diagram.png --size 2048 ``
Themes: -t default|dark|neutral|forest (base works only inside a %%{init: {'theme':'base'}}%% directive). SVG/PDF: same command with .svg/.pdf output.
- Self-check (vision) — a clean compile only proves the syntax; read the exported PNG and judge it as its audience will. Mermaid positions everything itself, so look for content defects, not overlaps:
| Check | What to look for | Fix | |
|---|---|---|---|
| Label truncation | Node / edge text clipped | Shorten the label, or wrap with <br/> | |
| Cramped density | Nodes crammed together; tangled lines | Flip direction (TD↔LR), split into subgraphs, or reduce nodes | |
| Wrong aspect | Far too wide or too tall to read | Change flowchart TD↔LR (or set direction in class/state) | |
| Edge spaghetti | Many crossings, hard to follow | Reorder declarations so connected nodes sit adjacent; group with subgraph | |
| Wrong diagram type | Type doesn't suit the content | Re-pick from the step-1 table | |
| Low contrast | Text blends into the node fill | classDef (FLOWCHART § Styling) or -t theme |
Re-compile after every fix. Max 2 self-check rounds, then show the user regardless. No vision available → skip straight to the review loop.
- Review loop — show the image, apply the minimal
.mmdedit per request, re-compile. Overwrite the same.mmd/.pngeach round — nov1,v2, … After 5 rounds, suggest the user fine-tune at mermaid.live.
- Report — give the user the
.mmdand image paths.
Gotchas
| Symptom | Fix | |
|---|---|---|
| Parse error on a label with special chars | Quote it: A["Label: value"] | |
| Parse error on a subgraph name with spaces | Quote it: subgraph "My Layer" | |
Maximum text size in diagram exceeded | Split the diagram; last resort -c config.json {"maxTextSize": 200000} | |
| Chrome crashes in CI / as root | -p puppeteer.json with {"args": ["--no-sandbox"]} | |
Architecture icons missing / logos: name not found | Register the pack at compile time — ARCHITECTURE § Icons | |
| Blank or tiny PNG | Add --size 2048 |
Security
- Security scanPassed, 90 of 100Pattern checks on the skill's files.
- Model reviewPassed
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingskills/mermaid-skill/SKILL.md
Popularity
- 7-day trend score
- 0
Not refreshed yet.
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/yassimba/loom/mermaid-skill)