Mermaid

Produce a valid, well-structured Mermaid diagram from the user's source material and intent. Pick the right diagram type, keep it readable, and emit it in a fenced mermaid code block.

Reference Files

File	Read When
`references/diagram-selection.md`	Default: choosing the right diagram type; one-line samples for each
`references/common-diagrams.md`	Generating flowchart, sequence, class, state, ER, or C4 diagrams
`references/specialized-diagrams.md`	Generating mindmap, gantt, timeline, journey, gitGraph, pie, quadrant, requirement, xychart, sankey, block, or architecture diagrams
`references/generation-guidelines.md`	Any diagram: label limits, structure rules, source-type strategies, anti-patterns
`references/styling-and-output.md`	Theming, classDef styling, frontmatter config, export, rendering targets

Workflow

Copy this checklist to track progress:

Mermaid progress:
- [ ] Step 1: Pick diagram type
- [ ] Step 2: Gather source material
- [ ] Step 3: Draft structure
- [ ] Step 4: Generate Mermaid syntax
- [ ] Step 5: Validate
- [ ] Step 6: Present output

Step 1: Pick diagram type

If the user named a type (e.g. "sequence diagram"), use it. Otherwise load references/diagram-selection.md and pick from its matrix. Common defaults:

Content	Diagram
Process with decisions, branching logic	`flowchart`
Messages between actors/systems over time	`sequenceDiagram`
States and transitions (order lifecycle, workflow)	`stateDiagram-v2`
Database / domain entities with relationships	`erDiagram`
System architecture by level	`C4Context` / `C4Container` / `C4Component`
Concept hierarchy, brainstorm, overview	`mindmap`

Ask the user to confirm only if two types fit equally well.

Step 2: Gather source material

For topics or specs, ask one clarifying question if scope is ambiguous (e.g., "overview or detailed breakdown?"), then commit. For codebases, files, or conversations, pull only what's needed to name the nodes and relationships — don't scan more than needed.

Step 3: Draft structure

Sketch the diagram's bones before writing syntax:

List the nodes/entities/steps.
List the relationships/messages/transitions.
For flowchart, sequence, state: identify start and end. For ER, class: identify multiplicity. For C4: pick the right level.
Apply the size and label rules in references/generation-guidelines.md — typically 5-20 nodes for readability.

Step 4: Generate Mermaid syntax

Open the relevant reference file for the chosen type and write the diagram inside a fenced mermaid code block. Keep syntax minimal; add styling only when requested (see references/styling-and-output.md).

Step 5: Validate

Run the common checks below, plus type-specific checks from the corresponding reference file. Revise and re-check if any fail.

Common checks:

First line declares the correct diagram type (e.g. flowchart TD, sequenceDiagram)
No unescaped ()[]{} or stray colons inside plain text labels
No blank lines inside the diagram block (breaks some renderers)
Every referenced node / participant / entity is defined before use where required
Direction and flow are consistent (top-down or left-right, not both)
Label text is short enough to render (aim for under 6 words; see per-type rules)
Total size is within the readable limit for the type (15-20 for flowchart/class; up to 40 for mindmap)

Step 6: Present output

Output the diagram in a fenced mermaid code block. Note where it renders: GitHub/GitLab markdown, Mermaid Live Editor (mermaid.live), Notion, Obsidian, VS Code with a Mermaid preview extension.

If the user requests a file, write to a .md file containing the fenced block. For image export, point them to Mermaid Live Editor or @mermaid-js/mermaid-cli (see references/styling-and-output.md).

Translation examples

How to turn typical source material into a diagram. Pattern: identify the type first, then extract nouns (nodes) and verbs (edges / transitions).

Prose → sequenceDiagram

Source: "User hits Checkout. Server validates the cart, reserves stock, and charges Stripe. On success, the order is saved and a confirmation email is queued."

sequenceDiagram
    actor User
    participant Server
    participant Stripe
    participant Email
    User->>+Server: POST /checkout
    Server->>Server: Validate cart
    Server->>Server: Reserve stock
    Server->>+Stripe: Charge
    Stripe-->>-Server: Success
    Server->>Server: Save order
    Server->>+Email: Queue confirmation
    Email-->>-Server: Queued
    Server-->>-User: 200 Order confirmed

Spec → stateDiagram

Source: "Orders start as Pending. Payment success moves to Paid; failure moves to Cancelled. Paid orders ship, then complete. Cancelled orders are terminal."

stateDiagram-v2
    [*] --> Pending
    Pending --> Paid : payment success
    Pending --> Cancelled : payment failed
    Paid --> Shipped : ship
    Shipped --> Completed : deliver
    Completed --> [*]
    Cancelled --> [*]

Scoring matrix → quadrantChart

Source: "Rate campaigns on reach (low/high) and engagement (low/high). A is high-reach, mid-engagement. B is low-reach, low-engagement. C is high-reach, high-engagement."

quadrantChart
    title Campaign reach vs engagement
    x-axis Low Reach --> High Reach
    y-axis Low Engagement --> High Engagement
    quadrant-1 Expand
    quadrant-2 Promote
    quadrant-3 Re-evaluate
    quadrant-4 Improve
    A: [0.8, 0.5]
    B: [0.2, 0.2]
    C: [0.8, 0.8]

Anti-patterns

Forcing everything into one diagram — split into multiple focused diagrams instead.
Sentences as labels — extract keywords or move detail into child nodes / notes.
Mixing two diagram types in one fenced block (not supported).
Icons or Font Awesome references when the rendering target doesn't support them.
Over-styling before the structure is validated.
Guessing syntax — load the reference file for the chosen type.

Related skills

presentation-creator for slide decks that include diagrams
define-architecture for detailed architecture decisions beyond visual diagrams
docs-writing for embedding diagrams into technical documentation