Visual Communication

Visual Communication

A diagram is not decoration — it is compression. If three paragraphs describe the same flow a diagram shows in 20 seconds, the diagram wins. If a diagram takes ten minutes to read, the paragraphs won.

When a diagram beats prose

| Signal | Diagram | |---|---| | The reader needs to understand a process with branches | Flowchart | | The reader needs to see who talks to whom and when | Sequence diagram | | The reader needs to understand state transitions | State diagram | | The reader needs to see what depends on what | Component / dependency diagram | | The reader needs the data model | ER diagram | | The reader needs the big picture of a system | C4 diagram | | The reader needs a 2×2 trade-off / matrix | Quadrant / matrix |

When a diagram does NOT beat prose:

• Simple linear sequences (1 → 2 → 3) — a list is clearer.
• Content-heavy ideas — diagrams force compression that loses nuance.
• When the diagram would have 30+ nodes — split or describe.
• When the reader needs exact values — a table beats a chart.

Picking the right diagram

Flow-oriented (what happens in what order)?
├── With branches/decisions → flowchart
├── Between actors/components over time → sequence diagram
└── With explicit states → state diagram

Structure-oriented (how is it organized)?
├── Data → ER diagram
├── Code → class diagram
├── System at high level → C4 diagram
└── Components + dependencies → dependency graph

Comparison-oriented?
├── 2 dimensions, 4 quadrants → quadrant chart
└── N options, M criteria → matrix / table (often not a diagram)

Full decision tree with worked examples in references/diagram-selection-matrix.md.

Mermaid is the default

Mermaid is the best default for engineering diagrams because:

• Text-based → versionable in git.
• Renders natively in GitHub, GitLab, Notion, Obsidian, VS Code.
• Low cost to edit; no image file to re-export.
• Adequate for 90% of diagrams engineering teams draw.

Use PNG/SVG diagrams only when Mermaid cannot express the idea (complex physical diagrams, highly custom layouts).

Mermaid templates

Flowchart

flowchart TD
    Start([Start]) --> Check{Is input valid?}
    Check -->|No| Error[Return 400]
    Check -->|Yes| Lookup[Lookup in cache]
    Lookup --> Hit{Cache hit?}
    Hit -->|Yes| Return[Return cached]
    Hit -->|No| Fetch[Fetch from DB]
    Fetch --> Store[Store in cache]
    Store --> Return
    Return --> End([End])
    Error --> End

Rules:

• TD (top-down) or LR (left-right). Avoid BT / RL — readers expect top-down or left-right.
• Decision nodes {} have explicit |Yes| / |No| labels.
• Start and end nodes use ().
• Node IDs short; labels readable.

Sequence diagram

sequenceDiagram
    autonumber
    participant U as User
    participant API as API Server
    participant DB as Database
    participant Cache as Redis

    U->>API: POST /login
    API->>Cache: Get session
    alt Cache hit
        Cache-->>API: session
    else Cache miss
        API->>DB: Query user
        DB-->>API: user row
        API->>Cache: Set session
    end
    API-->>U: 200 OK + token

Rules:

• autonumber adds step numbers — makes the diagram referenceable.
• Use alt / else for branches, loop for retries, par / and for parallelism.
• Arrows: ->> solid (sync), -->> dashed (async/return).

State diagram

stateDiagram-v2
    [*] --> Draft
    Draft --> Reviewing: submit
    Reviewing --> Draft: reject
    Reviewing --> Accepted: approve
    Accepted --> Superseded: new ADR
    Accepted --> [*]
    Superseded --> [*]

Rules:

• [*] = start/end.
• Transition labels name the event that triggers the transition.
• Every state has at least one outbound transition or is terminal.

ER diagram

erDiagram
    USER ||--o{ ORDER : places
    ORDER ||--|{ LINE_ITEM : contains
    ORDER }o--|| ADDRESS : "ships to"
    USER {
        uuid id PK
        string email UK
        string name
    }
    ORDER {
        uuid id PK
        uuid user_id FK
        timestamp created_at
    }

Cardinality:

• ||--o{ one-to-many (1 to 0-or-more)
• ||--|{ one-to-many (1 to 1-or-more)
• ||--|| one-to-one
• }o--o{ many-to-many

Full Mermaid cheat sheet with class, gitGraph, pie, and quadrant in references/mermaid-cheatsheet.md.

The C4 model

C4 structures software architecture diagrams at four zoom levels:

| Level | Shows | Audience | |---|---|---| | Context | Your system + users + external systems | Any stakeholder | | Container | Applications / services inside your system | Tech leads, architects | | Component | Major components inside a container | Developers | | Code | Classes / modules inside a component | Developers (rarely needed) |

Rules:

• Most teams need only Context + Container. Component diagrams for complex services. Code-level rarely.
• Each level should fit on one screen. If not, the scope is wrong.
• Legend required — what is a box, what is an arrow, what are the styles.

C4 can be rendered in Mermaid using C4Context / C4Container, or via dedicated tools like Structurizr. See references/c4-model-guide.md for the four levels with worked examples.

Diagram-as-code principles

• Checked into the repo with the code it describes. Diagrams drift faster than code; proximity helps.
• Updated in the PR that changes the system. Reviewers catch staleness.
• Consistent style across diagrams. Same shape = same concept.
• Labeled arrows — an unlabeled arrow is a puzzle.
• Direction matters — request arrows point one way, data flow another; pick conventions and stick to them.

Anti-patterns

• Decoration diagrams — added because docs look incomplete without one. If the diagram conveys nothing, delete it.
• Mystery boxes — boxes without labels or with jargon-only labels. Every box has a purpose that must be readable.
• Unlabeled arrows — arrows without direction semantics. The reader guesses.
• Everything-on-one-diagram — 40 boxes, 80 arrows, unreadable. Split into zoom levels (C4) or per-flow diagrams.
• Wrong diagram type — a sequence diagram where a flowchart was needed, or vice versa. Match diagram to the question.
• Out-of-date diagrams — worse than no diagram. Readers act on stale information.
• Image files instead of code — PNG/JPG in the repo. Nobody updates them.
• C4 without legend — readers don't know what the shapes mean.

Workflow

1. Decide whether a diagram helps. If prose is ≤3 paragraphs and linear, skip.
2. Pick the diagram type. Flow? Sequence? State? Structure? Use the decision tree.
3. Draft in Mermaid. Simple case first — no styling.
4. Review for readability. Can a reader unfamiliar with the system follow it in 60 seconds?
5. Label everything. Arrows, nodes, groups. No orphans.
6. Check the diagram into the repo next to the code it describes.
7. Update with the PR that changes the system.

References

| File | Contents | |---|---| | references/mermaid-cheatsheet.md | Full Mermaid syntax for flowchart, sequence, state, ER, class, gitGraph, pie, quadrant | | references/diagram-selection-matrix.md | When to use which diagram type; worked examples; when NOT to diagram | | references/c4-model-guide.md | The four C4 levels, when to stop, worked examples, Mermaid templates |

Related skills

• structured-writing — diagrams inside a BLUF-style doc.
• stakeholder-alignment — architecture diagrams inside RFCs.
• documentation-discipline — when to check a diagram into the repo.
• frontend-design — UI/UX mockups and visual design (different domain).
• frontend-slides — slide decks and presentations.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: viktorbezdek
• Source: viktorbezdek/skillstack
• License: MIT

Install and usage instructions live in the source repository linked above.