diagram

Diagram Skill

Diagram Type Selection

Determine the subtype before writing any DSL or Mermaid. Match the user's intent to the table below, then load the corresponding reference files.

User Intent	Subtype	Load
C4, architecture, system/container/component/deployment diagram	likec4-c4	references/likec4-c4/
Sequence, flow, temporal, dynamic view, service interaction	likec4-dynamic	references/likec4-dynamic/
Data flow, DFD, data movement/transformation	data-flow	references/data-flow/

REQUIRED: After selecting the subtype from the table above, use the Read tool to load ALL markdown files from references/{subtype}/. These files are your complete reference for generating the artifact. Do not proceed without reading them.

---

View Type Decision Tree

What question is the user asking?
│
├── "What systems interact with X and who uses it?"
│   └── likec4-c4 — Context view
│
├── "What's inside system X?" / "What containers does it have?"
│   └── likec4-c4 — Container view
│
├── "What's inside container X?" / "What components does it have?"
│   └── likec4-c4 — Component view
│
├── "How is this deployed on infrastructure?"
│   └── likec4-c4 — Deployment view
│
├── "How does a request flow through the system?" / "Show the sequence"
│   └── likec4-dynamic — Dynamic view
│   └── (if no model exists yet, run likec4-c4 first)
│
└── "Where does data move and transform?" / "Show me the DFD"
    └── data-flow — Mermaid flowchart DFD

---

Shared Workflow

Step 1 — Gather Context

Ask (or infer from context) before writing any code:

1. What is the system being modeled? (name, brief purpose)
2. What is the scope? (use the decision tree above)
3. Who are the actors? (human users, external systems)
4. Are there key containers, components, or data stores to include?

Step 2 — Check for Existing Model Files

Glob **/*.likec4
Glob **/*.c4

If existing files are found, read them before writing. LikeC4 merges all .likec4 files in a project — do not duplicate element definitions. For data-flow diagrams, check docs/architecture/ for existing Mermaid files.

Step 3 — Select Subtype from Dispatch Table

Apply the dispatch table and decision tree above. Confirm the subtype before loading references.

LikeC4 Model Dependency: If the user wants a likec4-dynamic view but no .likec4 model exists yet, route to the likec4-c4 subtype first. Dynamic views reference elements defined in the model — they cannot invent new elements. Build specification { } and model { } blocks first, then return to create the dynamic view.

Step 4 — Load ALL Reference Files

Using the Read tool, load ALL markdown files from the selected subtype's references/ directory. Use Glob to enumerate the files if needed:

Glob references/{subtype}/*.md

These files are your complete syntax and convention reference. Do not generate the diagram without reading them.

Step 5 — Generate the Diagram

Follow the subtype-specific syntax, conventions, and file placement from the loaded references:

• likec4-c4: Write .likec4 files under docs/architecture/. Use specification { }, model { }, views { } blocks in that order. Follow naming conventions: camelCase identifiers, 'Title Case' display names, kebab-case.likec4 filenames.
• likec4-dynamic: Write to docs/architecture/views/flows/{scenario-name}.likec4. Reference existing model element identifiers only — do not re-declare specification or model.
• data-flow: Write a Mermaid flowchart block in a markdown file at docs/architecture/data-flow-{level}.md. Apply classDef to distinguish entities, processes, and data stores.

Step 6 — Review and Refine

After writing, present a summary:

• Diagram type and scope
• Elements included (count and kinds)
• Key relationships or flows modeled
• Which file was written

Ask whether the user wants to adjust scope, add styling, drill into a sub-system, or create a variant.

---

Anti-Patterns

Duplicating element definitions across files — Each LikeC4 identifier must be declared once. All .likec4 files in a directory are merged; duplicate definitions cause errors.

Overloading one dynamic view — Combining happy path, error path, and edge cases in a single dynamic view produces an unreadable tangle. Create one view per scenario variant.

Including too many elements in one C4 view — A context view with 30 systems is unreadable. Use exclude predicates or scope with view of X.

Connecting external entities directly to data stores in a DFD — External entities must always interact through a process. A direct entity-to-store connection means a process is missing.

Mixing temporal with topological concerns — DFDs show topology, not sequence. Do not add step numbers or attempt ordering. Use likec4-dynamic for temporal flows.

Skipping the specification block — Element kinds without a specification entry default to generic rectangles. Define kinds explicitly so diagrams share consistent vocabulary and styling.

Using vague DFD flow labels — sends data, calls service are not data names. Label flows with the data itself: order request, payment result, session token.

---

References

• references/likec4-c4/ — DSL syntax, view examples (context/container/component/deployment), style guide
• references/likec4-dynamic/ — Dynamic view DSL syntax, flow examples (request/response, fan-out, saga, error paths)
• references/data-flow/ — DFD notation mapped to Mermaid, DFD examples (context, Level 1, multi-boundary)