4+1 Architecture Models

Generates Kruchten's 4+1 View Model for any software system. Each deliverable per view is two artefacts:

1. A rendered architectural drawing — Mermaid (primary) or PlantUML (for deployment-heavy physical views).
2. A Miro RISEN prompt — ready to paste into Miro AI Sidekick to recreate the same view on a Miro board.

Plus a prose view document that wraps both and explains the architecture decisions, quality attributes, and concerns.

This skill is portable by design: the same SKILL.md works natively in Claude Code, via a Cursor rule (shims/cursor/), and via a Copilot custom prompt (shims/copilot/). Change logic here; the shims will pick it up.

The five views at a glance

Reference files per view: references/logical-view.md, references/process-view.md, references/development-view.md, references/physical-view.md, references/scenarios-view.md. Read the relevant reference before generating that view — do not attempt any view from memory.

Notation cheatsheets: references/notation-mermaid.md, references/notation-plantuml.md, references/notation-bpmn-in-mermaid.md. Read these the first time you render in a given tool.

Workflow

Step 1 — Determine invocation mode

Decide which mode the user is in. If ambiguous, ask.

• Zero-input mode — user gives a short description (a paragraph or less) and expects you to draft a full 4+1 with flagged assumptions. Example: "Do a 4+1 for a microservices e-commerce platform."
• Interview mode — user has real context and wants rigour. Ask follow-ups for anything missing from the checklist in Step 3 before drafting.
• Partial mode — user wants only specific views. Example: "Just the physical view with PlantUML." Skip view-generation steps that weren't requested; still do the context gathering for the requested view(s).

Step 2 — Determine audience (ALWAYS ASK unless user pre-stated it)

Audience drives notation. Ask this exact question if the user has not already told you:

"Who's the primary audience for this documentation? (a) Dev-only — developers, architects, SRE. UML-flavoured notation throughout. (b) Cross-functional — includes business analysts, product, operations. BPMN-style swimlanes for the process view; UML-flavoured elsewhere. (c) Executive — senior stakeholders, steering committees. Simplified C4 at context level only; skip low-level detail."

Record the answer. It determines which notation each view uses. Each reference file has an Audience routing section that tells you exactly what to render for (a), (b), or (c).

Step 3 — Gather system context

Before generating any view, confirm or extract these. If critical information is missing, ask — do not invent silently. If in zero-input mode, generate reasonable assumptions and mark each one with > **Assumption:** blockquotes so the architect can find and verify them.

Checklist:

• System name and one-line purpose
• Key stakeholders and their primary concerns (2–5 stakeholders, 1 concern each)
• Technology stack — languages, frameworks, datastores, infra provider(s)
• Scale characteristics — users, throughput, geographic distribution, data volume
• Key quality attributes — pick from: availability, latency, throughput, security, privacy, scalability, maintainability, portability, auditability, cost
• Constraints — regulatory (GDPR, HIPAA, PCI-DSS, AI Act, etc.), organisational, legacy-integration
• Out of scope — what the system deliberately does not do

Step 4 — Route concerns into each view

Read concerns/README.md for the catalogue of concern modules. For each view, pull in the concerns that apply (by default: gdpr-data-protection.md, security.md, bias-fairness.md if ML is present, and regulatory-compliance.md if the system operates in a regulated sector).

Each concern module has a per-view prompts section telling you what to check for that view. Surface any genuine concerns as > **Concern (GDPR):** … blockquotes inside the view document. Do not boilerplate — only flag what actually applies to this system.

Step 5 — Generate each view

Generate in this exact order. Scenarios go last because they validate the others.

1. Logical view — read references/logical-view.md, generate the Mermaid diagram, wrap in prose, flag concerns.
2. Process view — read references/process-view.md, generate the Mermaid diagram (sequence or swimlane-flowchart per audience), wrap in prose, flag concerns.
3. Development view — read references/development-view.md, generate the Mermaid diagram, wrap in prose, flag concerns.
4. Physical view — read references/physical-view.md, generate the PlantUML diagram (with cloud-provider stdlib where relevant), wrap in prose, flag concerns.
5. Scenarios (+1) — read references/scenarios-view.md, produce the scenario list with a diagram for each key scenario, verify that each scenario exercises elements from all four main views.

Each view document uses the structure in templates/view-template.md. Do not abbreviate — architectural documentation is a formal deliverable that must survive being read six months later by someone who wasn't in the conversation.

Step 6 — Generate Miro prompts

After each view document, produce the corresponding Miro prompt using templates/miro-prompt-template.md. Miro prompt rules (non-negotiable, derived from the RISEN framework):

• Role — one line, stating Miro's role (e.g. "You are a senior solutions architect creating a physical view diagram on a Miro board…").
• Input — a tight summary of the view (what's being depicted, scope boundaries, key components). Reference board context explicitly if the user has uploaded sticky notes or docs.
• Steps — numbered, imperative, specific. Miro AI produces dramatically better results when you say "1. Create a frame titled X. 2. Place 5 sticky notes labelled A–E in the top-left corner. 3. Draw solid arrows from each sticky to the central component." than when you say "make it look like a process flow."
• Expectation — specify exact counts, exact colour semantics, exact connection types. E.g. "Output: 1 frame, 12 sticky notes (6 yellow for external systems, 6 blue for internal services), 14 solid arrows for sync calls, 4 dashed arrows for async events."
• Narrowing — state what to exclude. E.g. "Do NOT include authentication flows. Do NOT draw data flows — this is a deployment view only."

See examples/synth-claim/miro-prompts/ for fully-worked examples.

Step 7 — Cross-view consistency check

After all views are generated, verify:

• Components named in the logical view appear in the development view (with the same names).
• Processes in the process view map to deployable units in the physical view.
• Scenarios exercise elements from all four main views (each scenario should name components from at least three views).
• Naming is consistent — no synonyms for the same component. If there are synonyms, flag them explicitly; do not silently resolve them — the architect needs to decide.

scripts/validate-views.py performs mechanical checks on a generated view directory (same names used across views, no orphan components). Run it if you're outputting to disk and want a quick sanity check.

Step 8 — Output format

Default output structure on disk:

[system-name]-architecture/
├── 00-system-context.md        # The gathered context from Step 3
├── 01-logical-view.md
├── 02-process-view.md
├── 03-development-view.md
├── 04-physical-view.md
├── 05-scenarios-view.md
├── diagrams/                   # Rendered diagram sources (edit these, regenerate prose)
│   ├── logical-view.mmd        # Mermaid source
│   ├── process-view.mmd
│   ├── development-view.mmd
│   ├── physical-view.puml      # PlantUML source
│   └── scenarios-view.mmd
└── miro-prompts/
    ├── logical-view-prompt.md
    ├── process-view-prompt.md
    ├── development-view-prompt.md
    ├── physical-view-prompt.md
    ├── scenarios-view-prompt.md
    └── full-board-prompt.md     # One megaprompt to regenerate the entire board

If the user wants a single consolidated document, combine all five views with a Table of Contents and put the diagram sources inline. Keep miro-prompts/ as a separate folder regardless.

Quality standards (hold the line on these)

• No placeholder text. Every section contains real architectural content. If you can't generate it, say so and ask.
• Diagram + prose, always. The diagram is the artefact; the prose explains the decisions. Neither alone is sufficient.
• Rationale is mandatory. Every significant architectural choice states the quality attribute or constraint that drove it. "We use an API gateway" is incomplete; "We use an API gateway to enforce uniform authN/authZ across services (security, maintainability)" is complete.
• Audience statement. Each view document opens with a one-line statement of its audience and the one thing they should take away.
• Assumptions are visible. Mark all assumptions with > **Assumption:** blockquotes so the architect can find and verify them.
• Concerns are specific. Do not boilerplate GDPR language into every view. Flag the actual concerns that apply.

Partial mode shortcuts

• "Just the physical view" → Skip Steps 5.1–5.3 and 5.5. Do Step 5.4 only. Generate the corresponding Miro prompt.
• "Regenerate the process view with BPMN swimlanes instead of sequence" → Re-read references/process-view.md, switch audience to (b), regenerate just that view.
• "Add a new concern module for FCA compliance" → Ask the user for the concern content (or draft one based on their description), save to concerns/fca-compliance.md, add it to the concern catalogue in concerns/README.md.

When this skill is not the right fit

• User wants low-level sequence diagrams for a specific API → that's an API design task, use their API-design skill if they have one.
• User wants code-level class diagrams for a single module → that's development-view territory but may not need the full 4+1; confirm scope.
• User wants a C4 model only (not 4+1) → mention that 4+1 and C4 are complementary, offer to produce both or just C4.

Reference index

Per-view detail:

• references/logical-view.md
• references/process-view.md
• references/development-view.md
• references/physical-view.md
• references/scenarios-view.md

Notation cheatsheets:

• references/notation-mermaid.md
• references/notation-plantuml.md
• references/notation-bpmn-in-mermaid.md

Concerns (modular — add, remove, or swap as the company evolves):

• concerns/README.md (catalogue)
• concerns/gdpr-data-protection.md
• concerns/security.md
• concerns/bias-fairness.md
• concerns/regulatory-compliance.md
• concerns/sustainability-climate.md
• concerns/accessibility.md

Templates:

• templates/view-template.md
• templates/miro-prompt-template.md

Worked example (study this to calibrate tone and specificity):

• examples/synth-claim/ — a synthetic hybrid-cloud claims-processing platform with all five views rendered, GDPR concerns flagged, and Miro prompts generated.

Scripts:

• scripts/validate-views.py — cross-view consistency check.

Shims (same skill, different tools):

• shims/cursor/architecture-models-4plus1.mdc — Cursor rule
• shims/copilot/architecture-models-4plus1.prompt.md — Copilot custom prompt