Architecture Documentation

Help the user create and maintain architecture documentation that is version-controlled, multi-layered, and useful. Good architecture docs answer "what exists and how does it connect?" — not by drawing everything, but by drawing the right things at the right abstraction level.

Workflow

Step 1: Assess Documentation Scope

Understand what exists and what's needed before creating anything:

• Current state: Are there existing diagrams, READMEs with architecture sections, or whiteboard photos? Scan the repo for docs/, architecture/, .puml, .d2, or Mermaid blocks in markdown.
• Audience: Who will read this? New hires need conceptual overviews. Ops teams need runtime topology. Architects need structural boundaries.
• System complexity: How many services, data stores, and external dependencies? This determines which abstraction levels are worth creating.

A simple monolith may only need one conceptual diagram. A distributed system with 5+ services likely needs canonical and runtime levels too.

Split the ask before you start. "Onboarding documentation" usually bundles two different deliverables, and only one is this skill's. A reader who cannot tell which service calls which needs a structural view — that is the work below. A reader who cannot get the system running needs setup instructions, which live in the README and belong to project-documentation. Say which half you are producing; a diagram does not fix a broken local setup, and a README does not substitute for a container view.

Step 2: Choose Abstraction Levels

Architecture documentation works at three distinct levels. Not every system needs all three — apply YAGNI.

Guidelines for selection:

• Always start with Conceptual — even a quick sketch helps align understanding
• Add Canonical when the system has multiple services, teams, or bounded contexts
• Add Runtime when deployment topology, data flows, or infrastructure matter for operations

See references/abstraction-levels.md for detailed guidance on each level, traceability rules, and the promotion model.

Step 3: Select Tooling and Establish Conventions

Choose tools that fit the team's stack. The framework is tool-agnostic — what matters is consistent usage, not specific tools.

Recommended defaults by level:

Conventions to establish:

• Shared vocabulary: Create a glossary of system names. Every diagram must use these names exactly — no aliasing or renaming per diagram.
• File naming: Use a consistent pattern like {level}-{scope}.{ext} (e.g., canonical-system-context.puml, runtime-checkout-flow.d2)
• Repository structure: Set up docs/architecture/ using the template at templates/architecture-doc.md
• Docs-as-code: All diagrams must be text-based, version-controlled, and reviewable in PRs

See references/diagram-tooling.md for tool comparison and syntax examples.

Step 4: Create Diagrams

Work top-down — start at the highest level and add detail only where needed:

1. System Context (canonical): The system as a black box, showing users and external dependencies. This is the single most valuable diagram.
2. Container diagram (canonical): Zoom into the system — services, data stores, message brokers, frontends.
3. Runtime views (as needed): Data flow diagrams, sequence diagrams for key interactions, deployment/infrastructure topology.
4. Component diagrams (rarely): Internal structure of a single service. Only create when a service is complex enough to warrant it.

Per diagram:

• Define scope — one diagram, one concern
• Identify elements and relationships
• Add brief annotations explaining non-obvious connections
• Ensure traceability: every element in a detailed diagram should trace to a parent in a higher-level diagram

Linking to decisions: When an ADR changes architecture, reference the affected diagrams in the ADR and update them. For ADR creation, use the architecture-design skill.

Step 5: Validate

Run through the validation checklist before considering documentation complete:

• All services/systems in the codebase appear in canonical diagrams
• Runtime diagrams reference only elements defined in canonical diagrams (infrastructure-only components like load balancers are exceptions)
• Naming is consistent across all diagrams and matches the glossary
• ADRs that changed architecture reference affected diagrams
• Diagrams are version-controlled and render correctly
• Conceptual diagrams are clearly labeled as non-canonical
• A new team member can orient themselves using only these docs

Step 6: Establish Maintenance Governance

Documentation that isn't maintained becomes misleading. Define when and how diagrams get updated:

Update triggers:

• New service or data store added
• Service boundary changed or service split/merged
• New external dependency integrated
• ADR accepted that affects architecture (use architecture-design skill)

Promotion model: Diagrams mature over time:

Conceptual sketch (idea / RFC)
   ↓  proves valuable, referenced repeatedly
Runtime diagram (structured, tooled)
   ↓  represents stable structural change
Canonical diagram (source of truth)

Review cadence:

• Canonical diagrams: review quarterly or after major architecture changes
• Runtime diagrams: review when deployment topology changes
• Conceptual diagrams: no formal review — treat as disposable

Principles Applied

• KISS: Start with one system context diagram. Add layers only when they serve a clear audience.
• YAGNI: A monolith doesn't need four C4 levels. Document what exists, not what might exist.
• Separation of Concerns: Each abstraction level serves a different purpose. Don't mix structural truth with runtime behavior in the same diagram.
• DRY: Canonical diagrams are the single source of truth for system structure. Other diagrams reference — never redefine — those elements.

Cross-Skill References

• architecture-design: For making and documenting architectural decisions (ADRs). Use when a documentation effort reveals undocumented or contested decisions.
• project-documentation: For README, CONTRIBUTING, API docs, and other non-architecture documentation.
• infrastructure-as-code: For aligning runtime/deployment diagrams with actual IaC definitions.
• containerization: For Kubernetes-specific topology diagrams that align with actual manifests.