# Architecture Designer

> Use this skill when designing system architecture, making technology stack decisions, evaluating trade-offs between architectural patterns, or writing Architecture Decision Records (ADRs). Trigger phrases: 'design a system for', 'what architecture should I use', 'monolith vs microservices', 'how should I structure this'. Not for implementation-level code design or UI/UX design.

- **Type:** Skill
- **Install:** `agentstack add skill-xcrrr-claude-skills-architecture-designer`
- **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior
- **Seller:** [xcrrr](https://agentstack.voostack.com/s/xcrrr)
- **Installs:** 0
- **Category:** [Agent Skills](https://agentstack.voostack.com/c/agent-skills)
- **Latest version:** 0.1.0
- **License:** MIT
- **Upstream author:** [xcrrr](https://github.com/xcrrr)
- **Source:** https://github.com/xcrrr/claude-skills/tree/main/skills/coding/architecture-designer

## Install

```sh
agentstack add skill-xcrrr-claude-skills-architecture-designer
```

Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent.

## About

# Architecture Designer

## Overview
The Architecture Designer skill guides the creation of well-structured, scalable, and maintainable software systems. It covers selecting the right architectural pattern for the problem (monolith, microservices, event-driven, layered/hexagonal), conducting trade-off analysis using established frameworks, writing Architecture Decision Records (ADRs) to document decisions, and reasoning about scalability, reliability, and operational complexity. Good architecture serves the current and anticipated needs of the system without over-engineering for hypothetical future scale.

## When to Use
- Starting a new system or service and choosing the overall structure
- Evaluating whether to split a monolith into microservices
- Designing the data flow between components (sync vs. async, events vs. REST)
- Making technology choices with lasting consequences (messaging, database type, framework)
- Documenting architectural decisions for team alignment and future reference

## When NOT to Use
- Implementation-level code structure within a single service (use refactorer skill)
- Database schema design for a single service (use sql-expert skill)
- API endpoint design (use api-designer skill)
- Infrastructure and DevOps decisions (deployment, networking, cloud IAM)

## Quick Reference
| Pattern | Best For | Trade-off |
|---------|----------|-----------|
| Monolith | Small teams, early product, CRUD-heavy | Simple to develop; harder to scale independently |
| Microservices | Large teams, independent scaling, different tech stacks | Complex operations; network overhead; distributed tracing needed |
| Modular Monolith | Growing team, future microservice migration | Best of both worlds at moderate scale |
| Event-Driven | Async workflows, decoupled producers/consumers | Eventual consistency; harder to debug; ordering issues |
| Layered (N-tier) | Standard CRUD apps with clear separation | Can create rigid coupling between layers |
| Hexagonal (Ports & Adapters) | Testable, domain-first design; framework-agnostic core | More boilerplate; learning curve |
| CQRS | Read-heavy systems where reads/writes have different models | Complexity; sync challenges between command and query sides |
| Saga | Distributed transactions across services | Complex compensation logic; eventual consistency |

## Instructions

1. **Gather requirements**
   - **Functional**: What does the system do? Core use cases, user types, integrations.
   - **Non-functional**: Expected load (requests/sec, users), latency targets (p99 > Write: feed reads outnumber posts ~50:1 — optimize for reads
- Latency: feed load 100K followers) make fan-out on write expensive.

## Decision
Use hybrid fanout: fan-out on write for users with <10K followers;
fan-out on read for celebrity accounts, merged at read time.

## Alternatives Considered
| Option | Pros | Cons |
|--------|------|------|
| Fan-out on write only | Fast reads | Celebrity posts spike write load |
| Fan-out on read only | Simple writes | Slow feed loads for followed-many accounts |
| Hybrid (chosen) | Balances read speed and write cost | More complex implementation |

## Consequences
Positive: p99 feed load < 200ms; write path scales independently.
Negative: Two code paths to maintain; requires follower count metadata.
```

---

### Example 2: Design a payment processing service

**Input:** "Design a payment processing microservice that handles charges, refunds, and webhooks from Stripe."

**Output:**

**Key Drivers:**
- Correctness over performance — a missed payment or double charge is a business-critical failure
- Idempotency — retries must not cause duplicate charges
- Auditability — every state transition must be logged
- Compliance — PCI-DSS scope must be minimized

**Architecture: Event-Sourced Saga**

```
┌──────────────┐     ┌──────────────────┐     ┌────────────────┐
│ Order Service│────▶│ Payment Service  │────▶│  Stripe API    │
│              │     │                  │◀────│  (external)    │
└──────────────┘     │  - Idempotency   │     └────────────────┘
                     │  - State machine │
                     │  - Audit log     │     ┌────────────────┐
                     └────────┬─────────┘     │ Webhook Handler│
                              │ events        │  (Stripe →us)  │
                     ┌────────▼─────────┐     └────────┬───────┘
                     │  Payment Events  │              │
                     │  DB (append-only)│◀─────────────┘
                     └──────────────────┘
```

**Idempotency pattern:**
```python
# Every charge request includes an idempotency key
def charge(order_id: str, amount_cents: int, idempotency_key: str):
    existing = db.payments.find_by_idempotency_key(idempotency_key)
    if existing:
        return existing  # Return cached result, don't re-charge

    result = stripe.charge.create(
        amount=amount_cents,
        idempotency_key=idempotency_key
    )
    db.payments.create(
        order_id=order_id,
        stripe_charge_id=result.id,
        idempotency_key=idempotency_key,
        status='succeeded'
    )
    return result
```

**Payment state machine:**
```
PENDING → PROCESSING → SUCCEEDED
                     ↘ FAILED
SUCCEEDED → REFUND_PENDING → REFUNDED
```

**Webhook handling (idempotent):**
```python
@app.route('/webhooks/stripe', methods=['POST'])
def stripe_webhook():
    event = stripe.Webhook.construct_event(
        request.data,
        request.headers['Stripe-Signature'],
        WEBHOOK_SECRET  # Verify authenticity
    )
    # Idempotent: process only if not already handled
    if not db.webhook_events.exists(event.id):
        handle_event(event)
        db.webhook_events.mark_processed(event.id)
    return '', 200
```

## Best Practices
- Match the architecture to the team, not to the trend — a two-person startup doesn't need Kubernetes and 20 microservices
- Data consistency requirements drive architecture more than anything else: identify them early
- Design for the failure case — every network call can fail; every service can be slow
- Make services independently deployable from day one if you're doing microservices
- Keep the operational burden in scope — microservices are operationally expensive

## Common Mistakes
- Jumping to microservices before the domain is well understood (premature decomposition)
- Sharing a database between services (creates tight coupling at the data layer)
- Ignoring operations: who deploys, monitors, and debugs this in production?
- Building for 100x current scale on day one (over-engineering adds real cost and complexity now)
- Not writing ADRs — future team members have no context for past decisions

## Tips & Tricks
- Draw the architecture on a whiteboard first; validate with the team before writing code
- "You must be this tall to use microservices" — start with a modular monolith; extract services when team/scaling pain is concrete
- Use the C4 model (Context → Container → Component → Code) for layered architecture diagrams
- Martin Fowler's "Strangler Fig" pattern is the safest way to migrate a monolith incrementally
- Treat ADRs as living documents — update them when decisions are revisited

## Related Skills
- [api-designer](../api-designer/SKILL.md)
- [sql-expert](../sql-expert/SKILL.md)
- [security-auditor](../security-auditor/SKILL.md)
- [refactorer](../refactorer/SKILL.md)

## Source & license

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

- **Author:** [xcrrr](https://github.com/xcrrr)
- **Source:** [xcrrr/claude-skills](https://github.com/xcrrr/claude-skills)
- **License:** MIT

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

## Pricing

- **Free** — Free

## Security capabilities

Automated source analysis of v0.1.0 — what this tool can access:

- **Network access:** no
- **Filesystem access:** no
- **Shell / process execution:** no
- **Environment & secrets:** no
- **Dynamic code execution:** no

*"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.*


## Versions

- **0.1.0** — security scan: passed — Imported from the upstream source.

## Links

- Listing page: https://agentstack.voostack.com/l/skill-xcrrr-claude-skills-architecture-designer
- Seller: https://agentstack.voostack.com/s/xcrrr
- Browse the marketplace: https://agentstack.voostack.com/browse

---
Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.