← ClaudeAtlas

swe-architectlisted

Design and draw high-level software/system architecture diagrams as code. Use whenever someone wants to design or diagram a system, service, or product ("Design a news feed", "Design TinyURL", "architect a chat app", "diagram our payment flow", "how would you build Uber's backend"), study system design, or review an architecture. Runs a short requirements interview, then generates a professional diagram (real AWS/Azure/GCP/K8s icons plus Redis, Kafka, RabbitMQ, Celery), a labelled data flow, and an animated GIF. Can also reverse-engineer a first-draft diagram from a codebase ("diagram our repo"), render multiple C4 levels, and build a narrated explainer video using the browser's built-in text-to-speech. Trigger even if the user only says "design", "architecture", "diagram", "explain this codebase", or "walk through the architecture".
singh-sumit/swe-architect · ★ 0 · AI & Automation · score 70
Install: claude install-skill singh-sumit/swe-architect
# SWE Architect Turn a system-design question into (1) a professional **architecture diagram**, (2) a **labelled data flow**, and (3) an **animated GIF** that walks the flow step by step. The golden rule: **interview before you draw.** A good architecture answer is driven by requirements and scale, not by a stock picture. Do not jump straight to a diagram. --- ## Workflow Follow these steps in order. ### 1. Interview — "grill" the requirements first Never generate a diagram from a one-line prompt. First ask a *small, sharp* set of clarifying questions so the design reflects real constraints. Ask about the things that actually change the architecture — do **not** ask trivia. Ask **4–7 questions** grouped into these buckets (pick the ones that matter for the request; skip what's already been answered): - **Core features / scope** — what must the system actually do? What's explicitly out of scope? - **Scale** — DAU/MAU, requests per second, read:write ratio, data volume, growth. - **Latency & availability** — p99 targets, uptime SLA, is stale data acceptable? - **Consistency** — strong vs eventual? Any transactional invariants? - **Data shape** — object sizes, hot keys, retention, structured vs blob vs time-series. - **Constraints** — preferred cloud/stack, budget, team size, compliance, on-prem vs cloud. Present the questions using whatever interactive mechanism the host tool offers (e.g. tappable choices) or as a short numbered list otherwise. Keep it tight — one rou