infra-designer: Skill for Claude Code

System design

Part of infra-designer — a plugin that plans, deploys, and audits cloud infrastructure from one reproducible spec. This skill is the standalone judgment layer; install the full plugin for the plan → deploy → audit workflow.

This skill is the judgment layer for planning infrastructure: deciding which components a workload needs and how to size them, before any component is placed in a spec. The per-component SKILL.md files in components/ cover the mechanics of one resource; this skill covers the cross-cutting decisions between them.

How to use this

1. Always start with references/blueprint-triage.md — classify the app into known patterns before composing any nodes. Most real specs are a handful of patterns combined.
2. Then read the reference(s) for the decisions the workload actually raises. Don't read all 17 — pull the ones the requirements implicate.
3. Size with references/numbers-to-know.md before adding nodes, caches, or queues. Most workloads need far less than intuition suggests.

Reference index

Start here

• blueprint-triage.md — classify the app into known patterns BEFORE designing.
• numbers-to-know.md — single-instance ceilings (2026). Do the arithmetic before scaling out.

Data

• data-modeling.md — entities → tables; Postgres-vs-DynamoDB choice.
• db-indexing.md — when/what to index; composite/covering; DynamoDB key & GSI design.
• cap-theorem.md — consistency vs availability, chosen per feature.
• dealing-with-contention.md — locking, races, optimistic vs pessimistic.
• large-blobs.md — bytes in object storage, metadata in the DB, compute never moves bytes.

Scale

• scaling-reads.md — replicas, caching, read ladders.
• scaling-writes.md — the write ladder; the queue is the main write-scaling tool.
• sharding.md — when data outgrows one Postgres; DynamoDB partition-key design.
• consistent-hashing.md — background on even key distribution.
• caching.md — when to cache, what, and invalidation rules.