lh-infrastructurelisted
Install: claude install-skill jrodrigopuca/local-agents
## Critical Patterns
### 1. Infrastructure as code — the repo IS the source of truth
Every piece of infra is declared in version-controlled code (Terraform,
Pulumi, CloudFormation, whatever), reviewed like application code, and applied
through the pipeline. Two consequences that matter: prod can be rebuilt from
zero without anyone's memory, and every change has a diff, an author, and a
rollback. The click-ops server configured by hand in a console is the thing
that takes the whole team down and nobody can say why — because there's no
record of what it was.
### 2. No snowflakes — environments are cattle, not pets
Servers/containers are identical, disposable, and replaceable, not lovingly
hand-maintained individuals. If fixing a sick instance means logging in to
nurse it, the design is wrong — kill it and let automation bring up a fresh
one. Immutable infrastructure (rebuild-and-replace over modify-in-place) makes
"what state is this box in?" a question with a known answer: the one the code
declares.
### 3. Dev/prod parity — shrink the "works on my machine" gap
The closer dev, staging, and prod are, the fewer surprises ship. Same OS, same
runtime versions, same dependencies (containers make this cheap); differences
live ONLY in injected config, never in code paths or versions. The gaps that
remain (prod has real scale, real data volume, real network latency) are named
explicitly so nobody mistakes "passed staging" for "will work at 10x load".
### 4. Config in the environ