Go With AI

Go-with-AI

A `CLAUDE.md` template that teaches [Claude Code](https://docs.anthropic.com/en/docs/claude-code) modern Go backend best practices for services running on AWS or NVIDIA GPU hosts. Drop it into any Go HTTP / gRPC API, worker, or model-serving gateway and Claude will follow production-grade conventions for project layout, the stdlib `net/http` server, `context` propagation, `pgx` + `sqlc` database access, AWS SDK v2, AWS Lambda (`provided.al2023` + arm64), ECS Fargate, **NVIDIA GPU workloads (Triton / NIM / NVCF / DGX Cloud)**, **arm64 vs amd64 architecture choice**, OpenTelemetry, `slog`, security, and testing — automatically, on every interaction.

Why

Claude Code reads `CLAUDE.md` from your working directory and applies its rules to every conversation (including sub-agents). Without guardrails, AI-generated Go code can drift in subtle ways that don't show up until production: a missing `ctx` argument that silently disables cancellation, a `fmt.Sprintf`-built SQL query (instant injection), `panic` in a request handler (process restart on every bad input), the wrong AWS SDK (v1 reached end-of-support July 2025), `ReadHeaderTimeout` left at zero (Slowloris bait), `math/rand` for tokens (predictable session IDs), bare goroutines with no termination path (memory leaks), `Error`-level logs for everything (alerting becomes useless). This template codifies the conventions experienced Go teams enforce in code review — small interfaces in the consumer package, the domain-package store pattern, errgroup for fan-out, slog with JSON in prod, AWS SDK v2 singletons, multi-stage distroless arm64 Docker images — so Claude follows them from the first line.

Two Versions

Pick the one that fits your project.