CI/CD + infrastructure baseline

GitHub Actions, Docker, Terraform, and observability that pages you correctly.

Run it as a prompt

Paste this into any AI agent, or fetch it: curl -s https://uplift.page/api/v1/prompts/ci-cd-infra/raw

prompt.md

# CI/CD + infrastructure baseline

Every merge to main deploys automatically, infrastructure is reviewable code, and you find out about problems before users tweet them.

## Recommended stack

- **GitHub Actions** - CI/CD. Lives next to the code; matrix builds and environments cover most pipelines.
- **Docker** - packaging. One artifact that runs identically in CI, staging, and prod.
- **Terraform (or OpenTofu)** - infra-as-code. Reviewed plans instead of console clicking; state explains what exists and why.
- **Grafana + Prometheus/Loki** - observability. Metrics, logs, and alerts in one place with sane defaults.

## Build steps

1. Pipeline order: lint → test → build image → deploy staging → smoke test → promote prod.
2. Make deploys boring: immutable image tags, one-command rollback to the previous tag.
3. Terraform everything externally visible (DNS, buckets, databases); plan on PR, apply on merge.
4. Define 3–5 SLO-based alerts (error rate, latency, saturation); delete alerts nobody acts on.
5. Document the break-glass path: how to deploy when CI itself is down.

## Watch out for

- Secrets in CI logs or env dumps - use OIDC + a secrets manager.
- Alert fatigue from threshold alerts on every metric.
- Snowflake servers that drift from the Terraform state.

## Definition of done

- Merge → production without a human touching a server
- Rollback in under 2 minutes, practiced once
- On-call can diagnose a 500 from dashboards alone

The full prompt

Every merge to main deploys automatically, infrastructure is reviewable code, and you find out about problems before users tweet them.

Recommended stack

GitHub Actions - CI/CD. Lives next to the code; matrix builds and environments cover most pipelines.
Docker - packaging. One artifact that runs identically in CI, staging, and prod.
Terraform (or OpenTofu) - infra-as-code. Reviewed plans instead of console clicking; state explains what exists and why.
Grafana + Prometheus/Loki - observability. Metrics, logs, and alerts in one place with sane defaults.

Build steps

Pipeline order: lint → test → build image → deploy staging → smoke test → promote prod.
Make deploys boring: immutable image tags, one-command rollback to the previous tag.
Terraform everything externally visible (DNS, buckets, databases); plan on PR, apply on merge.
Define 3–5 SLO-based alerts (error rate, latency, saturation); delete alerts nobody acts on.
Document the break-glass path: how to deploy when CI itself is down.

Watch out for

Secrets in CI logs or env dumps - use OIDC + a secrets manager.
Alert fatigue from threshold alerts on every metric.
Snowflake servers that drift from the Terraform state.

Definition of done

Merge → production without a human touching a server
Rollback in under 2 minutes, practiced once
On-call can diagnose a 500 from dashboards alone

Served from the uplift.page library and refreshed within 5 minutes of every update.