DevOps docs

Container, cluster & observability kits

Six standalone kits — copy the one you need. Run & deploy first (Docker → Swarm → K8s), then observe with Zabbix, ELK, or Grafana/Loki.

Note

Examples use .NET 10 Sample prompts and Dockerfile.dotnet target .NET 10 Minimal API. Skills work for other stacks — specify the runtime in the prompt.

Run & deploy Observe Install

kits

skills

tracks · run + observe

risky cmds agent runs

Track A

Run & deploy

From a single container to a full cluster. Graduate only when Swarm’s missing primitives become a real constraint.

docker-kit available

Docker

Single-host containerization — Dockerfile, Compose, CI. No orchestrator.

/dockerfile /compose /ci

Example: /dockerfile .NET 10 Minimal API --prod

docker-swarm-kit available

Docker Swarm

Multi-node stacks that stay in Compose syntax — deploy:, not a new API.

/stack /service /secret

Use when multi-node without full K8s.

k8s-kit available

Kubernetes

Manifests, Helm charts, Prometheus rules — when Swarm primitives run out.

/deploy /helm /monitor

Example: /deploy container-api --namespace prod

Migration path run

docker-kit→ docker-swarm-kit→ k8s-kit

Start simple. Move to Swarm for multi-node Compose. Graduate to K8s for HPA, Ingress, RBAC, operators.

Which container kit?

1 host → docker-kit
Multi-node, keep compose → docker-swarm-kit
Autoscaling / Ingress / RBAC → k8s-kit

Track B

Observe — metrics, logs, alerts

Classic infra (Zabbix), full-text logs (ELK), dashboards + Loki (Grafana). Each kit ships a brain + check skills.

zabbix-kit observe

Zabbix

Hosts, templates, triggers, read-only problem investigation.

/zabbix /zabbix-check /zabbix-trigger

Brain: ZABBIX-BRAIN.md

elk-kit observe

ELK Stack

Elasticsearch, Logstash, Kibana — first-class log investigation.

/log-check /es-query /logstash

Primary log skill for KQL / full-text.

grafana-kit observe

Grafana + Loki

Dashboards, alerts, Explore, LogQL log checks.

/loki-query /grafana-dashboard /grafana-alert

Brain: GRAFANA-BRAIN.md

Which log skill?

Need	Kit	Skill
Full-text logs in Elasticsearch / Kibana	elk-kit	`/log-check`
Logs in Loki (Grafana Explore)	grafana-kit	`/loki-query`
Zabbix problem / log item	zabbix-kit	`/zabbix-check`
PrometheusRules YAML on Kubernetes	k8s-kit	`/monitor`

Anatomy

Structure of every kit

Same layout as the rest of the repo. Observability kits also add references/ and a *-BRAIN.md.

{kit-name}/
├── README.md
├── .claude/
│ ├── skills//{skill}/SKILL.md
│ └── rules//*-conventions.md, approval-gate.md
├── _templates/
├── references/ ← optional (obs kits)
└── *-BRAIN.md ← optional (obs kits)

Design

Shared principles

Agent never runs the risky command

docker build, kubectl apply, Zabbix imports, ES deletes, Grafana API — human reviews and runs. See each kit’s approval-gate.md.

Rules are the single source of truth

Skills reference shared *-conventions.md instead of repeating security and resource rules in every skill file.

Templates are copy-and-fill

Known-good baselines for both AI and humans. Docker samples assume .NET 10 unless you override.

Evidence-only investigations

Log and problem checks do not invent hit counts or metric values — TBD when data is missing.

Install any kit

Copy .claude/ and _templates/ into your project. No package manager.

Example below uses docker-kit with a .NET 10 prompt. Swap the folder name for other kits.

# from repo root
cp -r kits/devops/docker-kit/.claude/ your-project/.claude/
cp -r kits/devops/docker-kit/_templates/ your-project/_templates/

# in Claude Code (or compatible agent)
/dockerfile .NET 10 Minimal API --prod

# observability example
cp -r kits/devops/elk-kit/.claude/ your-project/.claude/
/log-check booking-api --env prod --since 1h