Apple Container Skill
jcordon5/apple-container-skill
Expertise em containers Apple para macOS com resultados medidos (100% vs 68%)
Ver no GitHubSkill que torna o Claude especialista na CLI de containers Apple — executa contêineres Linux e máquinas Linux persistentes em Macs com Apple Silicon. Reduz alucinações corrigindo conceitos Mac-específicos (networking, container machines, limites do macOS 15) que o Claude base erra ou inventa.
Instalação
npx convez-ai jcordon5/apple-container-skill1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Executar ambientes de desenvolvimento Linux persistentes em Macs M-series
- Troubleshoot networking e conectividade de contêineres em macOS
- Build multi-arch e push de imagens para registros
Veredito

Skill bem medido e focado em preencher uma lacuna real: corrige alucinações do Claude sobre Apple Container CLI que plain Claude confunde com Docker/Lima/Colima. Vale para quem usa Apple silicon e precisa de ambientes Linux nativos sem Docker Desktop.
README
A tradução para o português deste README ainda está na fila.
apple-container-skill
A Claude skill that turns the model into a hands-on expert for Apple's
containerCLI — the open-source tool that runs Linux containers and persistent Linux "container machines" as lightweight VMs on Apple-silicon Macs.
Apple's container is new enough that a general-purpose assistant gets it confidently wrong: it invents a brew install --cask container, reaches for Docker Desktop / Lima / Colima, and misses the Mac-specific gotchas that actually bite (per-container IPs, the macOS 15 vs 26 networking gate, container machine not even existing in its training data).
This skill fixes that. It's distilled from the entire official documentation set (release 1.x) into a SKILL.md plus four focused references, and — importantly — it's measured, not vibes. 👇
📊 Does it actually help? (measured)
Benchmarked on 4 realistic tasks, each answered with the skill and without it (plain Claude), then graded against objective assertions.
| Metric | With skill | Without skill |
|---|---|---|
| Assertions passed | 22 / 22 — 100% | 15 / 22 — 68% |
Per task (assertions passed):
| Task | With skill | Without (plain Claude) |
|---|---|---|
| Run Postgres for local dev (ports + persistent volume) | 6 / 6 | 6 / 6 |
| Persistent Linux dev environment | 6 / 6 | 1 / 6 |
| Multi-arch image build + push to a registry | 5 / 5 | 5 / 5 |
| Networking troubleshooting on macOS | 5 / 5 | 3 / 5 |
| Total | 22 / 22 | 15 / 22 |
The interesting part is where the skill earns its keep. On tasks that mirror Docker (run a Postgres, multi-arch build) both scored 100% — base Claude already knows those, because container run/build copy Docker's flags. The gap opens on the genuinely Mac-specific tasks:
- Persistent Linux dev environment (1/6 → 6/6). Without the skill Claude doesn't know
container machineexists, so it pivots to Lima / Colima / Multipass and inventsbrew install --cask container. With the skill it correctly uses a container machine — boots the image's init system (sosystemdworks), persists across restarts, auto-mounts your Mac$HOMEand user. - Networking troubleshooting (3/5 → 5/5). Without the skill Claude prescribes a bridge-network fix that silently fails on macOS 15. With the skill it flags the macOS 15-vs-26 container-to-container limitation up front and gives the port-publish workaround.
🚀 Install
npx skills add https://github.com/jcordon5/apple-container-skill
Or install it manually by cloning into your skills directory:
git clone https://github.com/jcordon5/apple-container-skill.git \
~/.claude/skills/apple-container
Either way, Claude auto-discovers it. Next time you ask about running containers or a Linux environment on your Mac, the skill kicks in. (To verify, ask Claude something like "how do I run Postgres on my M3 Mac without Docker Desktop?")
Requires the actual tool to do anything: a Mac with Apple silicon and Apple
container(signed.pkgfrom the releases page). macOS 26+ is recommended —containerruns on macOS 15 but with real networking limitations the skill will warn you about.
🧠 What it covers
- The mental model that explains every difference from Docker: each container is its own micro-VM, with its own IP — not packed into one shared VM.
- container vs. container machine — a decision table so you pick the right tool (an app to run, vs. a Linux box to live in).
- Core workflows:
run,build(incl. multi-arch via Rosetta),exec,logs,stats, lifecycle. - Container machines: persistent dev environments, home-mount modes, bring-your-own systemd image, the VS Code remote workflow.
- Networking & storage: publishing ports (v4/v6), user-defined networks (macOS 26+), local DNS, named/anonymous volumes (and the anonymous-volume cleanup trap), bind mounts, SSH-agent forwarding.
- Config & troubleshooting:
config.tomlschema, Rosetta toggle, the memory-ballooning caveat, and a symptom → fix table including the macOS-15 networking failures.
📁 What's in here
SKILL.md # the skill: mental model, macOS gate, install, workflows
references/
├── command-reference.md # the full CLI surface, every flag
├── container-machines.md # persistent Linux dev environments + VS Code
├── networking-storage.md # ports, networks, DNS, volumes, bind mounts
└── config-and-troubleshooting.md # config.toml + symptom→fix table
evals/evals.json # the 4 benchmark tasks used above
🔬 How it was built
Researched from the official apple/container docs, then iterated with the
skill-creator workflow: write the skill → run each
test task with and without it via independent subagents → grade against objective assertions →
review → improve. The benchmark numbers above come straight from that loop.
📄 License & disclaimer
MIT. This is an unofficial, community skill — not affiliated with or endorsed by
Apple. container and the Containerization project are Apple's. No warranty; you are
responsible for what you run on your machine.