Home Assistant Manager skill for Claude Code: deploy, verify automations, build dashboards

What they did

The author packaged a Claude Code skill called home-assistant-manager that turns Claude into a Home Assistant specialist rather than a generic YAML generator. The skill encodes an opinionated workflow: edit configuration locally, validate before applying, push changes to the live instance with scp for instant testing, then commit to git only once things are stable. It teaches Claude to decide between a reload and a full restart based on what changed, so routine tweaks don't cause downtime. For automations, the skill steers Claude toward modern 2024.10+ syntax (triggers/conditions/actions, action: instead of service:) rather than the deprecated style LLMs tend to emit, and it includes a verification protocol: trigger the automation manually, then inspect logs and traces for success or error indicators, including the gotcha that automation.trigger skips conditions. It also covers blueprint structure, choosing mode: restart versus single, spotting template errors and type mismatches, and an iterative fix-and-retest loop. On top of that, the skill supports Lovelace dashboard development, with an example tablet-optimized dashboard showing touch-friendly controls, color-coded status, and a responsive grid. Remote access runs through SSH and hass-cli, and the repo ships demo videos of a skill check and a full end-to-end workflow at 3x speed.

How it went

The repo presents a polished, demo-backed skill: two videos show Claude recognizing the skill and running a complete automation cycle of deployment, testing, log analysis, and git commit, plus a screenshot of a finished tablet dashboard. The emphasis on validation-before-apply and log/trace verification suggests fewer broken deploys, though results are self-reported rather than benchmarked.

Worth knowing

Requires an SSH-reachable Home Assistant instance with hass-cli installed and credentials configured; nothing here works against a purely local or cloud-only setup. The skill assumes a git repo for configs and a working scp path. Modern-syntax guidance is tied to 2024.10+ HA versions, so older installs may need adjustment. No automated test suite is described, and the demos are curated rather than exhaustive.

Wiring

Clone the repo and place skills/home-assistant-manager under your Claude Code skills directory so Claude loads it on demand. Configure SSH access to the Home Assistant host (key-based auth recommended) and install hass-cli on the machine Claude runs on, pointing it at the HA URL and a long-lived access token. Keep your HA configuration in a git repo locally. The loop is: Claude edits YAML locally, runs a config validation step, scp's files to the HA config directory, calls hass-cli to reload the relevant domain (or restart if required), then triggers automations and pulls logs/traces via hass-cli or the HA API to confirm behavior before committing to git.

Try it yourself

Install the home-assistant-manager skill from komal-SkyNET/claude-skill-homeassistant into my Claude Code setup. Then read my Home Assistant config repo and my SSH/hass-cli connection details, and: (1) audit my automations for deprecated syntax and convert them to 2024.10+ triggers/conditions/actions form; (2) flag any automation that should use mode: restart instead of single; (3) validate the config, deploy via scp, and reload rather than restart where possible; (4) for each changed automation, trigger it manually and check logs and traces for errors, remembering that automation.trigger skips conditions; (5) summarize what changed and what still needs a restart.