Home Assistant Manager Skill
A Claude Code skill that turns Claude into a home automation expert: safely deploy configs, verify automations end to end, and rapidly build tablet-optimized Lovelace dashboards.
The README asserts validate-before-apply and reload-vs-restart discipline, but the core SKILL.md is not provided, so these safety practices cannot be confirmed. The workflow requires root SSH ([email protected] in examples), scp deployment, and direct writes to .storage/ — a broad permission surface with no least-privilege or confirmation protocol documented; HASS_TOKEN is passed via environment variables with no leakage guidance. Deductions are tied to evidence: root SSH, direct .storage writes, no documented confirmation or rollback procedures. Author attribution and MIT license justify source_attribution of 2 (unverified publisher earns no extra points per the publisher rule).
Prerequisites (hass-cli, SSH keys, HASS_SERVER/HASS_TOKEN, optional Context7 MCP) are clearly documented, so dependency_availability earns 2. But the claimed verification protocol, log-analysis patterns, and error detection cannot be checked because SKILL.md is absent; self_consistency and failure_messages score 1 as unsupported assertions.
Audience and scenarios are well covered (automation development, tablet dashboards, template debugging, ops troubleshooting) — 3. The MCP-vs-skill section explicitly bounds capability (fallback to SSH/hass-cli without MCP) — 2. Trigger precision depends entirely on the missing SKILL.md frontmatter, so 1. Environment fit is described but assumes strong preconditions (root SSH, git-ified /config) — 2.
Install notes are the repo's strongest point: three install paths, marketplace slash commands, verification steps — 3. Complete MIT license text — 3. Repository structure, consistent naming, named maintainer, issue/discussion channels, and contribution guidelines — 2 each. Explicit gaps: no known-limitations section (1) and no version numbers or CHANGELOG anywhere (0).
Three reproducible usage examples and a clear workflow narrative support output_usability and marginal_value (HA-specific procedural knowledge, not generic git advice) at 2 each; these are claims rather than verifiable artifacts, so full marks are not warranted.
All effectiveness claims (demo videos, production dashboard examples, verification protocol) point to the absent SKILL.md and external videos, so claim_traceability is 1; no second source exists for cross-corroboration (0); facts and inferences are mostly separated but with weak demonstration (1).
- The core SKILL.md was not within this review's provided files; all safety and workflow claims are unverifiable statically — inspect the prompt content yourself before installing.
- Examples use root SSH into Home Assistant and direct writes to .storage/; run it first in an isolated environment and keep an independent backup of your config directory.
- HASS_TOKEN is a long-lived access token with no rotation, scoping, or leak-prevention guidance in the README — protect this credential carefully.
- The repository has no versioning or changelog, so marketplace updates cannot be audited incrementally; diff before upgrading.
- Publisher identity is unverified and maintenance is by an individual; do not rely on it in production without your own testing.
What does this agent do, and when should you use it?
This repository (komal-SkyNET/claude-skill-homeassistant) is a Claude Code skill following the official Claude skills specification, and it doubles as a self-hosted plugin marketplace. It transforms Claude into a Home Assistant configuration expert across three domains: remote access via SSH and hass-cli, deployment workflows (git for stable changes, scp for rapid iteration, smart reload-vs-restart decisions), and Lovelace dashboard development. The skill validates configuration before applying changes, writes modern automation YAML syntax by default (triggers:/conditions:/actions:), and closes the loop by manually triggering automations and checking logs and traces. It installs into the user's Home Assistant config repository via a .claude/skills directory or plugin marketplace slash commands, with optional Context7 MCP integration for official documentation. It does not provide live device-control tools — that is an MCP server's job — instead encoding procedure and judgment for working on your config.
The skill directs Claude through end-to-end Home Assistant config management: reading your /config repository, generating automation YAML in modern 2024.10+ syntax (triggers:/conditions:/actions: with action: instead of service:), deploying changes via scp for instant testing, and deciding whether to reload or restart based on change type. It validates configuration before applying to prevent downtime, then verifies by manually triggering automations and checking logs and traces via hass-cli — including the automation.trigger-skips-conditions gotcha. For dashboards, it creates dashboard files in .storage/, registers them in lovelace_dashboards, and builds touch-friendly layouts for 7"/11"/13" tablets using Mushroom, Tile, and Template cards with Jinja2 patterns (door/window counting with color coding, time/state-conditional display); dashboard changes go live with just a browser refresh. Stable work is committed to git. Remote operations run SSH-based ha CLI commands, and hass-cli connects using HASS_SERVER and HASS_TOKEN environment variables.
- A home automation hobbyist using Claude Code who wants to rapidly develop and verify automations (e.g., notify when the front door stays open over 5 minutes) without restarting the whole HA instance
- A user with a wall-mounted tablet (7"/11"/13") who needs touch-friendly, color-coded Lovelace dashboards
- A user debugging template type errors (e.g., a TypeError comparing str and int) who needs a systematic log-analysis, fix, and re-verify loop
- A developer versioning HA config in git who wants a hybrid scp-for-iteration plus git-for-stable workflow
- A user worried about LLMs emitting deprecated automation syntax who wants modern 2024.10+ YAML enforced by default
- An existing HA MCP server user who wants to add procedural judgment for config work (validate-before-restart, verify-from-logs loops)
What are this agent's strengths and limitations?
- Enforces modern automation syntax (triggers:/conditions:/actions:), avoiding the deprecated service: syntax LLMs tend to emit
- Includes a complete verification protocol: manual triggering plus log and trace checks, covering the automation.trigger-skips-conditions gotcha
- Smart reload-vs-restart decisions with validate-before-apply discipline reduce the risk of downtime from bad configs
- Composes with MCP servers: the skill prefers MCP tools for live state and service calls, falling back to SSH/hass-cli without them
- Dashboard changes deploy via scp and go live with just a browser refresh — no restart — enabling fast iteration
- Ships working examples from real production tablet dashboards and ready-to-use Jinja2 templates
- Depends on the Claude Code runtime; it is not usable with other AI coding tools or chat interfaces
- Requires SSH access to the HA instance (typically as root) and a git repository on /config, which may not be feasible on managed or restricted installs
- Requires local hass-cli plus HASS_SERVER/HASS_TOKEN long-lived access token setup — a nontrivial environment checklist
- Provides no live device-control capability; an MCP server is needed separately to read entity states or call services
- Context7 documentation integration requires an additional API key
- No releases, test suite, or adoption evidence independent of the author's environment are provided in the README
How do you install or deploy this agent?
Prerequisites: Claude Code installed and configured; a Home Assistant instance with SSH access enabled and a git repository connected to /config; hass-cli installed locally (pipx install homeassistant-cli); SSH key authentication configured; environment variables HASS_SERVER and HASS_TOKEN set.
Recommended (plugin marketplace, two slash commands from anywhere):
/plugin marketplace add komal-SkyNET/claude-skill-homeassistant
/plugin install home-assistant-manager@claude-skill-homeassistantOption 2 (clone into your HA config repo):
cd /path/to/your/homeassistant/config
mkdir -p .claude/skills && cd .claude/skills
git clone [email protected]:komal-SkyNET/claude-skill-homeassistant.git home-assistant-manager-repo
ln -s home-assistant-manager-repo/skills/home-assistant-manager home-assistant-managerOption 3 (download and extract):
cd /path/to/your/homeassistant/config
mkdir -p .claude/skills/home-assistant-manager && cd .claude/skills/home-assistant-manager
curl -L https://github.com/komal-SkyNET/claude-skill-homeassistant/archive/main.tar.gz | tar xz --strip-components=2 claude-skill-homeassistant-main/skills/home-assistant-managerVerify: run /plugin to confirm home-assistant-manager is listed and enabled, or start Claude Code in your HA repo and the skill loads automatically.
How do you use this agent?
Start Claude Code inside your Home Assistant config repository; the skill loads automatically. Issue tasks in natural language, e.g., "Create an automation that sends a notification when the front door is left open for more than 5 minutes" — Claude generates the YAML, deploys via scp, reloads automations, manually triggers to test, checks logs, verifies the notification, and commits to git when working. Or "Create a dashboard for my 11-inch tablet with lights, thermostat, and door status" — Claude creates the dashboard file in .storage/, registers it in lovelace_dashboards, applies a 3-column grid with Mushroom cards, and previews instantly via scp. Optional enhancements: install hass-cli and export HASS_SERVER/HASS_TOKEN, verify SSH with ssh [email protected] "ha core info", and add Context7 MCP via claude mcp add --transport http context7 https://mcp.context7.com/mcp for official documentation.
How does this agent compare with similar options?
Versus Home Assistant MCP servers (the official mcp_server integration or community ha-mcp): MCP servers give Claude live tools — read entity state, call services, control devices — but don't teach Claude how to work on your config. This skill encodes the procedure and judgment: the git/scp deploy pipeline, reload-vs-restart decisions, validate-before-restart discipline, the verify-from-logs-and-traces loop, template type-safety rules, dashboard storage-cache gotchas, and modern automation syntax. Without the skill, an MCP-equipped Claude can flip your lights but may still restart HA on an unvalidated config or write deprecated YAML; without MCP, this skill falls back to SSH/hass-cli. Together is best.