Manage Home Assistant configuration safely and fast — edit and deploy YAML (automations, blueprints, scripts, scenes, templates, MQTT), validate with ha core check, deploy via git or rapid scp, reload-vs-restart correctly, verify changes from logs, traces and entity state, and build Lovelace dashboards.
SkillScore73Maintained
Automatic
◆DevOps & SRE pack
Last verified 8/14/2026
What it changes
Passed 14 of 19 test cases with the skill, and 10 of 19 without it.
Without skill55%
With skill73%
Checks it passed that the model alone missed
Assigns a top-level triggers key and a stable id key.
Instructs visual inspection in the web browser after deploying.
Uses scp to directly transfer the updated dashboard file to /config.
and 3 more in the full report
5 fixed11 passed either way5 still failing1 broke
Pass-rate lift
+18 pp
Tokens
+61%
Agent turns
0%
Response time
-44%
Model gemini-3.6-flashMeasured 8/14/2026v116 of 22 cases passed with skill · 19 comparable
Tokens, turns and time describe resource use, not a dollar price.
See it in action
One of 5 test cases the skill fixed. Same model, same prompt, with and without the skill.
1 of 5
Prompt
What specific SSH command filters recent core logs for errors or entries related to 'climate_control'?
What was checked, without and with the skill
Checked
Without skill
With skill
Uses grep -iE with error and climate_control piped to tail -20 on ha core logs.
Missed
Passed
case-20 · gemini-3.6-flash · 8/14/2026 · answers are capped and redacted
What's inside
The instructions your model reads when the skill loads.
SKILL.md122 linessingle file
Home Assistant Manager
Operate a remote Home Assistant instance precisely: make a change, get it live, prove it
worked. Optimize for the fewest safe round-trips.
Assumptions
The repo you're editing is the HA /config dir, git-connected to the instance.
Edits aren't live until pulled on the instance.
[email protected] in examples is a placeholder. Resolve the real user/host once
(project CLAUDE.md, ~/.ssh/config, or ask) and if it isn't recorded in the project
CLAUDE.md yet, add it so future sessions skip this step.
Access via one or more of: hass-cli (REST), SSH ha, or an MCP server (see below).
Only edit .yaml/.yml/.md. Never read/write .env or secrets.yaml; use !secret.
Remote access — pick the right tool
SSH ha — always works, needs no local env. Use for ha core check|restart|logs|info.
hass-cli (REST) — state/service calls, but needs HASS_SERVER/HASS_TOKEN in the
shell before the session starts. If they're unset, hass-cli falls back to the wrong
host (localhost) and errors — don't retry, check [ -n "$HASS_TOKEN" ] once, then use
SSH or MCP instead.
MCP (preferred when available) — first-class tools for live state/control, no env
juggling. Official mcp_server integration (HA core ≥2025.2) or community ha-mcp
(richer, 80+ tools). Use it instead of shelling out when present.
The deploy pipeline (the one canonical flow)
Changes are not live until step 4.
Edit YAML locally.
Validate: ssh [email protected] "ha core check" (slow, ~30-60s — see "when to
skip" below).
Apply: reload if possible, else restart (table below).
Verify (next section).
Rapid iteration: skip git and scp straight to the instance, then reload — good for
dashboards and tight test loops. Commit to git only once stable.
scp automations.yaml [email protected]:/config/ → reload.
When to skip ha core check: it parses the whole config and is slow. For an isolated
YAML edit you're confident in, a domain reload surfaces errors faster and the logs tell