t3chnaztea/home-assistant-skills/ha-context-map
Claude Code and agent skill by t3chnaztea/home-assistant-skills
- community
- ext:[{"type":"plugin","description":"Custom plugin integration"}]
Checking that the SKILL.md file is still in the repository.
Use when an agent keeps guessing wrong entity ids, when starting recurring agent work against a Home Assistant instance, or when asked to "map my Home Assistant", "build an entity inventory", "document my smart home", or "why does the agent keep using entities that don't exist". Covers writing and maintaining the instance map file: entity inventory, add-ons, integrations, helpers, notify targets, and a running Gotchas list, plus the provenance checks that keep it honest. Not for making the initial connection (ha-connect) or writing automations (ha-automations).
Install
npx vskill@latest install t3chnaztea/home-assistant-skills/ha-context-mapSemi-extensibleMentions customization, but needs skill-specific knowledge. Read the extensible skills standard
SKILL.md
HA Context Map
The part people skip. An agent working against a Home Assistant instance without a map will guess entity ids, and Home Assistant fails silently around bad ids often enough that guessed automations look fine and never fire. One markdown file fixes this. Assumes the connection lanes from ha-connect.
Why a map file beats querying every time
/api/states on a mature instance returns hundreds of entities with machine-generated names (light.living_room_2, because the first living_room was deleted and re-added). The map is where a human confirms, once, that light.living_room_2 is the actual living room lamp, and where the agent learns it without burning a discovery round-trip every session. State is queryable; meaning is not.
Keep it wherever your agent reads context from: the repo where you do HA work, or directly on the box as /config/CLAUDE.md (or AGENTS.md) so any agent that lands there finds it.
What goes in it
A section per category, every section dated. Suggested skeleton:
Building the inventory without clicking through the UI
The queries produce the candidates; the human pass that says which media_player is the actual family TV is what makes the file worth having. Do that pass once, in the file, and mark it verified.
The Gotchas list
The highest-value section, and the one nobody writes. Every instance accumulates local traps that no doc can predict. Real examples of the shape of entry that belongs there:
- "
light.kitchenis a group of only two of the four kitchen lights." - "The lamp on the switched outlet reads
unavailablewhen the switch is
off. Normal. Do not flag it."
- "The old phone's
notify.mobile_app_*service no longer exists;
automations calling it abort silently at that step."
- "Renamed entities keep their old id unless you rename the entity id, not
just the friendly name."
- "
unavailableon anautomation.*entity means an orphaned registry
entry, not a broken automation."
Append with a date, never rewrite history. When an agent hits something that cost more than five minutes, the fix isn't finished until the gotcha is in the map.
Keeping it honest: provenance checks
A map that silently rots is worse than no map: the agent trusts it. Two rules:
- Date every snapshot section. "Add-ons (as of 2026-07-01)" tells a
future agent exactly how much to trust it.
- Ship re-verify one-liners with the claims. For each volatile section,
include the command that checks it, so any session can cheaply re-run:
If a check disagrees with the map, trust the live output, fix the map, and re-date the section. That loop is the whole maintenance burden: a few minutes, whenever a check fails.
What NOT to put in it
- The token, MQTT passwords, or any credential. The map gets pasted into
agent context freely; it must be safe to leak.
- Full automation YAML (it lives in
automations.yaml; the map holds the
index and the gotchas).
- Anything the agent can derive fresh more cheaply than you can maintain it
(full state dumps, attribute lists).
Security
- Security scanPassed, 95 of 100Pattern checks on the skill's files.
- Model reviewPassed
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingskills/ha-context-map/SKILL.md
Popularity
- Repository stars
- 1
- 7-day trend score
- 3
Stars and forks are for the whole GitHub repository, not this skill alone. Not refreshed yet.
Extension points
pluginCustom plugin integration
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/t3chnaztea/home-assistant-skills/ha-context-map)