git:20260426.40b0ace to git:20260511.c0b6c74

3 added, 3 removed. Audit A to A.

---

name: ha-refresh-context

description: Fetch and normalize the full Home Assistant state into durable artifacts. Use before drafting automations or when context is stale.

allowed-tools:

- Bash

- Read

- Write

- mcp__homeassistant__GetLiveContext

---

# HA Refresh Context

## Steps

1. Run `${CLAUDE_PLUGIN_ROOT}/bin/ha-agent-lab ha refresh-context`.

2. Read the output JSON — note entity count and base_url_source.

3. Read `.claude-code-hermit/raw/snapshot-ha-normalized-latest.json` to understand what's in the house.

- 4. Compare with the previous entity count (check `MEMORY.md` House Profile section).

- 5. If new entities, areas, or domains appeared, update `MEMORY.md` House Profile.

- 6. If entities disappeared or became unavailable, note it in `MEMORY.md` Known Issues.

+ 4. Compare with the previous entity count (check your auto memory for a stored house profile).

+ 5. If new entities, areas, or domains appeared, update your auto memory with the house profile changes.

+ 6. If entities disappeared or became unavailable, note it in your auto memory.

## Output

- `.claude-code-hermit/raw/snapshot-ha-context-<date>.json` + `snapshot-ha-context-latest.json` — raw HA API snapshot

- `.claude-code-hermit/raw/snapshot-ha-normalized-latest.json` — processed entity/service index (fixed name)

- `.claude-code-hermit/raw/audit-ha-context-refresh-<date>.md` + `audit-ha-context-refresh-latest.md` — audit entry

**House profile (first run or when profile changes):** if new areas, domains, or significant entity changes are observed in step 5, write a durable house profile summary to `.claude-code-hermit/compiled/context-house-profile-<YYYY-MM-DD>.md` with frontmatter `type: context`, `foundational: true`, `title: "House Profile — <date>"`, `created: <ISO8601>`, `session: <session_id or null>`, `tags: [ha-context, house-profile]`. Append `[[compiled/context-house-profile-<date>]]` to SHELL.md `### Artifacts produced this session` under `## Monitoring`.

## When to Use

- Session start (via ha-boot) when context is stale

- Before building any automation

- After the operator reports changes to their HA setup

- Periodically to detect drift