agent-skillsCAUTION

From the repository's own README, as read at the audited commit. Badges and raw HTML are left out.

The instruction file, verbatim from the audited commit — this is the text the model reads, and the surface the audit's instruction layer examines. Quoted here so you can judge it without cloning anything.

---
name: lakehousebox-home-assistant
description: Move a Home Assistant's history into LakehouseBox and keep it flowing in - the recorder's last 10 days of states and its hourly long-term statistics into Iceberg tables (no account needed to start, lhbox new), then the person claims the catalog and every state change is posted once a minute by rest_command and one automation. Use when someone wants to keep, move, back up or query Home Assistant history, energy or sensor data for longer than the recorder keeps it. Builds on the lakehousebox and lakehousebox-keep-fresh skills.
---

# Home Assistant history on LakehouseBox

The person wants their Home Assistant history kept for years and answers about it. The whole recipe, every command
and its recorded output, is in [the instructions](references/instructions.md); this page is the order and the
decisions.

## Before you start

- **Find Home Assistant.** You need its configuration folder (`configuration.yaml`, `home-assistant_v2.db`) and a way
  to restart it: a container (`docker restart homeassistant`), a Core install's service, or on Home Assistant OS the
  Terminal & SSH add-on (`ha core restart`, the folder is `/config`). If the person told you where, use that. If you
  cannot reach it, say so and do what you can from a copy of the database file.
- **Check the recorder.** `recorder: db_url:` pointing at MariaDB or PostgreSQL means the history move (step 2) is not
  covered; skip to the live feed after the claim.

## The order

1. **Access** (instructions §1). A saved key (`lhbox whoami`) or a person at the keyboard (`lhbox login`): use that.
   Otherwise `lhbox new`, no account: a catalog named `lakehouse` you can write now. Tell the person it accepted the
   Terms for them, and that it is deleted after 72 hours unless they keep it.
2. **Move the history** (§2): snapshot the database with SQLite's backup (never the live file), run the SQL, import
   `ha.states` and `ha.statistics`. Leave out `device_tracker` (locations) unless the person said yes. Answer one
   real question from the data before anything else: that is what the person came for.
3. **The claim** (§3) is the one thing only the person can do: `lhbox claim`, give them the link and the code and
   tell them to pick **read and write** (the page offers read first), then `lhbox claim --resume --wait`. If it comes
   back `downgraded`, ask again with `lhbox login --new --level write --catalog lakehouse:write --no-wait`. If they
   are not around, stop here and tell them exactly that, with the deadline.
4. **The live feed** (§4–§6), after the claim: the sink, the secret, the `rest_command` and automation **exactly as
   written** (the payload survives integrations whose attributes are not JSON), the "uploads stalled" sensor, one
   restart, rows arriving, then fill the gap between the snapshot and the restart.

## Rules that save a retry

- Back up `configuration.yaml`, `secrets.yaml` and `automations.yaml` before you edit them, and check the YAML with
  Home Assistant's own check before the restart when you can (`ha core check`, or
  `hass --script check_config -c <folder>`).
- The sink's address is a password: it goes in `secrets.yaml`, never in the conversation or `configuration.yaml`.
- One restart, after all the YAML is in place: a new `rest_command` needs it, and it is the minute that posts
  every entity at once.
- Name the catalog from `lhbox catalog list` after the claim: a claim can rename it (`lakehouse_2`).
- The first question: `lhbox query "<sql>"` (runs on LakehouseBox's side, nothing saved on disk). A DuckDB secret saved
  with `lhbox duckdb --persist` before the claim holds the old key.
- Queries, energy per day and what each domain holds: instructions §7 and "Energy". A live dashboard: the person
  connects LakehouseBox in Claude and ticks the catalog ("See it on a dashboard"). Daily summaries on a schedule or a
  morning email: the `lakehousebox-hosted-agents` skill.

CAUTIONgrade B · trust 89/100 Install with care. The audit found things worth knowing before you trust its output.

What the source does

Filesystem

none-observed

Network

none-observed

Shell

none-observed

Dependencies

pinned

Secrets in source

none-found

Findings (10)

curl -fsSL https://lakehousebox.com/install.sh | sh   # one Python 3 file, no dependencies

**`weather.sql`** must be one DuckDB SELECT over an http(s) source. This example uses Open-Meteo, which needs no API key:

Answer a first question to show the person their history is there. lhbox query runs it on LakehouseBox's side, read-only, with nothing to install and no credential written to disk (an unclaimed catalo

- A public folder is world-readable. When the catalog is public and its admin published a folder of this bucket (lhbox catalog publish <name> --confirm --files public/, docs/public-warehouses.md, "Pub

Read lets you list and query tables, get the catalog's read-only connection recipe, browse and preview the catalog's files, and list its uploaders and sinks. Write adds creating, changing and dropping

The key is never shown. Do not ask for it, do not look for it, do not cat the credentials file. Every lhbox command and the connection recipe below read it themselves. If the person wants to see or re

- The lhbox command line: curl -fsSL https://lakehousebox.com/install.sh | sh.

curl -fsSL https://lakehousebox.com/install.sh | sh          # one Python 3 file, no dependencies

curl -fsSL https://lakehousebox.com/install.sh | sh       # a single Python 3 file, no dependencies

2026-10-04 09:21:00.249 DEBUG (MainThread) [homeassistant.components.rest_command] Calling post https://in.lakehousebox.com/d/<token> with headers: {'Content-Type': 'application/json'} and payload: b'

Gates applied: no_behavioural_pass.