Home Assistant writes an OS update and leaves the reboot to someone, so a scheduled
one never finished unattended. The schedule now asks for the reboot once, follows it
through Home Assistant going away, and counts it done when the Supervisor reports a
different version with nothing pending. A machine that doesn't come back or falls
back to the old version fails after 20 minutes, which ends the slot and raises
update_failed. Install on the page and through MCP still leave the reboot to the person.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The tile shows Home Assistant's own used figure beside the total, so it matches
Home Assistant's storage page. An OS update doesn't reboot the machine: Supervisor
writes the image, reports the new version and raises a repair asking for a reboot.
The confirmation, schedule warning, MCP docs and plugin docs now say so.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The connection's job is keeping Home Assistant running. Each poll now reads open
repairs, Supervisor health, disk, recovery mode and add-ons that should be running
over its websocket, for the page, the overview and the problems_open, disk_free_pct
and update_failed metrics. That needs a websocket client, so websockets joins the
dependencies.
A weekly slot, set on the page or through MCP and off until then, installs add-ons,
then Core, then the OS, each after its own confirmed backup, ending at the first
failure. The slot arithmetic moves from the host agent's cadence to steward.core.weekly
so both share it.
The device scrape goes: the page's Metrics section and a target's token_from from
4c9ed02 are removed, and a migration deletes targets using it, as they'd otherwise ask
with no token. A refused token still stops a scrape.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A Prometheus target's token_from names the plugin that supplies its token at each
scrape. Read its metrics makes the target; Use this sign-in moves one at that address
across, keeping its name, filters and history. A refusal stops the connection, so one
Try again or sign-in starts everything, and the sign-in is only sent to its own address.
A 404 says how to turn on Home Assistant's Prometheus integration, as Steward doesn't
change Home Assistant. Refused also joins the volatile columns a backup check ignores.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The invite test asserted EDT for a time a day from now, so it failed from today:
a one-day link now ends after New York's clocks change on 1 November. It would
have stayed red until March.
The Home Assistant test used a fixed backup time, which put the unsent install
past its deadline at once.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Replaces the read-only token and the webhook automation, which every install
would have had to paste into its own Home Assistant. Connect now goes to Home
Assistant's own sign-in page, and Steward keeps the refresh token that comes
back, encrypted, and calls backup.create_automatic and update.install directly.
The connection page warns that the stored credential has the same access as
the user who signed in, and recommends a service account so Steward's actions
are easy to tell apart in Home Assistant's logs. A user who isn't an
administrator gets the status page and no Install.
Those actions answer only when finished, so a request cut off once sent counts
as sent and the outcome is read from the entities. Removing the connection, or
signing in again, ends the old sign-in on Home Assistant.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A new plugin reads Home Assistant with a read-only token every 5 minutes and
shows which Core, OS and add-on updates are waiting and how old the last backup
is. Install takes a backup through a local-only webhook, confirms Home
Assistant's last-backup time moved, installs, and follows the update to its end,
through Home Assistant's and Steward's own restarts. One runs at a time. Each
ends as one timeline entry.
Steward holds nothing that can act on the house: the webhook's automation can
only take a backup or install an allowed update (System 231's ruling). A refused
token stops polling, as Home Assistant bans an address after a few refusals.
Milestone 576 steps 1 to 5.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
bvandeusenandClaude Opus 5.5