P1: The registration was a silent no-op in production — it read
hass.data["lovelace"].dashboards as the DashboardsCollection and
awaited a sync async_items(), and passed "strategy"/"mode" keys the
real create schema rejects. Per the real lovelace source
(components/lovelace/__init__.py, _create_map_dashboard): the metadata
collection lives at hass.data["lovelace"]["dashboards_collection"],
async_items() is a SYNC @callback, the create payload carries metadata
only (url_path/title/icon/show_in_sidebar/allow_single_word), and the
dashboard CONFIG (the strategy reference) is saved as a SECOND step
through the content manager registered under
hass.data["lovelace"]["dashboards"][url_path]. The strategy is saved
only for a dashboard created in this run (or whose config already
carries ours); an existing dashboard skips both steps, so a user's
dashboard config is never overwritten. Collection unavailable / save
raising still warns and continues, never crashing setup.
P2: Declare "lovelace" in manifest.json dependencies (alongside
frontend/http) so hass.data["lovelace"] is populated before setup
instead of relying on boot order.
P3: Guard the check-then-create with a per-hass asyncio.Lock so two
concurrent entry setups cannot both pass the exists-check and race on
the url_path create. Per-hass (attribute on the hass object) rather
than a bare module-level lock, which binds to the first event loop
that awaits it.
The lifecycle tests' fake is rebuilt to mirror the real API shape
(sync async_items, async async_create_item, a dashboards content-
manager dict whose entries expose async async_save), with assertions
that the strategy config was saved with the custom:nestquest-party
type, that a user's existing config is not overwritten, that a failing
save still leaves setup green, and that concurrent registrations
create the dashboard exactly once (verified to fail without the
guard).
A fresh HA install with children configured should show the panel
with no YAML and no manual dashboard creation: setup now registers
ONE storage-mode Lovelace dashboard whose config is the bundle's
custom:nestquest-party strategy (frontend/src/strategy.ts), under the
stable url_path 'nestquest' with a title, icon and sidebar visibility.
Registration is idempotent — the url_path is the idempotency key, so
a reload or a second entry never duplicates the dashboard, and a
dashboard already occupying the url_path (ours or the user's) is left
completely alone (never overwritten, never deleted). An unavailable
Lovelace dashboards collection (YAML mode, lovelace not loaded) or a
refused create logs a clear warning and setup continues — never a
crash. Tests cover create-on-setup, no-duplicate-on-reload, the
untouched user dashboard, the missing collection and a refusing one.
HA event triggers expose the payload as trigger.event.data, not
trigger.payload; fix the blueprint, its tests, and the README
examples, with a regression guard.
Child ids sharing a prefix no longer suppress each other (split and
compare entries for equality), and queued mode keeps a second child's
day-complete event from being dropped while the first notify runs.
Triggers on nestquest_child_day_complete (never a zero-quest day),
notifies with the payload's child name, and dedupes per child per
day via an input_text helper holding <child_id>:<today> entries
pruned to today — a re-cleared day does not re-fire, but a second
child clearing the same day still celebrates.
A missing or non-numeric remaining sensor reads as -1 in the
all-clear guard, so a degraded day reaches the parent instead of
looking like a clear day.
Defaults to 20:00 local — deliberately before the midnight missed
sweep, reading the same day's data. Every present child gets a
completion line; owed lines name still-open quests; absent children
are omitted; nothing sends when all clear unless opted in.
The modern HA action schema passes title and message to the notify
service only under data; fix the morning summary and afternoon
reminder blueprints and the test accessors accordingly.
Lists only present children who still owe quests; sends nothing at
all when every present child is clear (a false template condition
stops the action). Blueprint test helpers move to a shared module.
Duplicate child names get HA registry suffixes (_2); the template now
iterates the due sensors (which carry present/child_id/child_name as
attributes) and pairs counts by child_id, so suffixed duplicates are
listed and the household rollup is excluded by the presence guard.
Iterates present children from the Feature 10 presence sensors, lists
what each owes today from the remaining sensor, omits absent children
entirely. Blueprint inputs (time, notify service, send toggle); the
notify target is the household's choice, never hard-coded. PyYAML and
Jinja2 join the dev-only dependencies for blueprint template tests.
Notify target (empty default, household fills their own service),
three strict HH:MM automation times (08:00/15:00/20:00 — the report
deliberately before the midnight sweep), and four independent enable
toggles, all leniently stored and round-tripped.
The whole watermark read-query-fire-write span runs under a
per-database sweep lock; a racing second sweep re-reads the watermark
inside the lock and announces nothing.
Fires one nestquest_quest_missed event per past-due open instance at
the configured rollover time and at startup, watermark-guarded in the
new nestquest_meta_state table (migration 7, strictly additive).
Read-only over the domain tables; never reports an instance that was
ever completed.
The completed event's was_on_time travels on the CompletionResult
(decided with the write) instead of re-reading latest state a racing
reversal could replace; unload clears coordinator listeners before
the database close.
NestQuestCoordinator.async_refresh holds a per-entry lock so
overlapping service-path refreshes never interleave; the queued pass
always publishes after the in-flight one.
complete_quest and uncomplete_quest force a coordinator refresh when
the call appended, so entity states land the moment the service call
returns; scheduled internals are unaffected.
complete/uncomplete return a CompletionResult str subclass carrying
appended (decided with the write under the lock), so duplicate
concurrent calls fire one transition event; day-complete only
evaluates for an instance due today.
quest_completed and quest_uncompleted fire on actual transitions with
documented payloads; child_day_complete fires when a completion clears
the child's day; quest_missed is defined as Feature 11's contract.
Payloads documented in the README for automation authors.
Household quests due/completed today sum the active children's
snapshots; cycle day reports the first scheduled child's cycle
position (0 when nobody is scheduled), anchor-date arithmetic only.
all_done is on exactly when the child owed quests today and owes none
remaining (a zero-quest day stays off so celebrations never fire on
an empty day); present_today mirrors the presence engine's answer
for the HA-local date.
State carries the earliest open quest's title truncated under the
255-character limit with the full shape in attributes; the documented
"none" sentinel is reported when the day is clear.
A False platform unload now returns False and retains the runtime
record so live entities keep their database. Sensor names and device
labels re-derive from the current snapshot so child renames
propagate; unique_ids stay child-id based.
Four sensors per active child read the coordinator snapshot: quests
due today (carrying the panel payload that omits missed instances plus
the admin payload that includes them, D-009), completed, remaining, and
completion percentage (100 on a zero-quest day). Platform forwarding
wired through the harness's entity registry stand-ins.
The setup except path closes the database; the coordinator created
above it now shuts down first so no listener can fire against the
closed connection. Regression test included.
One refresh pass builds the whole entity snapshot: active children,
today's instances with derived open/done/missed states, per-child
presence via the PresenceEngine (records converted to model objects),
rollup counts, and the custody cycle day. Interval configurable via
the new update_interval option, default five minutes.
permission_gate wraps all registered services at registration time:
admin-only policy reads call.context.user_id against admin_allowlist
is_admin (fail closed on no context); complete_quest stays open.
overseer
dev_agent
grok
orchestrator-agent