Fixes 4 Important findings from the final whole-branch review:
- /inventory never called WorldStore.load(), leaving the TopBar stuck on
"loading" and characterLevel() silently defaulting to 1 for any character
above level 1. Mirrors the same guard already used in HuntPageComponent.
- The combat/equipment snapshot-immutability integration test asserted only
status/round, never the actual playerState snapshot the whole test claims
to prove is untouched after a post-fight equip.
- Documented (comment only, no behavior change) that the demo character's
armor dropping from the old hardcoded 6 to 0 is an intentional,
spec-sanctioned tradeoff (Slice 0.5 spec Section19), not a bug.
- inventory.store.spec.ts's equip test used an identical inventory fixture
before and after equip(), so a regression dropping the post-equip
inventory re-fetch would still have passed. Now asserts the refetched
fixture is actually reflected.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The equippedItemInSelectedSlot computed is the riskiest logic this
component owns, but every existing test mocked selectedItem() as null,
so the slot-matching branch never ran and a regression to "any equipped
item" would have gone unnoticed. Adds a test with three items across two
slots that asserts the detail panel receives the same-slot equipped item,
not just any equipped item.
The CharacterItem idempotency check was looking up by the seed's own
literal id instead of the (characterId, itemDefinitionId) unique index
that CharacterItem actually enforces. If the demo character had already
looted a worn-short-sword naturally, re-running the seed would miss
that row and try to insert a colliding duplicate, breaking seeding
instead of being a safe no-op. Look up by the real domain key and reuse
whatever id is found when wiring up the CharacterEquipment row.
Reorders the Verbrannte Straße POIs to plan §8's authored sequence
(hunt, investigate, search, then the scout) — purely a keyboard tab-order
fix, since hotspots are placed by percentage, not list order.
Rewrites the location page's viewport-height reserve as a calc() over the
same rem values the top bar and footer already declare as their own
min-block-size, with file:line pointers to both, instead of an opaque
191px constant. Doesn't remove the underlying coupling (still no
ResizeObserver / shared token), but a future edit to either component's
minimum height now has a documented, unit-matching term to update instead
of an unexplained magic number.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Brings in the encounter-status feature (cleared/resumed hunt encounters)
and its own independent Verwilderter Straßenhund / Verkohlter Plünderer
assets. Both branches added the same two monsters at the same time;
resolved by keeping master's asset set as canonical (images/combat/icons/,
images/combat/sprites/) rather than maintaining a parallel copy under
images/monsters/icons/ — dropped that directory and pointed the seed's
MonsterDefinition.iconPath, the local-view test fixtures and the frontend
icon lookup at the existing combat/icons paths instead. Kept this
branch's COMBAT_MONSTER_SCALE entries for the two monsters, since master
never added them.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The shell sizes itself with min-block-size everywhere, which is a floor,
not a ceiling. Against that indefinite ancestor, the location page's
minmax(0, 1fr) scene row fell back to content-based sizing instead of
being bounded, so the artwork could grow tall enough to push the action
bar off screen — confirmed visually at 1920 and 1024px widths, where the
bar was fully or partially clipped.
Gives the location page its own definite, viewport-bounded height
(reserve = stable top bar + footer + own padding) instead of touching the
shared shell, which other screens still size freely. The narrow/tablet
breakpoint had a second instance of the same class of bug: the sidebar's
auto-sized row claimed its full content height before the 1fr main row
saw any space at all, collapsing the action bar to 0px height. Swapping
which row is auto vs. 1fr — main first — fixes it the same way.
Also re-anchors the four Verbrannte Straße hotspots to painted detail in
the real artwork (cart, roadside grave, road, cracked stones) rather than
the composition-reference coordinates, and lets primary-action labels
stay on one line via clamp() instead of wrapping unevenly across widths.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
A finished journey now opens the location view instead of leaving the
player on the map, and backing out of the hunt returns to the place the
hunt happens in. The victory and defeat screens gain "Zum Ort" alongside
"Weiter jagen", so the location is always reachable without costing the
hunt loop its one-click rhythm.
The store raises the arrival only after the server-owned current location
has been re-read, and does not navigate itself — timers, arrival times and
the server-side completion are untouched; only the screen that shows the
result changed.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The screen a player stands on between activities: name, region, scene
artwork with hotspots pinned by percentage, a four-button action bar and
a context sidebar covering identity, danger, encounters, interactions and
rewards.
It owns no knowledge of any particular place. Hotspots and actions are
routed by interaction type: HUNT and MAP hand off to the existing hunt
and map screens, and everything that reveals text goes through the
server-authoritative interaction endpoint. A second location therefore
renders by supplying different content, which the Südtor case in the page
spec exercises.
The shell drops its generic area rail on /location, where the screen's
own sidebar says the same thing better, and Ort joins the navigation as
its first entry. Root and unknown routes now land on the location rather
than the map: arriving somewhere should mean arriving at a place.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Brings in the First Loot slice. Resolved additively:
- MonsterDefinition keeps both the new iconPath and master's lootTableId.
- The seed keeps the four-monster pool and the local view content, and
gives the two new monsters existing loot tables — the road dog shares
the beast table, the charred looter the raider table.
- The local location view migration moves to 1788700000000 so it orders
deterministically after the loot migration, which claimed the same
timestamp.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The hunt screen kept whatever roll was last in memory, so a player coming
back from a fight saw every encounter as fresh. Encounters now carry their
own status, which the combat module advances as fights start and end.
- hunt_encounters.status replaces consumed_at, which only recorded that a
fight had begun and could not distinguish a win from a loss
- a lost fight hands the encounter back as AVAILABLE, so it can be retried;
the unique index tying one combat to one encounter goes with it
- GET /hunts/active serves the resumable hunt, which the hunt page adopts on
entry rather than trusting its in-memory roll
- defeated encounters are crossed out and lose their hover and attack action
- a fresh page load rejoins a combat the server still holds open
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Mirrors the extended current-location payload and the interaction endpoint
in the web API client. Replaces the per-spec CurrentLocationResponse
literals with one shared fixture factory, so a contract change is answered
in a single place. Adds the drawn glyph set used by hotspots, actions and
reward previews.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Adds the server side of the local location view: location_definitions
carries region naming, a location type, a scene-level description and
artwork, plus JSONB points of interest, primary actions and a reward
preview. Locations become content, so a second location renders through
the same components with different data.
GET /api/world/current-location gains those fields, a recommendation
label and a danger rating derived from the weighted average of the
location's own monster pool — a rare elite no longer makes a beginner
road read as lethal. The encounter preview is derived from that same
pool rather than duplicating it.
POST /api/world/current-location/interactions/:key reveals a hotspot's
authored result. The location is resolved from the character, never from
the request, and result text never ships with the location payload, so a
caller cannot read or trigger a hotspot it has not travelled to.
Seeds the Verbrannte Straße with its four hotspots and the Südtor with
its own transition content. Adds Verwilderter Straßenhund and Verkohlter
Plünderer to the road's pool, including combat sprites, so the preview
shows encounters the hunt can actually roll. Medallion icons move to
images/monsters/icons, where both combat and the location view read them.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Adds a real-SQL-inspection test for CreateLootAndRewards1788600000000
(mirroring the visible-vertical-slice pattern) so the one-reward-per-combat
unique index and other DB-level invariants can't be silently deleted
without failing a test, and asserts the TopBar renders character.silver
and character.experience values already present in the app.spec.ts fixture.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Fix a lock-order inversion Task 8 introduced: performAction locked the
combat row first and, inside grantVictoryRewards, the character row
second -- the opposite order to startCombat (character, then combat),
creating a deadlock cycle for two concurrent requests on the same
character. performAction now locks the character first via the
existing lockCharacter helper, matching startCombat; the later re-lock
inside grantVictoryRewards is a no-op within the same transaction.
Also strengthen the test that guards the transaction contract for
grantVictoryRewards: expect.anything() would have passed even if the
data source were handed over instead of the transaction manager, since
CombatRewardService has no runtime guard against that substitution.
The test now asserts on the captured argument's identity. Verified
this is load-bearing by temporarily passing the data source in place
of the manager and confirming the test fails.
Finally, make the rollback test's unchanged-XP/silver assertions real:
the fake grantVictoryRewards now writes through the transaction
manager before throwing, so the assertions prove the rollback
discarded those writes instead of passing vacuously.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
