OnTokenValidated + manual access-token base64 decode was unreliable.
Switch to GetClaimsFromUserInfoEndpoint=true and extract realm_access.roles
in OnUserInformationReceived where the UserInfo JSON is already parsed into
ctx.User (JsonDocument) — no manual JWT decoding needed.
Generated with [Claude Code](https://claude.ai/code)
via [Happy](https://happy.engineering)
Co-Authored-By: Claude <noreply@anthropic.com>
Co-Authored-By: Happy <yesreply@happy.engineering>
The admin pod runs plain HTTP; the ingress terminates TLS. Without forwarded
headers, ASP.NET Core built the redirect_uri as "http://control.aeternalisgames.com/signin-oidc"
but Keycloak only accepts the https:// variant, causing "invalid_redirect_uri".
Add UseForwardedHeaders() with XForwardedProto/Host/For before UseAuthentication
so the OIDC middleware sees the correct public https scheme and hostname when
constructing the callback URL.
Generated with [Claude Code](https://claude.ai/code)
via [Happy](https://happy.engineering)
Co-Authored-By: Claude <noreply@anthropic.com>
Co-Authored-By: Happy <yesreply@happy.engineering>
Admin panel now delegates authentication entirely to Keycloak using the
Authorization Code flow. The local Identity password form and
IdentityRevalidatingAuthenticationStateProvider are removed. Access requires
the "admin" Keycloak realm role, mirroring the game's role-based auth model.
- Program.cs: AddOpenIdConnect("Keycloak") with cookie session; /auth/challenge
triggers the OIDC flow; /auth/logout signs out of both cookie and Keycloak SSO
- Login.razor: replaced HTML form with an immediate redirect to /auth/challenge
- IdentityRevalidatingAuthenticationStateProvider.cs: deleted (no longer needed)
- appsettings.json: Keycloak:Authority / ClientId / ClientSecret section added
- Directory.Packages.props: pinned Microsoft.AspNetCore.Authentication.OpenIdConnect 8.0.0
- SOR.Admin.csproj: added OpenIdConnect package reference
Generated with [Claude Code](https://claude.ai/code)
via [Happy](https://happy.engineering)
Co-Authored-By: Claude <noreply@anthropic.com>
Co-Authored-By: Happy <yesreply@happy.engineering>
Server — JWT audience validation failing:
Keycloak does not include the client ID in the `aud` claim by default
(uses "account"). ValidateAudience=true with ValidAudience="sor-game"
rejected every token. Disabled audience validation — issuer + signature
+ realm role check is sufficient for a single-realm deployment.
Bumped JWT validation failure log to Information so it appears in default logs.
Client — DISCONNECTED instead of AUTH_FAILED on rejected auth:
When the server sends an error packet immediately followed by a close frame,
the socket transitions to STATE_CLOSING before the next _process frame.
Network.gd returned early during STATE_CLOSING without draining packets,
leaving the pending request unresolved. _on_ws_closed then failed it as
DISCONNECTED. Now drains packets during STATE_CLOSING so the error packet
is resolved with the correct error code before the close is processed.
Also added "invalid_token", "role_required", "invalid_payload",
"auth_service_unavailable", "account_creation_failed" to the close reason
→ error code mapping so they surface as AUTH_FAILED / AUTH_SERVICE_UNAVAILABLE
rather than DISCONNECTED even in the fallback path.
Generated with [Claude Code](https://claude.ai/code)
via [Happy](https://happy.engineering)
Co-Authored-By: Claude <noreply@anthropic.com>
Co-Authored-By: Happy <yesreply@happy.engineering>
PC — scene reloading on auth error:
Auth.gd's keycloak error handler called change_scene_to_file(LoginScreen)
after emitting auth_failed, which reloaded the scene and wiped the error
message before the user could see it. Removed the scene change — the
login screen is already active and handles auth_failed itself.
Web — token_received signal never handled after page redirect:
login_screen._ready() returned early when Keycloak.auth_in_progress was
true, before connecting Keycloak.token_received. When the exchange
completed nobody was listening, so the game sat frozen. Signals are now
connected before the early-return check.
Web — fragile redirect URI and sessionStorage injection:
JavaScriptBridge.eval return value could stringify as "<null>". Now uses
separate origin + pathname evals with a null check. sessionStorage items
and the final navigation use uri_encode() + decodeURIComponent() instead
of manual quote-escaping, preventing injection from PKCE values.
Generated with [Claude Code](https://claude.ai/code)
via [Happy](https://happy.engineering)
Co-Authored-By: Claude <noreply@anthropic.com>
Co-Authored-By: Happy <yesreply@happy.engineering>
PC native — connection lost, retrying:
_send_keycloak_auth's DISCONNECTED handler was clearing _pending_keycloak_token
(taken at top of function) without restoring it, so the next _on_network_connected
had no token to retry with. Restore it: _pending_keycloak_token = token.
Web export — could not bind a local redirect port:
Godot's HTML5 export cannot create TCP server sockets. Add a separate web
code path in Keycloak.gd:
- start_auth_flow() detects OS.get_name() == "Web" and navigates the page
to Keycloak, persisting PKCE state (verifier, state, redirect_uri) in
sessionStorage before the page leaves.
- _check_web_callback() (called in _ready()) detects the ?code= params on
return, recovers state from sessionStorage, cleans the URL via
history.replaceState, then calls _exchange_code().
- auth_in_progress flag lets login_screen show "Completing sign-in..."
while the token exchange is in flight after page reload.
Requires the web game's origin to be registered as a valid redirect URI
in the Keycloak sor-game client.
Generated with [Claude Code](https://claude.ai/code)
via [Happy](https://happy.engineering)
Co-Authored-By: Claude <noreply@anthropic.com>
Co-Authored-By: Happy <yesreply@happy.engineering>
The server closes idle unauthenticated connections after AuthTimeoutMs
(10 s). A user authenticating in the browser takes longer than that, so
the socket the client held while the browser was open is already dead
when token_received fires. Sending auth.keycloak on that stale
connection produces no reply and the scene never changes.
Fix: login_with_keycloak() now calls Network.force_reconnect() instead
of sending immediately. _on_network_connected() then sends auth.keycloak
as the guaranteed-first message on the fresh connection.
Also: DISCONNECTED errors in the auth.keycloak error callback no longer
clear keycloak_refresh_token, preserving silent re-auth on transient drops.
Generated with [Claude Code](https://claude.ai/code)
via [Happy](https://happy.engineering)
Co-Authored-By: Claude <noreply@anthropic.com>
Co-Authored-By: Happy <yesreply@happy.engineering>
Server:
- Add TeamCount, PlayerOneTeamIndex, PlayerTwoTeamIndex to GameSession so
team assignments survive the lobby→game transition without recomputation
- GameLobbyReadyHandler.StartGameAsync: apply TeamAssignmentsJson (host custom
reassignments) for all seats; store P1/P2 team indices on the session
- GameReadyHandler N-player path: use session.PlayerOneTeamIndex /
PlayerTwoTeamIndex instead of hardcoded teamCount=1 (which put both P1 and
P2 on team 0 in every non-FFA game)
- EF migration: AddGameSessionTeamFields
Client (game_screen.gd):
- _rebuild_ownership: treat same-team allies as friendly instead of enemy
- Unit color loop: use state.teamColors palette per team instead of hardcoded
blue (self) / red (everyone else)
- _on_game_over: derive i_won from player_outcomes[me].outcome == "Winner" so
a teammate's final kill correctly shows "Victory!" for the whole team
Also: skip absent hex tiles in BoardTemplateBuilder SVG render
Generated with [Claude Code](https://claude.ai/code)
via [Happy](https://happy.engineering)
Co-Authored-By: Claude <noreply@anthropic.com>
Co-Authored-By: Happy <yesreply@happy.engineering>
- HexSides now stores radius+1 (for hex coordinate math); HexPolygonSides stores the corner count (3–10)
- Board.IsTemplateBased: template boards use tile existence for IsInBounds/GetNeighbors instead of the hexagonal formula, so N-gon and other custom shapes work correctly
- HexGrid.AllNeighbors: unfiltered 6-neighbor generator for template board use
- ApplyHexConfig: generates a regular N-gon on the hex canvas — tiles outside the polygon are marked absent; N corner zones auto-generated for each player count 2..N
- Admin builder UI: separate "Polygon Sides" (3–10 corners) and "Radius (tiles)" inputs
- BoardTemplate entity and migration: HexPolygonSides column added to board_templates
- BoardTemplateDto exposes HexPolygonSides to client
Generated with [Claude Code](https://claude.ai/code)
via [Happy](https://happy.engineering)
Co-Authored-By: Claude <noreply@anthropic.com>
Co-Authored-By: Happy <yesreply@happy.engineering>
- Admin builder: SVG hex canvas renders when GridType=Hexagon (pointy-top hexagons, correct axial layout); square CSS grid otherwise
- Admin builder: ApplyHexSides / OnGridTypeChanged handle dimension/tile pruning; canvas header shows N/diameter for hex
- BoardTemplateDto now includes GridType and HexSides so clients can filter
- rules_editor_dialog: Board Width/Height/Shape hidden when hex selected; Hex Sides hidden when square; template picker filtered to matching grid type with live toggle
Generated with [Claude Code](https://claude.ai/code)
via [Happy](https://happy.engineering)
Co-Authored-By: Claude <noreply@anthropic.com>
Co-Authored-By: Happy <yesreply@happy.engineering>
_idle_rotation now:
- Finds the nearest enemy body in _unit_bodies by XZ distance
- Returns atan2 toward that position minus the model's base angle
- Falls back to the original north/south heuristic when no enemy exists yet
Also refreshes idle rotation for all units on every state rebuild (step 4 in
_rebuild_units), skipping any unit currently in a move tween, so rotations
stay correct as units reposition across turns.
Generated with [Claude Code](https://claude.ai/code)
via [Happy](https://happy.engineering)
Co-Authored-By: Claude <noreply@anthropic.com>
Co-Authored-By: Happy <yesreply@happy.engineering>
Server - MakeHexagon starting zones:
- Zones are now corner-based instead of full-edge: each zone is the corner
tile plus (N-1)/2 tiles along each of the two edges meeting at that corner
- 2 players → opposite corners (0, 3 = right/left); 3 → every-other corner
(0,2,4); 4 → (0,1,3,4); 5 → (0-4); 6 → all 6 corners
Client - game_screen.gd:
- Add HEX_TILE_SPACING = TILE_SPACING * 0.5 constant
- _coord_to_world: use HEX_TILE_SPACING for hex world positions
- _add_tile: tile cylinder radius scaled by HEX_TILE_SPACING
- _add_highlight: highlight cylinder radius scaled by HEX_TILE_SPACING
Generated with [Claude Code](https://claude.ai/code)
via [Happy](https://happy.engineering)
Co-Authored-By: Claude <noreply@anthropic.com>
Co-Authored-By: Happy <yesreply@happy.engineering>
Server:
- CustomRulesOverride: add HexSides (int?) field
- CustomLobby entity: add HexSidesOverride (int?) column
- GameUpdateLobbyRulesHandler: save HexSidesOverride, include HexSides in
effective rules preview
- GameLobbyReadyHandler: set rules.HexSides from lobby.HexSidesOverride on
game start
- GameSavePresetHandler: capture BoardShape + HexSides in saved presets
- Migration 20260401222603_AddCustomLobbyHexSides: adds HexSidesOverride to
custom_lobbies; applied to pg-forge-01 via dotnet ef database update
Client:
- rules_editor_dialog: send null (not "") for boardTemplateIdOverride when no
template selected — empty string crashed Guid? deserialisation on the server
Generated with [Claude Code](https://claude.ai/code)
via [Happy](https://happy.engineering)
Co-Authored-By: Claude <noreply@anthropic.com>
Co-Authored-By: Happy <yesreply@happy.engineering>
Replaces burying "Hexagon" in the Board Shape list with a dedicated
Grid Type dropdown at the top of the BOARD section. When "Hexagon" is
selected, boardShape is forced to "Hexagon" on confirm; square boards
use the Board Shape field as before. Hex Sides remains visible for
the sides setting.
Generated with [Claude Code](https://claude.ai/code)
via [Happy](https://happy.engineering)
Co-Authored-By: Claude <noreply@anthropic.com>
Co-Authored-By: Happy <yesreply@happy.engineering>
When no Hex CDN model exists for an affinity, hex tiles now keep their
coloured CylinderMesh (radial_segments=6) placeholder rather than loading
a Square model or texture. Square boards are unchanged.
Generated with [Claude Code](https://claude.ai/code)
via [Happy](https://happy.engineering)
Co-Authored-By: Claude <noreply@anthropic.com>
Co-Authored-By: Happy <yesreply@happy.engineering>
- rules_editor_dialog: add "Hexagon" to Board Shape picker and a Hex Sides
spinner (3–10, default 5) so custom lobbies can configure hex boards
- game_screen _add_tile: has_model check now looks for Hex-shape CDN models
(falls back to Square), so hex tiles get height variance when a Hex model is
registered on the CDN instead of always using the flat placeholder height
- game_screen _try_load_tile_model: texture fallback now checks shape_key
("Hex") first, then falls back to "Square" — consistent with model loading
Generated with [Claude Code](https://claude.ai/code)
via [Happy](https://happy.engineering)
Co-Authored-By: Claude <noreply@anthropic.com>
Co-Authored-By: Happy <yesreply@happy.engineering>
DB columns are PascalCase (e.g. BoardWidth, not board_width). The previous
migration used snake_case names — renamed grid_type→GridType and hex_sides→HexSides
on pg-forge-01, and updated the migration file to match so future re-runs are correct.
Generated with [Claude Code](https://claude.ai/code)
via [Happy](https://happy.engineering)
Co-Authored-By: Claude <noreply@anthropic.com>
Co-Authored-By: Happy <yesreply@happy.engineering>
Movement tiles (_valid_move_tiles):
- Walk: BFS blocked by all occupied tiles (can't walk through units)
- Flight: BFS blocked only by other Flight units (flies over Walk units)
- Jump: Manhattan range, unoccupied only (was already correct)
- Teleport: Manhattan range + LOS to destination
Attack highlights (_attack_range_tiles, _valid_attack_target_coords,
_valid_attack_target_ids):
- Added AttackRequiresLos field to CardInstanceDto / ToCardInstanceDto
- Client now checks _client_has_los() for LOS-required attackers
- Tiles/units blocked by occupants are no longer highlighted as targets
Generated with [Claude Code](https://claude.ai/code)
via [Happy](https://happy.engineering)
Co-Authored-By: Claude <noreply@anthropic.com>
Co-Authored-By: Happy <yesreply@happy.engineering>
Previously, clicking an ArchSpirit showed the abilities panel AND a
separate OS-level PopupMenu for arch-spirit actions. Now everything
lives in one floating panel:
[Abilities] ← shown only if unit has active abilities
──────────
<ability buttons>
[ArchSpirit] ← shown only for friendly ArchSpirit
──────────
Sacrifice a Spirit (only when valid candidates adjacent)
Ascend (only when tier+1 card in hand)
Transition (only when tier-1 card in hand)
Removed the _show_arch_spirit_menu() PopupMenu call from _select_unit().
_show_ability_panel() now builds both sections and returns early only
when both would be empty.
Generated with [Claude Code](https://claude.ai/code)
via [Happy](https://happy.engineering)
Co-Authored-By: Claude <noreply@anthropic.com>
Co-Authored-By: Happy <yesreply@happy.engineering>
If sync_complete or sync_failed never fires (e.g. background refresh
_is_syncing guard eats the start_sync call and signals were already
emitted before init_screen connected them, or a server request is
in-flight with no reply), the player was silently stranded with no
retry path and had to relaunch.
Adds a SYNC_TIMEOUT_SEC = 20 watchdog timer started in _do_sync() and
reset on each sync_progress page. If it fires, surfaces the retry
button with a "Timed out" message.
Generated with [Claude Code](https://claude.ai/code)
via [Happy](https://happy.engineering)
Co-Authored-By: Claude <noreply@anthropic.com>
Co-Authored-By: Happy <yesreply@happy.engineering>
2-finger pinch/pan, 3-finger pan, MagnifyGesture, and PanGesture are
now gated behind gesture mode. In buttons mode only single-finger drag
(controlled by the active toggle button) affects the camera.
Generated with [Claude Code](https://claude.ai/code)
via [Happy](https://happy.engineering)
Co-Authored-By: Claude <noreply@anthropic.com>
Co-Authored-By: Happy <yesreply@happy.engineering>
Cumulative drag distance (_touch_drag_total) is the wrong metric for tap
detection: natural finger jitter (A→B→A) can exceed 36 px even on a clean
tap, silently suppressing board clicks (unit moves, card plays, etc.).
Switch the board-click gate to start-to-end distance (< 50 px = tap).
_touch_drag_active now only means "camera mode fired during this touch" and
is set by the gesture timer or when an active camera mode consumes a drag.
This means taps always reach _handle_board_click regardless of finger wobble.
Generated with [Claude Code](https://claude.ai/code)
via [Happy](https://happy.engineering)
Co-Authored-By: Claude <noreply@anthropic.com>
Co-Authored-By: Happy <yesreply@happy.engineering>
_touch_drag_active was set unconditionally on 36px drag, silently eating
board taps even when no camera action occurred. Now only set when a camera
mode (orbit/pan/zoom) is actually active. Gesture timer also immediately
marks drag-active on activation so the 3s hold doesn't fire a board click.
Buttons enlarged to 44px minimum height for touch usability.
Generated with [Claude Code](https://claude.ai/code)
via [Happy](https://happy.engineering)
Co-Authored-By: Claude <noreply@anthropic.com>
Co-Authored-By: Happy <yesreply@happy.engineering>
- Settings: new touch_camera_mode ("gesture"|"buttons"), persisted to camera section
- GameScreen: CamGestureTimer (3 s), CamCtrlPanel with Orbit/Pan/Zoom toggle buttons
- game_screen: gesture timer fires orbit/pan/zoom after 3-second stationary hold;
button mode shows on-screen toggles; single-finger drag now gated behind active mode
- InGameSettings: Camera Controls section with Touch Camera OptionButton
Generated with [Claude Code](https://claude.ai/code)
via [Happy](https://happy.engineering)
Co-Authored-By: Claude <noreply@anthropic.com>
Co-Authored-By: Happy <yesreply@happy.engineering>
Left-click drag was orbiting the camera on desktop, so any mouse movement
>2px during a click would suppress _handle_board_click. Moved camera orbit
to right-mouse drag (_orbit_active flag) leaving left-click exclusively for
board actions regardless of mouse movement. Touch single-finger orbit is
unchanged.
Generated with [Claude Code](https://claude.ai/code)
via [Happy](https://happy.engineering)
Co-Authored-By: Claude <noreply@anthropic.com>
Co-Authored-By: Happy <yesreply@happy.engineering>
4px per-event threshold was too sensitive — normal finger tremor during
a tap could exceed it, causing _touch_drag_active to suppress the click.
Now accumulates total drag distance since touch-down; requires 12px total
movement before classifying the touch as a camera orbit.
Generated with [Claude Code](https://claude.ai/code)
via [Happy](https://happy.engineering)
Co-Authored-By: Claude <noreply@anthropic.com>
Co-Authored-By: Happy <yesreply@happy.engineering>
Added _touch_drag_active flag set only when finger moves > 4px threshold.
Camera orbit now requires meaningful drag; taps no longer jiggle camera
or get swallowed — board clicks fire correctly for gameplay actions.
Generated with [Claude Code](https://claude.ai/code)
via [Happy](https://happy.engineering)
Co-Authored-By: Claude <noreply@anthropic.com>
Co-Authored-By: Happy <yesreply@happy.engineering>