WebSocket Message Schemas (WS-002):
- Add Pydantic models for all client/server WebSocket messages
- Implement discriminated unions for message type parsing
- Include JoinGame, Action, Resign, Heartbeat client messages
- Include GameState, ActionResult, Error, TurnStart server messages
Connection Manager (WS-003):
- Add Redis-backed WebSocket connection tracking
- Implement user-to-sid mapping with TTL management
- Support game room association and opponent lookup
- Add heartbeat tracking for connection health
Socket.IO Authentication (WS-004):
- Add JWT-based authentication middleware
- Support token extraction from multiple formats
- Implement session setup with ConnectionManager integration
- Add require_auth helper for event handlers
Socket.IO Server Setup (WS-001):
- Configure AsyncServer with ASGI mode
- Register /game namespace with event handlers
- Integrate with FastAPI via ASGIApp wrapper
- Configure CORS from application settings
Game Service (GS-001):
- Add stateless GameService for game lifecycle orchestration
- Create engine per-operation using rules from GameState
- Implement action-based RNG seeding for deterministic replay
- Add rng_seed field to GameState for replay support
Architecture verified:
- Core module independence (no forbidden imports)
- Config from request pattern (rules in GameState)
- Dependency injection (constructor deps, method config)
- All 1090 tests passing
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Create tests/conftest.py with testcontainers for Postgres and Redis
- Auto-detect Docker Desktop socket and disable Ryuk for compatibility
- Update tests/db/conftest.py and tests/services/conftest.py to use shared fixtures
- Fix test_resolve_effect_logs_exceptions: logger was disabled by pytest
- Fix test_save_and_load_with_real_redis: use redis_url fixture
- Minor lint fix in engine_validation.py
Tests now auto-start containers on run - no need for `docker compose up`
All 1199 tests passing.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Project plan updates:
- Mark all 14 Phase 3 tasks as completed
- Update acceptance criteria to met
- Update master plan status to Phase 4 next
- Add detailed deliverables list for Phase 3
Documentation updates:
- Add UNSET sentinel pattern to CLAUDE.md
- Document when to use UNSET vs None for nullable fields
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
UNSET sentinel pattern:
- Add UNSET sentinel in protocols.py for nullable field updates
- Fix inability to clear deck description (UNSET=keep, None=clear)
- Fix repository inability to clear validation_errors
Starter deck improvements:
- Remove unused has_starter_deck from CollectionService
- Add deprecation notes to old starter deck methods
Validation improvements:
- Add energy type validation in deck_validator.py
- Add energy type validation in deck schemas
- Add VALID_ENERGY_TYPES constant
Game loading fix:
- Fix get_deck_for_game silently skipping invalid cards
- Now raises ValueError with clear error message
Tests:
- Add TestEnergyTypeValidation test class
- Add TestGetDeckForGame test class
- Add tests for validate_energy_types utility function
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Critical fixes:
- Add admin API key authentication for admin endpoints
- Add race condition protection via unique partial index for starter decks
- Make starter deck selection atomic with combined method
Moderate fixes:
- Fix DI pattern violation in validate_deck_endpoint
- Add card ID format validation (regex pattern)
- Add card quantity validation (1-99 range)
- Fix exception chaining with from None (B904)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Implemented with Repository Protocol pattern for offline fork support:
- CollectionService with PostgresCollectionRepository
- DeckService with PostgresDeckRepository
- DeckValidator with DeckConfig + CardService injection
- Starter deck definitions (5 types: grass, fire, water, psychic, lightning)
- Pydantic schemas for collection and deck APIs
- Unit tests for DeckValidator (32 tests passing)
Architecture follows pure dependency injection - no service locator patterns.
Added CLAUDE.md documenting DI requirements and patterns.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Discord OAuth callback now redirects to frontend with tokens in URL fragment
(more secure than query params - fragment not sent to server)
- Error responses also redirect with error in query params
- Fix deps.py import: get_session_dependency -> get_session
Part of Phase 2 Authentication completion.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Complete OAuth-based authentication with JWT session management:
Core Services:
- JWT service for access/refresh token creation and verification
- Token store with Redis-backed refresh token revocation
- User service for CRUD operations and OAuth-based creation
- Google and Discord OAuth services with full flow support
API Endpoints:
- GET /api/auth/{google,discord} - Start OAuth flows
- GET /api/auth/{google,discord}/callback - Handle OAuth callbacks
- POST /api/auth/refresh - Exchange refresh token for new access token
- POST /api/auth/logout - Revoke single refresh token
- POST /api/auth/logout-all - Revoke all user sessions
- GET/PATCH /api/users/me - User profile management
- GET /api/users/me/linked-accounts - List OAuth providers
- GET /api/users/me/sessions - Count active sessions
Infrastructure:
- Pydantic schemas for auth/user request/response models
- FastAPI dependencies (get_current_user, get_current_premium_user)
- OAuthLinkedAccount model for multi-provider support
- Alembic migration for oauth_linked_accounts table
Dependencies added: email-validator, fakeredis (dev), respx (dev)
84 new tests, 1058 total passing
- Derive image_path from card ID instead of raw data image_file field
- Use simplified CDN paths: /<set>/<card>.webp (e.g., a1/033-charmander.webp)
- Energy cards use basic/<type>.webp paths
- Fix undefined variable bug in transform_trainer_card
- Update tests to match new path format
- Regenerate all 382 card definitions with correct image_url fields
The converter now falls back to using the source image_url from raw
data when image_file is not available (images not downloaded locally).
- Pokemon cards: use source image_url when image_file is null
- Trainer cards: same fallback behavior
- Regenerated all 382 card definitions with image URLs
Images point to pokemon-zone.com assets for now. When we host our
own images, cards with image_file will use our CDN instead.
- Add lifespan context manager to app/main.py with startup/shutdown hooks
- Wire startup: init_db(), init_redis(), CardService.load_all()
- Wire shutdown: close_db(), close_redis()
- Add /health/ready endpoint for readiness checks
- Add CORS middleware with configurable origins
- Disable docs in production (only available in dev)
- Export get_session_dependency from app/db/__init__.py for FastAPI DI
- Add game_cache_ttl_seconds to Settings (configurable, was hardcoded)
- Fix datetime.utcnow() deprecation (4 occurrences) -> datetime.now(UTC)
- Update test to match S3 image URL (was placeholder CDN)
All 974 tests passing.
The source website uses <span class='energy-text energy-text--type-fire'>
to render inline energy icons. BeautifulSoup's get_text() was stripping
these spans, losing the energy type information and causing merged text
like 'Discard aEnergy' instead of 'Discard a Fire Energy'.
Changes:
- Add ENERGY_TEXT_TYPES mapping for inline energy references
- Add replace_energy_text_spans() to convert spans to text before extraction
- Add extract_effect_text() helper with proper text joining (separator=' ')
- Update parse_attack(), parse_ability(), _parse_trainer_details() to use it
- Fix JSON encoding in convert_cards.py to use UTF-8 (ensure_ascii=False)
Before: 'Discard an Energy from this Pokémon'
After: 'Discard a Fire Energy from this Pokémon'
Re-scraped all 372 cards and regenerated 382 definitions.
Scraper fixes:
- Detect fossil cards (Helix/Dome Fossil, Old Amber) as Trainer/Item cards
- Add text artifact cleaning for stripped energy icons:
- 'aEnergy' -> 'an Energy'
- 'extraEnergy' -> 'extra Energy'
- 'BenchedPokémon' -> 'Benched Pokémon'
- And 20+ other common patterns
Converter improvements:
- Add evolution chain validation to detect broken evolves_from references
- Track conversion errors and validation warnings in _index.json
- Return errors from convert_set() for better debugging
Data fixes:
- Fixed 4 fossil cards (now correctly typed as trainer/item)
- Fixed text artifacts in 46 raw card files
- Regenerated all 382 card definitions
- All evolution chains now valid
Added fix_raw_text.py utility script for batch text cleanup.
- Rename data/cards/ to data/raw/ for scraped data
- Add data/definitions/ as authoritative card data source
- Add convert_cards.py script to transform raw -> definitions
- Generate 378 card definitions (344 Pokemon, 24 Trainers, 10 Energy)
- Add CardService for loading and querying card definitions
- In-memory indexes for fast lookups by type, set, pokemon_type
- search() with multiple filter criteria
- get_all_cards() for GameEngine integration
- Add SetInfo model for set metadata
- Update Attack model with damage_display field for variable damage
- Update CardDefinition with image_path, illustrator, flavor_text
- Add 45 tests (21 converter + 24 CardService)
- Update scraper output path to data/raw/
Card data is JSON-authoritative (no database) to support offline fork goal.
Phase 1 Database Implementation (DB-001 through DB-012):
Models:
- User: OAuth support (Google/Discord), premium subscriptions
- Collection: Card ownership with CardSource enum
- Deck: JSONB cards/energy_cards, validation state
- CampaignProgress: One-to-one with User, medals/NPCs as JSONB
- ActiveGame: In-progress games with GameType enum
- GameHistory: Completed games with EndReason enum, replay data
Infrastructure:
- Alembic migrations with sync psycopg2 (avoids async issues)
- Docker Compose for Postgres (5433) and Redis (6380)
- App config with Pydantic settings
- Redis client helper
Test Infrastructure:
- 68 database tests (47 model + 21 relationship)
- Async factory pattern for test data creation
- Sync TRUNCATE cleanup (solves pytest-asyncio event loop mismatch)
- Uses dev containers instead of testcontainers for reliability
Key technical decisions:
- passive_deletes=True for ON DELETE SET NULL relationships
- NullPool for test sessions (no connection reuse)
- expire_on_commit=False with manual expire() for relationship tests
- Fix is_ex detection to only match ' ex' suffix (fixes Exeggutor false positive)
- Fix trainer card detection with specific 'Trainer | Type' pattern
- Improve trainer effect text extraction from content body
- Scrape full A1 set: 286 cards
- Scrape full A1a set: 86 cards
- Total: 372 cards with images (some images failed due to rate limiting)
- Re-scraped affected cards to fix is_ex and trainer type issues
- Add --images flag to download images during scraping
- Add --download-images command to fetch images for existing card data
- Images saved to data/images/{set}/ directory
- Card JSON updated with image_file field (relative path)
- Uses Referer header for asset server compatibility
- Supports skip-if-exists for incremental downloads
- Add scrape_pokemon_pocket.py script to fetch card data from pokemon-zone.com
- Scrapes Pokemon, Trainer, and Energy cards with full metadata
- Includes image URLs for offline caching support
- Supports --set, --card, --limit, and --reindex CLI options
- Add beautifulsoup4 and requests as dev dependencies
- Create data/cards/ directory structure for card JSON files
Design Documentation:
- docs/ACTIVE_EFFECTS_DESIGN.md: Comprehensive design for persistent effect system
- Data model (ActiveEffect, EffectTrigger, EffectScope, StackingMode)
- Core operations (register, remove, query effects)
- Integration points (damage calc, energy counting, retreat, lifecycle)
- Effect categories from Pokemon Pocket card research (~372 cards)
- Example implementations (Serperior, Greninja, Mr. Mime, Victreebel)
- Post-launch TODO for generic modifier system
Module README Files:
- backend/app/core/README.md: Core engine overview and key classes
- backend/app/core/effects/README.md: Effects module index and quick reference
- backend/app/core/models/README.md: Models module with relationship diagram
Minor cleanup:
- Revert Bulbasaur weakness to Fire (was test change for Lightning)
- Clean up debug output in game walkthrough
- Add DamageCalculationResult model for transparent damage breakdown
- Implement _calculate_attack_damage with W/R modifiers (additive/multiplicative)
- Add _execute_attack_effect for future effect system integration
- Add _build_attack_message for detailed damage breakdown in messages
- Update _execute_attack to use new calculation pipeline
- Bulbasaur now properly weak to Lightning in walkthrough demo
New features:
- Weakness applies bonus damage (additive +X or multiplicative xN)
- Resistance reduces damage (minimum 0)
- State changes include weakness/resistance details for UI
- Messages show damage breakdown (e.g. 'base 10 +20 weakness')
Tests: 7 new tests covering additive/multiplicative W/R, type matching,
minimum damage floor, knockout triggers, and state change details
- Add await to engine.execute_action() calls (energy attach and attack)
- Update entry point to use asyncio.run()
- Fix game creation result handling (use game_creation.game)
- Fix AttachEnergyAction parameter name (energy_card_id)
Creates comprehensive documentation for AI agents working with app/core/:
- Architecture overview with component diagram
- Import patterns (correct enum imports from app.core.enums)
- Core patterns: game creation, action execution, effects, config
- Key classes: GameState, PlayerState, CardDefinition, CardInstance, Zone
- Testing patterns with SeededRandom for determinism
- Security rules for hidden information
- Module independence guidelines for offline fork support
- File organization and quick commands
Updates PROJECT_PLAN.json: 31/32 tasks complete.
Creates a comprehensive interactive demo that walks through:
- Card definition creation
- Rules configuration
- Deck building
- Game initialization via GameEngine
- Setup phase with Basic Pokemon placement
- Full turn cycle (draw, main, attack, end phases)
Uses colored terminal output and 'press Enter' prompts for
step-by-step exploration of the core game engine.
Architectural refactor to eliminate circular imports and establish clean
module boundaries:
- Move enums from app/core/models/enums.py to app/core/enums.py
(foundational module with zero dependencies)
- Update all imports across 30 files to use new enum location
- Set up clean export structure:
- app.core.enums: canonical source for all enums
- app.core: convenience exports for full public API
- app.core.models: exports models only (not enums)
- Add module exports to app/core/__init__.py and app/core/effects/__init__.py
- Remove circular import workarounds from game_state.py
This enables app.core.models to export GameState without circular import
issues, since enums no longer depend on the models package.
All 826 tests passing.
Merged unique content from CLAUDE.md into AGENTS.md:
- Project Overview section (campaign mode description)
- Full Tech Stack tables (frontend and backend)
- WebSocket Events pattern
- Game Engine Patterns (Card Effect System, Turn State Machine)
Removed CLAUDE.md to standardize on AGENTS.md for agent guidelines.
Relocated SYSTEM_REVIEW.md and PROJECT_PLAN_ENERGY_EVOLUTION.md from
backend/docs/legacy to project-level docs/legacy. Added docs/README.md
indexing all documentation including ARCHITECTURE.md and GAME_RULES.md.
Relocated SYSTEM_REVIEW.md and PROJECT_PLAN_ENERGY_EVOLUTION.md to
docs/legacy/ now that all issues are resolved. Added docs/README.md
with index of documentation files.
Issue #2 gap: Added 14 CardDefinition validation tests covering all
required field checks (hp, stage, pokemon_type, evolves_from, trainer_type,
energy_type) with both negative and positive test cases.
Issue #7 gap: Added 4 confusion attack engine tests covering heads/tails
outcomes, self-damage, self-KO with opponent scoring, and configurable
damage from RulesConfig.
Issue #13 documentation: Added TODO comments in engine.py and handlers.py
documenting the expected pattern for knockout detection when effect
execution is implemented. Effect handlers set knockout flags; engine
should process knockouts after all effects resolve.
825 tests passing (+17 new tests)
Effect handler exceptions now logged at ERROR level with full context:
- effect_id, source_player_id, source/target card IDs, params
- Full traceback via logger.exception()
Game still returns safe EffectResult.failure() to prevent crashes,
but debugging information is now preserved in logs.
Both deal_damage and attack_damage now check if the target is knocked out
after applying damage. If KO'd, EffectResult includes:
- details['knockout'] = True
- details['knockout_pokemon_id'] = target's instance_id
- Message includes 'knocked out!' notification
Knockout check correctly respects HP modifiers via effective_hp().
Added 9 tests covering knockout detection, HP modifier behavior,
weakness-triggered knockouts, and resistance preventing knockouts.
Added stadium_owner_id field to GameState to track who played the stadium:
- stadium_owner_id: str | None tracks the player who played the current stadium
- When a stadium is replaced, old stadium discards to OWNER's pile (not current player)
- Added stadium_owner_id to VisibleGameState for client visibility
- Updated existing test and added 2 new tests for stadium ownership
This fixes the bug where replacing an opponent's stadium would discard
to the current player's pile instead of the opponent's.
797 tests passing.
Changed forced_action from single item to FIFO queue to support
scenarios where multiple forced actions are needed simultaneously:
- forced_actions: list[ForcedAction] replaces forced_action: ForcedAction | None
- Added queue management methods:
- has_forced_action() - check if queue has pending actions
- get_current_forced_action() - get first action without removing
- add_forced_action(action) - add to end of queue
- pop_forced_action() - remove and return first action
- clear_forced_actions() - clear all pending actions
- Updated engine, turn_manager, rules_validator, and visibility filter
- Added 8 new tests for forced action queue including double knockout scenario
This fixes the bug where simultaneous knockouts (e.g., mutual poison damage)
would lose one player's select_active action due to overwriting.
795 tests passing.
Changed ability_uses_this_turn from int to dict[int, int] to track each
ability's usage independently:
- ability_uses_this_turn: dict[int, int] maps ability index to use count
- can_use_ability() now requires ability_index parameter
- Added get_ability_uses() and increment_ability_uses() helper methods
- reset_turn_state() clears the dict instead of setting to 0
This fixes the bug where using one ability could incorrectly block
another ability on the same Pokemon from being used.
789 tests passing.
Issues #2, #3, #4, #6 are still open:
- #2: CardDefinition field validation not added
- #3: end_turn() knockout processing not fixed
- #4: Win condition timing not fixed
- #6: Engine doesn't call process_knockout for status KOs
The refactor improved process_knockout() internals but didn't fix the
callers that should invoke it.
Mark issues #1, #5, #7, #8 as FIXED:
- #1: find_card_instance now includes energy_zone + attached cards
- #5: Energy stored as CardInstance objects, not just IDs
- #7: Confusion handling added to attack execution
- #8: discard_energy handler moves cards to owner's discard
Update test count to 766 and add change log section.
Major refactor to properly track attached cards and evolution history:
Model Changes (app/core/models/card.py):
- Change attached_energy from list[str] to list[CardInstance]
- Change attached_tools from list[str] to list[CardInstance]
- Add cards_underneath field for evolution stack tracking
- Update attach_energy/detach_energy to work with CardInstance
- Add attach_tool/detach_tool methods
- Add get_all_attached_cards helper
Engine Changes (app/core/engine.py):
- _execute_attach_energy: Pass full CardInstance to attach_energy
- _execute_evolve: Build evolution stack, transfer attachments, clear status
- _execute_retreat: Detached energy goes to discard pile
- Fix: Evolution now clears status conditions (Pokemon TCG standard)
Game State (app/core/models/game_state.py):
- find_card_instance now searches attached_energy, attached_tools, cards_underneath
Turn Manager (app/core/turn_manager.py):
- process_knockout: Discard all attached energy, tools, and evolution stack
Effects (app/core/effects/handlers.py):
- discard_energy: Find owner's discard pile and move detached energy there
- NEW devolve effect: Remove evolution stages with configurable destination
- Fix: Use EffectType.SPECIAL instead of non-existent EffectType.ZONE
Rules Validator (app/core/rules_validator.py):
- Update energy type checking to iterate CardInstance objects
Tests:
- Update existing tests for new CardInstance-based energy attachment
- NEW test_evolution_stack.py with 28 comprehensive tests covering:
- Evolution stack building (Basic -> Stage 1 -> Stage 2)
- Energy/tool transfer and damage carryover on evolution
- Devolve effect (single/multi stage, hand/discard destination, KO check)
- Knockout processing with all attachments going to discard
- find_card_instance for attached cards and evolution stack
All 765 tests pass.
