docs: adopt unified-3.0 Claude rule set + trim duplications

Adopts the four-file rule set the player added to the working tree:

- CLAUDE.md grows from a 114-line pointer doc to the 571-line
  `unified-3.0` rulebook: hard global constraints (§2), engine
  rules (§3), asset rules (§4), code standards (§5), build +
  verification (§6), git workflow (§7), the change-control
  ASK BEFORE list (§8), and the Context Injection System (§14).
- CLAUDE_SPEC.md — formal architecture spec: crate dependency
  graph with forbidden_deps, data ownership map, state-machine
  invariants ("52 cards always exist", "no duplicate IDs",
  "all cards belong to exactly one pile"), sync merge contract,
  server contract, validation checklist.
- CLAUDE_WORKFLOW.md — two-agent Builder/Guardian pipeline with
  hard-fail patterns that auto-reject (core uses IO/Bevy/network,
  GameState mutated outside GameLogicSystem, blocking async on
  main thread, duplicate logic, merge altered incorrectly).
- CLAUDE_PROMPT_PACK.md — task-type templates.

Three duplicate rule passages removed:

- CLAUDE_SPEC.md §0 dropped no_panics_in_core / core_is_pure /
  event_driven_engine — already canonical in CLAUDE.md §2.1, §2.3,
  §3.1. Kept single_source_of_truth and sync_is_additive (those
  describe data flow, not in CLAUDE.md).
- CLAUDE_SPEC.md §11 Prohibited Patterns now references CLAUDE.md
  §11 instead of restating the same five forbidden items.
- ARCHITECTURE.md Design Principles dropped the pure-core /
  no-panics / UI-first bullets — those are enforcement constraints
  living in CLAUDE.md §2.1, §2.3, §3.3; this file describes the
  design that motivates them. Kept the offline-first, one-language,
  and plugin-based-Bevy bullets (those are descriptive, not
  enforcement).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

CLAUDE.md

@@ -1,114 +1,571 @@
-# Solitaire Quest — Claude Code Instructions
+# CLAUDE.md
 
-See @ARCHITECTURE.md for full project design, crate responsibilities, data models, and API reference.
+version: unified-3.0
 
 ---
 
-## Project Layout
+# 0. Role of This File
 
-```text
-solitaire_core/    # Pure Rust game logic — NO Bevy, NO network, NO I/O
-solitaire_sync/    # Shared API types — NO Bevy, serde/uuid/chrono only
-solitaire_data/    # Persistence + SyncProvider trait + server client
-solitaire_engine/  # Bevy ECS systems, components, plugins
-solitaire_server/  # Axum sync server binary
-solitaire_app/     # Thin binary entry point
-assets/            # Source assets — embedded at compile time via include_bytes!()
+This document defines:
+
+* **Execution rules (what Claude must do)**
+* **System constraints (what Claude must never violate)**
+* **Operational architecture (how code is structured)**
+
+For full system design details:
+→ `ARCHITECTURE.md` (authoritative source of truth)
+
+This file overrides all conversational assumptions.
+
+---
+
+# 1. System Architecture (Authoritative Mapping)
+
+## 1.1 Crates
+
+```text id="crate_map"
+solitaire_core/    # PURE logic (no IO, no Bevy, deterministic)
+solitaire_sync/    # Shared API + merge logic
+solitaire_data/    # Persistence + sync client
+solitaire_engine/  # Bevy ECS + UI + gameplay orchestration
+solitaire_server/  # Axum backend (optional sync layer)
+solitaire_app/     # Entry binary
+assets/            # Runtime assets (except audio)
 ```
 
 ---
 
-## Build & Test Commands
+## 1.2 Architecture Source of Truth
 
-```bash
-# Dev run (fast compile via dynamic linking)
-cargo run -p solitaire_app --features bevy/dynamic_linking
+* Full system design: `ARCHITECTURE.md`
+* This file NEVER redefines system design
+* This file ONLY enforces behavior
 
-# Release build
-cargo build --workspace --release
+---
 
-# All tests — MUST pass before any commit
+# 2. Hard Global Constraints (NON-NEGOTIABLE)
+
+These override all other instructions.
+
+## 2.1 Core Determinism
+
+* `solitaire_core` MUST:
+
+  * be deterministic
+  * be side-effect free
+  * never depend on Bevy / IO / async
+
+---
+
+## 2.2 Sync Isolation
+
+* `solitaire_sync`:
+
+  * no Bevy
+  * no IO
+  * no engine dependencies
+* merge logic must be pure functions only
+
+---
+
+## 2.3 Error Policy
+
+* NO `unwrap()`
+* NO `panic!()` in runtime/game logic
+* All state transitions:
+
+```rust id="err_model"
+Result<T, MoveError>
+```
+
+---
+
+## 2.4 Threading Rules
+
+* Sync must run on `AsyncComputeTaskPool`
+* NEVER block Bevy main thread
+
+---
+
+## 2.5 Persistence Rules
+
+* atomic writes only:
+
+  * write `.tmp`
+  * rename atomically
+* no partial state writes allowed
+
+---
+
+## 2.6 Security Rules
+
+* credentials ONLY via `keyring`
+* NEVER store secrets in:
+
+  * files
+  * logs
+  * source code
+
+---
+
+## 2.7 Sync System Rules
+
+* All sync backends implement:
+
+```rust id="sync_trait"
+trait SyncProvider
+```
+
+* `SyncPlugin` MUST be backend-agnostic
+* NEVER match on backend inside ECS systems
+
+---
+
+# 3. Engine Rules (Bevy Layer)
+
+## 3.1 ECS Design
+
+* systems = single responsibility
+* communication = Events only
+* shared state = Resources only
+* per-entity state = Components only
+
+---
+
+## 3.2 Game State Authority
+
+* ONLY `GameStateResource` can mutate game state
+* UI systems MUST NOT directly modify core logic
+
+---
+
+## 3.3 UI-First Constraint (CRITICAL)
+
+Every player action MUST:
+
+* have a visible UI control
+* NOT rely solely on keyboard shortcuts
+
+Keyboard shortcuts are:
+→ optional accelerators only
+
+---
+
+## 3.4 Layout System
+
+* recompute on `WindowResized`
+* no fixed resolution assumptions
+
+---
+
+# 4. Asset System Rules
+
+## 4.1 Runtime Assets (AssetServer)
+
+Loaded via:
+
+* `CardImageSet`
+* `BackgroundImageSet`
+* `FontResource`
+
+Includes:
+
+* cards
+* backgrounds
+* fonts
+
+---
+
+## 4.2 Embedded Assets
+
+Only audio:
+
+```text id="audio_rule"
+include_bytes!()
+```
+
+---
+
+## 4.3 Test Compatibility Rule
+
+All asset loaders MUST accept:
+
+```rust id="asset_fallback"
+Option<Res<AssetServer>>
+```
+
+Must degrade gracefully under `MinimalPlugins`.
+
+---
+
+# 5. Code Standards
+
+## 5.1 Error Handling
+
+* use `thiserror`
+* no `Box<dyn Error>` in libraries
+
+---
+
+## 5.2 Public API Rules
+
+* prefer `Into<T>` over concrete types
+* all public items require doc comments
+
+---
+
+## 5.3 Derive Order
+
+```rust id="derive_order"
+#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
+```
+
+---
+
+## 5.4 Performance Rules
+
+* NO `clone()` in hot paths
+* profile before optimizing
+
+---
+
+## 5.5 SQL Rules
+
+* ONLY `sqlx::query!`
+* NO raw SQL strings
+
+---
+
+# 6. Build & Verification Rules
+
+These are mandatory before ANY commit.
+
+```bash id="build_rules"
 cargo test --workspace
-
-# Lint — MUST pass clean (zero warnings)
 cargo clippy --workspace -- -D warnings
-
-# Run sync server locally
-cargo run -p solitaire_server
-
-# Check a single crate
-cargo test -p solitaire_core
-cargo clippy -p solitaire_core -- -D warnings
 ```
 
 ---
 
-## Hard Rules
+# 7. Git Workflow Rules
 
-- `solitaire_core` and `solitaire_sync` must never gain Bevy or network dependencies.
-- No `unwrap()` or `panic!()` in game logic. All state transitions return `Result<_, MoveError>`.
-- Audio assets are embedded at compile time using `include_bytes!()` in `audio_plugin.rs`.
-- Card faces (52 PNGs in `assets/cards/faces/`), card backs (`assets/cards/backs/back_N.png`), board backgrounds (`assets/backgrounds/bg_N.png`), and the UI font (`assets/fonts/main.ttf`) are loaded at runtime via `AssetServer::load()` and stored as `Handle<Image>`/`Handle<Font>` in the `CardImageSet`, `BackgroundImageSet`, and `FontResource` resources. The `assets/` directory must ship alongside the binary.
-- Asset-loading systems take `Option<Res<AssetServer>>` so they degrade cleanly under `MinimalPlugins` (tests). When `CardImageSet` is absent, `card_plugin` falls back to a `Text2d` rank+suit overlay; when `BackgroundImageSet` is absent, the board falls back to a solid colour.
-- Atomic file writes only: write to `filename.json.tmp`, then `rename()`.
-- Passwords and tokens are stored in the OS keychain via the `keyring` crate — never in plaintext files or logs.
-- Sync runs on `AsyncComputeTaskPool` — never block the Bevy main thread.
-- All sync backends implement the `SyncProvider` trait. The `SyncPlugin` is backend-agnostic — never `match` on `SyncBackend` inside a Bevy system.
-- `cargo clippy --workspace -- -D warnings` must pass clean after every change.
-- `cargo test --workspace` must pass after every change.
+## Commit format
+
+```text id="commit_fmt"
+type(scope): description
+```
+
+Examples:
+
+* feat(core): add draw-three rules
+* fix(engine): correct drag z-order
+* test(core): undo boundary cases
 
 ---
 
-## Code Style
+## Commit conditions
 
-- Use `thiserror` for error types. Never `Box<dyn Error>` in library crates.
-- Prefer `Into<T>` over concrete types in public API function parameters.
-- All public items must have doc comments (`///`). Private items: comment only when non-obvious.
-- Derive order convention: `#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]`
-- Bevy systems: one responsibility per system. Use `Events` for cross-system communication, never shared mutable state.
-- SQL queries: use `sqlx::query!` macros (compile-time checked), not raw string queries.
-- No `clone()` calls in hot paths (game loop systems). Profile before optimising elsewhere.
+* tests must pass
+* clippy must be clean
+
+NEVER commit otherwise
 
 ---
 
-## Bevy Conventions
+# 8. Change Control (ASK BEFORE DOING)
 
-- One `Plugin` per major feature: `CardPlugin`, `AudioPlugin`, `AchievementPlugin`, `UIPlugin`, `SyncPlugin`.
-- Resources own shared state. Events communicate between systems. Components own per-entity data.
-- All UI screens are built with Bevy UI (`bevy::ui`). Never mix UI layout and game logic in the same system.
-- Layout is recomputed on `WindowResized` — never assume a fixed window size.
-- **UI-first.** Every player-triggered action (new game, undo, draw, pause, open stats / settings / help / profile / leaderboard, switch mode, etc.) must be reachable from a visible UI control. Keyboard shortcuts are optional accelerators — never the sole entry point. New gameplay features ship with the UI control alongside the system that backs it; do not merge a feature that is keyboard-only.
+Claude must request confirmation before:
+
+* adding dependencies
+* modifying `solitaire_sync`
+* changing DB schema
+* introducing `unsafe`
+* changing merge strategy
 
 ---
 
-## Git Workflow
+# 9. System Mental Model (IMPORTANT)
 
-- Commit after each passing phase, not after every file change.
-- Commit message format: `type(scope): description`
-  - `feat(core): add draw-three mode validation`
-  - `fix(engine): card z-order during drag`
-  - `test(core): undo stack boundary conditions`
-  - `chore(server): add sqlx migration 002`
-- Never commit with failing tests or clippy warnings.
-- Never commit secrets, `.env` files, or `*.db` files.
+```text id="mental_model"
+Core (rules + deterministic logic)
+    ↓
+Engine (Bevy orchestration)
+    ↓
+Data layer (persistence + sync)
+    ↓
+Server (optional external system)
+```
+
+Core is always the source of truth.
 
 ---
 
-## Ask Before Doing
+# 10. Known Platform Pitfalls
 
-- Adding a new crate dependency (discuss alternatives first).
-- Changing a type in `solitaire_sync` (breaking change on both client and server).
-- Altering the database schema (requires a new sqlx migration).
-- Introducing `unsafe` code anywhere.
-- Changing the merge strategy in `solitaire_sync::merge()`.
+Must always be handled explicitly:
+
+* Bevy `Time` uses `f32`
+* `sqlx::migrate!()` path is crate-relative
+* `dirs::data_dir()` may return `None`
+* Linux may lack keyring backend
 
 ---
 
-## Lessons Learned
+# 11. Forbidden Patterns
 
-> Add entries here when Claude makes a mistake so it isn't repeated.
+* game logic inside Bevy systems
+* duplication across crates
+* blocking async calls in ECS
+* insecure credential storage
+* bypassing core logic layer
 
-- Bevy's `Time` resource uses `f32` seconds; convert to `u64` only when writing to `StatsSnapshot`.
-- `sqlx::migrate!()` macro path is relative to the crate root, not the workspace root.
-- `keyring` on Linux requires a running secret service (e.g. GNOME Keyring or KWallet) — handle `Error::NoStorageAccess` gracefully and fall back to prompting the user.
-- `dirs::data_dir()` returns `None` on some minimal Linux environments — always handle the `None` case explicitly, do not unwrap.
+---
+
+# 12. Execution Rules for Claude
+
+When generating code:
+
+1. respect crate boundaries
+2. minimize diff size
+3. do not expand scope
+4. follow existing patterns
+5. preserve invariants
+
+If unclear:
+→ ask before acting
+
+---
+
+# 13. Relationship to ARCHITECTURE.md
+
+| File            | Role                      |
+| --------------- | ------------------------- |
+| CLAUDE.md       | execution + constraints   |
+| ARCHITECTURE.md | system design truth       |
+| Both combined   | full system understanding |
+
+---
+# 14. Context Injection System (AUTOMATIC SCOPE FILTER)
+
+## 14.1 Purpose
+
+Before generating any response, Claude MUST construct a **minimal relevant context set**.
+
+This prevents:
+
+* architectural drift
+* irrelevant spec loading
+* over-engineering
+* cross-crate confusion
+
+---
+
+## 14.2 Input Classification Step (MANDATORY)
+
+Every request MUST be classified into exactly one task type:
+
+```text id="task_types"
+feature
+bugfix
+refactor
+system_design
+bevy_system
+core_logic
+sync
+optimization
+test
+debug
+```
+
+If uncertain → ask clarification.
+
+---
+
+## 14.3 Context Selection Engine
+
+After classification, Claude MUST include ONLY the relevant sections below.
+
+---
+
+## 14.4 Context Map (CORE RULESET)
+
+### feature
+
+Include:
+
+* §2 Hard Global Constraints
+* §3 Engine Rules
+* ARCHITECTURE.md (crate of target feature only)
+* relevant data models (GameState, SyncPayload if needed)
+
+---
+
+### bugfix
+
+Include:
+
+* §2 Hard Global Constraints
+* §5 Code Standards
+* affected crate boundaries
+* relevant system (engine/core/sync only)
+
+---
+
+### refactor
+
+Include:
+
+* §3 Engine Rules
+* §5 Code Standards
+* §11 Forbidden Patterns
+* target crate boundaries
+
+---
+
+### system_design
+
+Include:
+
+* ARCHITECTURE.md (FULL)
+* §9 Mental Model
+* §1 System Architecture Mapping
+
+---
+
+### core_logic
+
+Include:
+
+* solitaire_core rules only
+* GameState model
+* MoveError model
+* §2.1–2.3 constraints
+
+---
+
+### bevy_system
+
+Include:
+
+* §3 Engine Rules
+* ECS rules (Events/Resources/Components)
+* UI-first constraint
+* relevant plugin system only
+
+---
+
+### sync
+
+Include:
+
+* SyncProvider trait
+* merge strategy rules
+* solitaire_sync models
+* §2.6 Sync Rules
+
+---
+
+### optimization
+
+Include:
+
+* target crate only
+* §5.4 Performance Rules
+* hot path constraints
+
+---
+
+### test
+
+Include:
+
+* §6 Build Rules
+* relevant module
+* expected invariants
+
+---
+
+### debug
+
+Include:
+
+* target file/module only
+* §2.3 Error Policy
+* runtime assumptions relevant to failure
+
+---
+
+## 14.5 Context Compression Rules
+
+Claude MUST obey:
+
+* never include full ARCHITECTURE.md unless system_design
+* max 2 crates per response unless explicitly required
+* prefer function-level context over file-level context
+* exclude unrelated plugins/systems
+
+---
+
+## 14.6 Context Priority Order
+
+When space is limited:
+
+1. Hard Constraints (§2)
+2. Target crate rules
+3. Data models
+4. Only then: architecture snippets
+
+---
+
+## 14.7 “No Context Pollution” Rule
+
+Claude must NOT include:
+
+* unrelated crates
+* unrelated plugins
+* unused data models
+* full architecture dumps
+* speculative systems
+
+---
+
+## 14.8 Self-Check Before Execution
+
+Before writing code, Claude MUST verify:
+
+* [ ] Is only relevant context included?
+* [ ] Is at least one hard constraint present?
+* [ ] Am I touching more than one crate unnecessarily?
+* [ ] Am I duplicating ARCHITECTURE.md content?
+
+If any fail → revise context selection.
+
+---
+
+## 14.9 Injection Output Format (Internal Model)
+
+Claude should behave as if it constructed:
+
+```text id="ctx_format"
+[SELECTED TASK TYPE]
+
+[MINIMAL REQUIRED RULES]
+
+[MINIMAL ARCHITECTURE SLICES]
+
+[RELEVANT MODELS]
+
+[REQUEST]
+```
+
+---
+
+## 14.10 Relationship to ARCHITECTURE.md
+
+* ARCHITECTURE.md = source of truth
+* CLAUDE.md = execution constraints
+* THIS SECTION = filtering layer between them
+
+---
+
+# END CONTEXT INJECTION SYSTEM