diff --git a/.github/workflows/pr-tests-and-format.yml b/.github/workflows/pr-tests-and-format.yml index 0dfd168..01c43a1 100644 --- a/.github/workflows/pr-tests-and-format.yml +++ b/.github/workflows/pr-tests-and-format.yml @@ -4,23 +4,48 @@ on: pull_request: jobs: - lint-and-test: + test-backend: runs-on: ubuntu-latest - steps: - uses: actions/checkout@v4 - - name: Set up uv uses: astral-sh/setup-uv@v2 with: python-version: "3.12" - cache: true - - name: Install dependencies run: uv sync --frozen + - name: Run pytest + run: uv run pytest + lint-backend: + runs-on: ubuntu-latest + steps: + - uses: actions/checkout@v4 + - name: Set up uv + uses: astral-sh/setup-uv@v2 + with: + python-version: "3.12" + - name: Install dependencies + run: uv sync --frozen - name: Check formatting with Black run: uv run black . --check - - name: Run pytest - run: uv run pytest + build-frontend: + runs-on: ubuntu-latest + steps: + - uses: actions/checkout@v4 + - name: Set up Node.js + uses: actions/setup-node@v4 + with: + node-version: '16' + - name: Cache Node.js modules + uses: actions/cache@v4 + with: + path: ui-web/frontend/node_modules + key: ${{ runner.os }}-node-${{ hashFiles('ui-web/frontend/package-lock.json') }} + restore-keys: | + ${{ runner.os }}-node- + - name: Install frontend dependencies + run: npm install --prefix ui-web/frontend --legacy-peer-deps + - name: Build frontend + run: npm run build --prefix ui-web/frontend diff --git a/ARCHITECTURE.md b/ARCHITECTURE.md index c49363c..8bfdda7 100644 --- a/ARCHITECTURE.md +++ b/ARCHITECTURE.md @@ -37,6 +37,10 @@ LieGraph follows a **state machine architecture** built on LangGraph, with clear │ │ Identity │ │ Speech │ │ Voting │ │ │ │ Inference │ │ Generation │ │ Logic │ │ │ └─────────────────┘ └─────────────────┘ └─────────────┘ │ +│ ┌─────────────────┐ ┌─────────────────┐ ┌─────────────┐ │ +│ │ Strategy │ │ Speech │ │ Vote │ │ +│ │ Coordination │ │ Tools │ │ Tools │ │ +│ └─────────────────┘ └─────────────────┘ └─────────────┘ │ ├─────────────────────────────────────────────────────────────┤ │ Configuration Layer │ │ ┌─────────────────┐ ┌─────────────────┐ ┌─────────────┐ │ @@ -77,9 +81,18 @@ LieGraph follows a **state machine architecture** built on LangGraph, with clear - `merge_private_states`: Incremental mindset updates - `add`: Append-only operations for immutable records -### 3. AI Strategy (`src/game/llm_strategy.py`) +### 3. AI Strategy System (`src/game/strategy/`) + +**Purpose**: Implements intelligent agent behavior through LLM reasoning with structured tools. + +**Core Components**: +- **Strategy Core (`strategy_core.py`)**: Main coordination for LLM-powered game intelligence +- **Prompt Builders (`builders/`)**: Context and prompt builders for different game phases +- **LLM Schemas (`llm_schemas.py`)**: Pydantic models for structured LLM outputs -**Purpose**: Implements intelligent agent behavior through LLM reasoning. +**Agent Tools (`src/game/agent_tools/`)**: +- **Speech Tools (`speech_tools.py`)**: Structured reasoning for tone, clarity, and tactical focus +- **Vote Tools (`vote_tools.py`)**: Evidence-based voting decisions with structured reasoning **Core Intelligence Systems**: - **Dynamic Identity Inference**: Real-time role analysis through conversation patterns @@ -191,4 +204,21 @@ Next Round or Game End ### Reliability - **Immutable State**: Append-only records for critical game actions - **Error Recovery**: Checkpoint-based state recovery -- **Validation**: Configuration and state validation at runtime \ No newline at end of file +- **Validation**: Configuration and state validation at runtime + +## Recent Architectural Improvements + +### Enhanced Strategy System +- **Modular Strategy Coordination**: Separated core coordination from prompt builders and LLM schemas +- **Structured Agent Tools**: Dedicated tools for speech planning and voting decisions with structured reasoning +- **Improved Serialization**: Enhanced serialization support across strategy modules + +### Metrics and Analysis +- **Historical Aggregation**: CLI tools for aggregating and analyzing game metrics summaries +- **Quality Scoring**: Deterministic and LLM-based quality evaluation methods +- **Comprehensive Tracking**: Enhanced metrics collection with detailed game progression analysis + +### Development Experience +- **Test Coverage**: Comprehensive test suite covering game rules, state management, nodes, and agent tools +- **Code Organization**: Better separation between workflow orchestration and AI intelligence +- **Error Handling**: Enhanced fallback mechanisms and error recovery strategies \ No newline at end of file diff --git a/README.md b/README.md index 465940f..f547b0f 100644 --- a/README.md +++ b/README.md @@ -14,7 +14,9 @@ LieGraph is a multi-agent implementation of the popular social deduction game "W - **Natural Language Interaction:** Agents communicate and reason in natural language throughout the game - **Probabilistic Belief System:** Sophisticated belief tracking with self-belief confidence and suspicions matrix - **Strategic Reasoning:** Advanced bluff detection, alliance formation, and long-term planning +- **LLM-driven Strategy:** Structured tools for speech planning and voting decisions - **Built-in Metrics:** Automatic quality tracking for win balance, identification accuracy, and speech diversity with JSON reports for prompt evaluation workflows +- **Historical Analysis:** CLI tools for aggregating and analyzing game metrics summaries ## 🚀 Quick Start @@ -200,7 +202,9 @@ LieGraph/ │ │ ├── state.py # Game state definitions │ │ ├── nodes/ # Graph node implementations │ │ ├── rules.py # Game logic and win conditions -│ │ └── llm_strategy.py # AI reasoning and speech generation +│ │ ├── strategy/ # AI strategy coordination and builders +│ │ ├── agent_tools/ # Structured tools for speech and voting +│ │ └── metrics.py # Game metrics and quality scoring ├── tests/ # Pytest test suite ├── ui-web/frontend/ # React web interface └── config.yaml # Game configuration diff --git a/README_zh.md b/README_zh.md index 3af487e..ed70355 100644 --- a/README_zh.md +++ b/README_zh.md @@ -15,7 +15,9 @@ LieGraph 是基于 LangGraph 构建的流行社交推理游戏 "谁是卧底" - **自然语言交互:** 智能体在整个游戏中以自然语言进行交流和推理 - **概率信念系统:** 具有自我信念置信度和怀疑矩阵的复杂信念追踪系统 - **战略推理:** 高级的虚张声势检测、联盟形成和长期规划 +- **LLM 驱动策略:** 用于发言规划和投票决策的结构化工具 - **内建指标:** 自动追踪胜率平衡、身份识别准确率与发言多样性，并生成 JSON 报告以供后续提示词评估使用 +- **历史分析:** 用于聚合和分析游戏指标摘要的 CLI 工具 ## 🚀 快速开始 @@ -201,7 +203,9 @@ LieGraph/ │ │ ├── state.py # 游戏状态定义 │ │ ├── nodes/ # 图节点实现 │ │ ├── rules.py # 游戏逻辑和胜利条件 -│ │ └── llm_strategy.py # AI 推理和发言生成 +│ │ ├── strategy/ # AI 策略协调和构建器 +│ │ ├── agent_tools/ # 用于发言和投票的结构化工具 +│ │ └── metrics.py # 游戏指标和质量评分 ├── tests/ # Pytest 测试套件 ├── ui-web/frontend/ # React 网页界面 └── config.yaml # 游戏配置