mzer0-yu · JasonNing96 · Mar 10, 2026
diff --git a/.thinkir-project-rules.md b/.thinkir-project-rules.md
@@ -0,0 +1,146 @@
+# ThinkIR 项目规则模板
+
+## 当在项目中使用 ThinkIR 时，请遵循以下规则：
+
+### 1. 代码组织规范
+
+#### ThinkIR 脚本位置
+```
+project/
+├── thinkir/
+│   ├── scripts/          # ThinkIR 脚本集合
+│   ├── templates/        # 脚本模板
+│   ├── examples/         # 示例输出
+│   └── rules.md         # 项目特定规则
+└── src/
+    └── generated/        # 生成的源代码
+```
+
+### 2. ThinkIR 使用流程
+
+#### Step 1: 需求分析
+- 明确编程目标
+- 确定约束条件
+- 选择合适的 ThinkIR 模式
+
+#### Step 2: 脚本编写
+- 使用模板开始
+- 添加逻辑节点
+- 设置约束和链接
+
+#### Step 3: 执行和验证
+- 选择执行模式
+- 验证生成结果
+- 调试和优化
+
+#### Step 4: 保存和归档
+- 保存 ThinkIR 脚本
+- 记录执行结果
+- 更新项目文档
+
+### 3. 项目集成规范
+
+#### 新项目初始化
+```bash
+# 创建项目结构
+mkdir -p thinkir/{scripts,templates,examples}
+cp templates/thinkir-template.cpp thinkir/templates/
+```
+
+#### 生成代码的规范
+- 所有生成的代码必须通过 ThinkIR 生成
+- 保留原始 ThinkIR 脚本以便复现
+- 添加必要的注释和文档
+
+### 4. 团队协作规范
+
+#### 代码审查
+- 审查 ThinkIR 脚本质量
+- 验证生成代码的正确性
+- 确保符合项目标准
+
+#### 知识共享
+- 记录常用的 ThinkIR 模式
+- 创建项目特定的 ThinkIR 库
+- 分享最佳实践
+
+### 5. 版本控制
+
+#### ThinkIR 脚本管理
+- 将 ThinkIR 脚本纳入版本控制
+- 使用有意义的文件名
+- 添加执行说明
+
+#### 生成代码追踪
+- 记录生成 ThinkIR 脚本的 commit
+- 标注代码来源
+- 便于追踪和回溯
+
+### 6. 质量保证
+
+#### ThinkIR 脚本质量
+- 使用清晰的逻辑节点
+- 添加适当的约束
+- 确保逻辑链完整
+
+#### 生成代码质量
+- 通过静态分析
+- 执行单元测试
+- 进行性能测试
+
+### 7. 项目特定规则
+
+根据项目类型，可以添加特定的 ThinkIR 规则：
+
+#### Web 项目规则
+```bash
+# 默认约束
+st("Use modern JavaScript/TypeScript")
+st("Include error handling")
+st("Follow project coding style")
+```
+
+#### 数据科学项目规则
+```bash
+# 默认约束
+st("Include data validation")
+st("Add performance metrics")
+st("Use pandas/numpy best practices")
+```
+
+#### API 项目规则
+```bash
+# 默认约束
+st("Include OpenAPI documentation")
+st("Add rate limiting")
+st("Implement proper error codes")
+```
+
+### 8. 快速命令
+
+#### 项目快捷命令
+```bash
+# 初始化 ThinkIR 环境
+/thinkir init
+
+# 查看项目状态
+/thinkir status
+
+# 切换模式
+/thinkir mode [command|exec|translate|silent]
+
+# 创建新脚本
+/thinkir new [script_name]
+```
+
+### 9. 故障排除
+
+#### 常见问题
+1. **LLM 兼容性** - 检查 LLM 是否支持 ThinkIR
+2. **输出格式** - 确认是否在正确的模式下
+3. **约束失效** - 检查语法和约束条件
+
+#### 调试步骤
+1. 使用 Translation 模式查看脚本
+2. 使用 Silent 模式查看干净输出
+3. 逐步添加节点进行测试
diff --git a/.thinkir-rules.md b/.thinkir-rules.md
@@ -0,0 +1,66 @@
+# ThinkIR 项目规则
+
+当在项目中启用 ThinkIR 模式时，以下规则将生效：
+
+## 自动规则
+
+### 1. 模式检测
+- 如果输入以 `//` 开头，自动进入 **Standard Mode**
+- 其他输入默认使用 **Command Line Mode**
+
+### 2. 输出格式化
+- Phase 1 使用 Markdown 代码块显示 ThinkIR 脚本
+- Phase 2 使用结构化格式显示拓扑和节点
+
+### 3. 环境初始化
+- 自动加载 ThinkIR.txt 规范
+- 输出 `// ThinkIR Environment Ready.` 作为确认
+
+## 快捷指令
+
+### 模式切换
+```
+// use execution mode          → 切换到 Execution Mode
+// use translation mode       → 切换到 Translation Mode
+// use silent mode            → 切换到 Silent Mode
+// use standard mode          → 切换到 Standard Mode
+```
+
+### 退出 ThinkIR
+```
+// [ThinkIR Exit]            → 完全退出 ThinkIR 模式
+```
+
+## 编程最佳实践
+
+### 推荐的 ThinkIR 结构
+```cpp
+#include "ThinkIR.hpp"
+
+LLMctx main() {
+    // 1. 设置上下文和目标
+    LLMctx logic = LLMctx::keyword("Domain Expert");
+    logic.purpose("Clear objective statement");
+
+    // 2. 添加约束（可选）
+    logic.st("Code must include error handling");
+
+    // 3. 创建逻辑节点
+    id t1 = logic.think("First logical deduction");
+    id t2 = logic.think("Second logical deduction");
+    logic.link(t1, t2);
+
+    // 4. 创建纯净容器（推荐）
+    LLMctx pure;
+
+    // 5. 生成最终输出
+    pure.push(logic.program("Generate code with specific requirements"));
+
+    return pure;
+}
+```
+
+### 节点类型说明
+- `[Think]` - 推理节点
+- `[Code]` - 代码节点
+- `[Create]` - 文档节点
diff --git a/CLAUDE.md b/CLAUDE.md
@@ -0,0 +1,67 @@
+# CLAUDE.md
+
+This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
+
+## Project Overview
+
+ThinkIR is a specification and documentation project for an intermediate scripting language designed to bridge natural language and programming. It's not a traditional software project but rather a conceptual framework for interacting with LLMs using structured intermediate representation.
+
+## Key Files
+
+- **README.md** - Comprehensive documentation about ThinkIR concept, usage examples, and comparison with natural language prompting
+- **ThinkIR.txt** - XML specification file (version 1.18.71) containing the complete language specification, operators, execution rules, and runtime pipeline
+- **BoostLicense** - Boost Software License file
+
+## Development Workflow
+
+Since this is a specification project, there are no traditional development commands:
+
+- **No build system** - This project doesn't use npm, yarn, pnpm, or any package manager
+- **No tests** - There is no test framework or testing infrastructure
+- **No scripts** - No package.json scripts or Makefiles
+
+## ThinkIR Language Architecture
+
+ThinkIR is based on C++26 syntax and operates through a dual-layer graph architecture:
+
+### Core Components
+1. **Logic Graph** (Bottom Layer) - Persistent reasoning nodes created with `think()` and connected with `link()`
+2. **Artifact Graph** (Top Layer) - Final deliverables (code, docs) created with `push()` and rendered with `create()`/`program()`
+
+### Execution Pipeline
+- **Phase 1**: Translation - Maps natural language to ThinkIR script
+- **Phase 2**: Serialization - Outputs topology and node content
+
+### Key Operators
+- `think()` - Creates persistent inference nodes
+- `st()` - Adds hard constraints
+- `link()` - Establishes logical dependencies
+- `keyword()`/`purpose()` - Sets context and goals
+- `push()` - Promotes artifacts to output
+- `create()` - Generates natural language text
+- `program()` - Generates executable code
+
+### Environment Modes
+- **Command Line Mode** (default) - Phase 1 + Phase 2
+- **Execution Mode** - Phase 2 only
+- **Translation Mode** - Phase 1 only
+- **Silent Mode** - Full pipeline, Phase 1 hidden
+- **Standard Mode** - Comment input exits ThinkIR mode
+
+## Working with ThinkIR
+
+When modifying this specification:
+1. **ThinkIR.txt is the source of truth** - All language specifications are defined here
+2. **XML structure must be maintained** - The specification follows strict XML formatting
+3. **Version tracking** - The specification version is tracked in the XML header
+4. **Update README.md** - When adding new features or changing behavior, update examples and documentation
+
+## Compatibility Notes
+
+The specification mentions compatibility with various LLMs:
+- Gemini 3 Fast/3 Thinking/3.1 Pro: Very good
+- DeepSeek v4: Very good
+- GPT-5.3: Good
+- Claude Sonnet 4.6: General
+
+When updating the specification, consider these compatibility constraints as different LLMs may have varying support for prompt injection techniques.