diff --git a/README.md b/README.md
index 77eb54d..f880634 100644
--- a/README.md
+++ b/README.md
@@ -2,7 +2,7 @@
Anthropic 官方 [Claude Code](https://docs.anthropic.com/en/docs/claude-code) CLI 工具的源码反编译/逆向还原项目。目标是将 Claude Code 大部分功能及工程化能力复现。虽然很难绷, 但是它叫做 CCB(踩踩背)...
-[项目解析文档在这里, 还非常初期想要](https://ccb.agent-aura.top/)
+[项目解析文档在这里, 还非常初期, 可以提想要解析的方向 issues](https://ccb.agent-aura.top/)
赞助商占位符
diff --git a/docs/context/compaction.mdx b/docs/context/compaction.mdx
index 6ce3a7b..ed7814a 100644
--- a/docs/context/compaction.mdx
+++ b/docs/context/compaction.mdx
@@ -15,6 +15,10 @@ description: "对话太长怎么办——优雅地'遗忘'不重要的信息"
不压缩的话,很快就会撞到 token 上限,对话被迫终止。
+
+
+
+
## 压缩的策略
Claude Code 提供了多层压缩机制:
diff --git a/docs/context/system-prompt.mdx b/docs/context/system-prompt.mdx
index b0038b3..4fdadf5 100644
--- a/docs/context/system-prompt.mdx
+++ b/docs/context/system-prompt.mdx
@@ -18,6 +18,10 @@ description: "AI 的'工作记忆'是如何在每次对话前被精心拼装的"
Claude Code 的 System Prompt 不是一段写死的文本,而是根据当前环境**实时组装**的:
+
+
+
+
| 组成部分 | 内容 | 来源 |
|----------|------|------|
| 基础人设 | 角色定义、行为准则 | 内置模板 |
diff --git a/docs/conversation/streaming.mdx b/docs/conversation/streaming.mdx
index c592ab5..1cd5857 100644
--- a/docs/conversation/streaming.mdx
+++ b/docs/conversation/streaming.mdx
@@ -18,16 +18,9 @@ description: "为什么 Claude Code 的回答是'打字机效果'而不是一整
一次 AI 响应中可能同时包含文字和工具调用。流式系统需要处理这种交织:
-```
-AI 开始输出文字 → "我来看看这个文件的内容..."
-AI 发出工具调用 → [FileRead: src/main.ts]
- ↓ 暂停流式输出
- 工具执行中...
- 结果返回给 AI
- ↓ 恢复流式输出
-AI 继续输出 → "这个文件里有一个 bug,第 42 行..."
-AI 发出下一个工具调用 → [FileEdit: src/main.ts]
-```
+
+
+
## 流式工具执行
diff --git a/docs/conversation/the-loop.mdx b/docs/conversation/the-loop.mdx
index 4b09ff5..fc8bba5 100644
--- a/docs/conversation/the-loop.mdx
+++ b/docs/conversation/the-loop.mdx
@@ -12,7 +12,9 @@ Claude Code 不一样:你说一个需求,它可能连续执行十几步操
这背后的机制叫做 **Agentic Loop**(智能体循环):
-{/* TODO: 插入 Loop 示意图 */}
+
+
+
diff --git a/docs/extensibility/mcp-protocol.mdx b/docs/extensibility/mcp-protocol.mdx
index 20e4326..ba8a98e 100644
--- a/docs/extensibility/mcp-protocol.mdx
+++ b/docs/extensibility/mcp-protocol.mdx
@@ -20,6 +20,10 @@ Claude Code 内置了 50+ 工具,覆盖了通用的软件开发需求。但每
**Model Context Protocol**(模型上下文协议)是 Anthropic 提出的开放标准,定义了 AI 与外部工具之间的通信方式。
+
+
+
+
类比:USB 是电脑连接外设的标准接口。MCP 是 AI 连接外部能力的标准接口。
## 工作原理
diff --git a/docs/images/agentic-loop.png b/docs/images/agentic-loop.png
new file mode 100644
index 0000000..2f5744e
Binary files /dev/null and b/docs/images/agentic-loop.png differ
diff --git a/docs/images/architecture-layers.png b/docs/images/architecture-layers.png
new file mode 100644
index 0000000..332be6c
Binary files /dev/null and b/docs/images/architecture-layers.png differ
diff --git a/docs/images/compaction.png b/docs/images/compaction.png
new file mode 100644
index 0000000..460b615
Binary files /dev/null and b/docs/images/compaction.png differ
diff --git a/docs/images/data-flow.png b/docs/images/data-flow.png
new file mode 100644
index 0000000..34cea77
Binary files /dev/null and b/docs/images/data-flow.png differ
diff --git a/docs/images/interaction-flow.png b/docs/images/interaction-flow.png
new file mode 100644
index 0000000..e8503e4
Binary files /dev/null and b/docs/images/interaction-flow.png differ
diff --git a/docs/images/mcp-architecture.png b/docs/images/mcp-architecture.png
new file mode 100644
index 0000000..d1c72cc
Binary files /dev/null and b/docs/images/mcp-architecture.png differ
diff --git a/docs/images/permission-layers.png b/docs/images/permission-layers.png
new file mode 100644
index 0000000..dc3c4e7
Binary files /dev/null and b/docs/images/permission-layers.png differ
diff --git a/docs/images/streaming-timeline.png b/docs/images/streaming-timeline.png
new file mode 100644
index 0000000..fab8f16
Binary files /dev/null and b/docs/images/streaming-timeline.png differ
diff --git a/docs/images/system-prompt-assembly.png b/docs/images/system-prompt-assembly.png
new file mode 100644
index 0000000..b60dd0f
Binary files /dev/null and b/docs/images/system-prompt-assembly.png differ
diff --git a/docs/introduction/architecture-overview.mdx b/docs/introduction/architecture-overview.mdx
index a2e34a4..e6ed31c 100644
--- a/docs/introduction/architecture-overview.mdx
+++ b/docs/introduction/architecture-overview.mdx
@@ -9,6 +9,10 @@ description: "五层架构,一条数据流"
Claude Code 从上到下分为五个层次,每一层职责清晰、边界分明:
+
+
+
+
| 层次 | 职责 | 关键词 |
|------|------|--------|
| **交互层** | 终端 UI、用户输入、消息展示 | React/Ink、REPL |
@@ -19,7 +23,9 @@ Claude Code 从上到下分为五个层次,每一层职责清晰、边界分
## 一条主数据流
-{/* TODO: 插入数据流序列图 */}
+
+
+
整个系统的运转可以浓缩为一条核心数据流:
diff --git a/docs/introduction/what-is-claude-code.mdx b/docs/introduction/what-is-claude-code.mdx
index e89c464..ea70a56 100644
--- a/docs/introduction/what-is-claude-code.mdx
+++ b/docs/introduction/what-is-claude-code.mdx
@@ -27,7 +27,9 @@ Claude Code 是一个运行在命令行终端里的 AI 编程助手。你用自
## 一次典型的交互流程
-{/* TODO: 插入一张交互流程示意图 */}
+
+
+
1. 你在终端输入自然语言需求
2. Claude Code 分析你的项目上下文
diff --git a/docs/safety/permission-model.mdx b/docs/safety/permission-model.mdx
index 7308a60..32c2ece 100644
--- a/docs/safety/permission-model.mdx
+++ b/docs/safety/permission-model.mdx
@@ -19,6 +19,10 @@ description: "三种行为 x 多级来源 = 灵活而严谨的权限体系"
规则可以来自多个来源,优先级从高到低:
+
+
+
+
用户在当前对话中手动授权的("对这个工具始终允许")