diff --git a/crates/workflow-context/Cargo.toml b/crates/workflow-context/Cargo.toml
index c70238f..dd39572 100644
--- a/crates/workflow-context/Cargo.toml
+++ b/crates/workflow-context/Cargo.toml
@@ -13,6 +13,7 @@ thiserror = "1.0"
 hex = "0.4"
 walkdir = "2.3"
 regex = "1.10"
+globset = "0.4.18"
 
 [dev-dependencies]
 tempfile = "3.8"
diff --git a/crates/workflow-context/src/lib.rs b/crates/workflow-context/src/lib.rs
index f41ea89..4e209cb 100644
--- a/crates/workflow-context/src/lib.rs
+++ b/crates/workflow-context/src/lib.rs
@@ -2,8 +2,10 @@ pub mod types;
 pub mod traits;
 pub mod vgcs;
 pub mod docos;
+pub mod worker_runtime;
 
 pub use types::*;
 pub use traits::*;
 pub use vgcs::Vgcs;
 pub use docos::{DocOS, DocManager};
+pub use worker_runtime::{WorkerContext, ContextShell, OutputFormat, FindOptions, NodeMetadata, GrepMatch, FileStats};
diff --git a/crates/workflow-context/src/worker_runtime.rs b/crates/workflow-context/src/worker_runtime.rs
new file mode 100644
index 0000000..9420413
--- /dev/null
+++ b/crates/workflow-context/src/worker_runtime.rs
@@ -0,0 +1,378 @@
+use std::path::Path;
+use std::sync::Arc;
+use std::env;
+use anyhow::{Result, Context, anyhow};
+use serde::{Serialize, Deserialize};
+use serde::de::DeserializeOwned;
+use globset::Glob;
+use regex::Regex;
+use crate::{DocOS, DocManager, Vgcs, DocNodeKind};
+
+// --- Data Structures ---
+
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub enum OutputFormat {
+    Text,
+    Json,
+}
+
+#[derive(Debug, Serialize, Deserialize, Clone, PartialEq)]
+pub struct NodeMetadata {
+    pub path: String,
+    pub kind: String, // "File" or "Dir"
+    pub size: u64,
+    // pub modified: bool, // TODO: Implement diff check against base
+}
+
+#[derive(Debug, Default, Clone)]
+pub struct FindOptions {
+    pub recursive: bool,
+    pub max_depth: Option<usize>,
+    pub type_filter: Option<String>, // "File" or "Dir"
+    pub min_size: Option<u64>,
+    pub max_size: Option<u64>,
+}
+
+#[derive(Debug, Serialize, Deserialize)]
+pub struct GrepMatch {
+    pub path: String,
+    pub line_number: usize,
+    pub content: String,
+}
+
+#[derive(Debug, Serialize, Deserialize)]
+pub struct FileStats {
+    pub path: String,
+    pub lines: usize,
+    pub bytes: usize,
+}
+
+// --- Trait Definition ---
+
+pub trait ContextShell {
+    fn tree(&self, path: &str, depth: Option<usize>, format: OutputFormat) -> Result<String>;
+    fn find(&self, name_pattern: &str, options: FindOptions) -> Result<Vec<NodeMetadata>>;
+    fn grep(&self, pattern: &str, paths: Option<Vec<String>>) -> Result<Vec<GrepMatch>>;
+    fn cat(&self, paths: &[String]) -> Result<String>;
+    fn wc(&self, paths: &[String]) -> Result<Vec<FileStats>>;
+    fn patch(&mut self, path: &str, original: &str, replacement: &str) -> Result<()>;
+}
+
+// --- WorkerContext Implementation ---
+
+pub struct WorkerContext {
+    doc: DocOS<Vgcs>,
+}
+
+impl WorkerContext {
+    pub fn from_env() -> Result<Self> {
+        let req_id = env::var("WORKFLOW_REQ_ID").context("Missing WORKFLOW_REQ_ID")?;
+        let commit = env::var("WORKFLOW_BASE_COMMIT").context("Missing WORKFLOW_BASE_COMMIT")?;
+        let data_path = env::var("WORKFLOW_DATA_PATH").context("Missing WORKFLOW_DATA_PATH")?;
+        
+        let vgcs = Vgcs::new(&data_path);
+        let doc = DocOS::new(Arc::new(vgcs), &req_id, &commit);
+        
+        Ok(Self { doc })
+    }
+
+    pub fn new(data_path: &str, req_id: &str, commit: &str) -> Self {
+        let vgcs = Vgcs::new(data_path);
+        let doc = DocOS::new(Arc::new(vgcs), req_id, commit);
+        Self { doc }
+    }
+
+    pub fn read_json<T: DeserializeOwned>(&self, path: impl AsRef<Path>) -> Result<T> {
+        let path_str = path.as_ref().to_string_lossy();
+        let content = self.doc.read_content(&path_str)?;
+        let data = serde_json::from_str(&content)
+            .with_context(|| format!("Failed to parse JSON from {}", path_str))?;
+        Ok(data)
+    }
+
+    pub fn read_text(&self, path: impl AsRef<Path>) -> Result<String> {
+        let path_str = path.as_ref().to_string_lossy();
+        self.doc.read_content(&path_str)
+    }
+
+    pub fn write_file(&mut self, path: impl AsRef<Path>, content: &str) -> Result<()> {
+         let path_str = path.as_ref().to_string_lossy();
+         self.doc.write_content(&path_str, content)
+    }
+    
+    pub fn attach_subsection(&mut self, parent: impl AsRef<Path>, name: &str, content: &str) -> Result<()> {
+        let parent_str = parent.as_ref().to_string_lossy();
+        self.doc.insert_subsection(&parent_str, name, content)
+    }
+
+    pub fn commit(&mut self, message: &str) -> Result<String> {
+        self.doc.save(message)
+    }
+
+    pub fn get_tool_definitions() -> serde_json::Value {
+        serde_json::json!([
+            {
+                "type": "function",
+                "function": {
+                    "name": "tree",
+                    "description": "List directory structure to understand the file layout.",
+                    "parameters": {
+                        "type": "object",
+                        "properties": {
+                            "path": { "type": "string", "description": "Root path to list (default: root)" },
+                            "depth": { "type": "integer", "description": "Recursion depth" },
+                            "format": { "type": "string", "enum": ["Text", "Json"], "default": "Text" }
+                        }
+                    }
+                }
+            },
+            {
+                "type": "function",
+                "function": {
+                    "name": "find",
+                    "description": "Find files by name pattern (glob). Fast metadata search.",
+                    "parameters": {
+                        "type": "object",
+                        "required": ["pattern"],
+                        "properties": {
+                            "pattern": { "type": "string", "description": "Glob pattern (e.g. **/*.rs)" },
+                            "recursive": { "type": "boolean", "default": true }
+                        }
+                    }
+                }
+            },
+             {
+                "type": "function",
+                "function": {
+                    "name": "grep",
+                    "description": "Search for content within files using regex.",
+                    "parameters": {
+                        "type": "object",
+                        "required": ["pattern"],
+                        "properties": {
+                            "pattern": { "type": "string", "description": "Regex pattern" },
+                            "paths": { "type": "array", "items": { "type": "string" }, "description": "Limit search to these paths" }
+                        }
+                    }
+                }
+            },
+            {
+                "type": "function",
+                "function": {
+                    "name": "cat",
+                    "description": "Read and assemble content of multiple files.",
+                    "parameters": {
+                        "type": "object",
+                        "required": ["paths"],
+                        "properties": {
+                            "paths": { "type": "array", "items": { "type": "string" } }
+                        }
+                    }
+                }
+            },
+            {
+                "type": "function",
+                "function": {
+                    "name": "patch",
+                    "description": "Replace a specific text block in a file. Use this for small corrections.",
+                    "parameters": {
+                        "type": "object",
+                        "required": ["path", "original", "replacement"],
+                        "properties": {
+                            "path": { "type": "string" },
+                            "original": { "type": "string", "description": "Exact text to look for. Must be unique in file." },
+                            "replacement": { "type": "string", "description": "New text to insert." }
+                        }
+                    }
+                }
+            }
+        ])
+    }
+}
+
+impl ContextShell for WorkerContext {
+    fn tree(&self, path: &str, depth: Option<usize>, format: OutputFormat) -> Result<String> {
+        let root_node = self.doc.get_outline()?; 
+        
+        let target_node = if path == "/" || path == "." {
+             Some(&root_node)
+        } else {
+             fn find_node<'a>(node: &'a crate::DocNode, path: &str) -> Option<&'a crate::DocNode> {
+                 if node.path == path {
+                     return Some(node);
+                 }
+                 for child in &node.children {
+                     if let Some(found) = find_node(child, path) {
+                         return Some(found);
+                     }
+                 }
+                 None
+             }
+             find_node(&root_node, path)
+        };
+        
+        let node = target_node.ok_or_else(|| anyhow!("Path not found: {}", path))?;
+
+        match format {
+            OutputFormat::Json => {
+                Ok(serde_json::to_string_pretty(node)?)
+            },
+            OutputFormat::Text => {
+                let mut output = String::new();
+                fn print_tree(node: &crate::DocNode, prefix: &str, is_last: bool, depth: usize, max_depth: Option<usize>, output: &mut String) {
+                     if let Some(max) = max_depth {
+                         if depth > max { return; }
+                     }
+                     
+                     let name = if node.path == "/" { "." } else { &node.name };
+                     
+                     if depth > 0 {
+                        let connector = if is_last { "└── " } else { "├── " };
+                        output.push_str(&format!("{}{}{}\n", prefix, connector, name));
+                     } else {
+                        output.push_str(&format!("{}\n", name));
+                     }
+                     
+                     let child_prefix = if depth > 0 {
+                         if is_last { format!("{}    ", prefix) } else { format!("{}│   ", prefix) }
+                     } else {
+                         "".to_string()
+                     };
+
+                     for (i, child) in node.children.iter().enumerate() {
+                         print_tree(child, &child_prefix, i == node.children.len() - 1, depth + 1, max_depth, output);
+                     }
+                }
+                print_tree(node, "", true, 0, depth, &mut output);
+                Ok(output)
+            }
+        }
+    }
+
+    fn find(&self, name_pattern: &str, options: FindOptions) -> Result<Vec<NodeMetadata>> {
+        let root = self.doc.get_outline()?;
+        let mut results = Vec::new();
+        
+        let glob = Glob::new(name_pattern)?.compile_matcher();
+        
+        fn traverse(node: &crate::DocNode, glob: &globset::GlobMatcher, opts: &FindOptions, depth: usize, results: &mut Vec<NodeMetadata>) {
+            if let Some(max) = opts.max_depth {
+                if depth > max { return; }
+            }
+            
+            let match_name = glob.is_match(&node.name) || glob.is_match(&node.path);
+            
+            let kind_str = match node.kind {
+                DocNodeKind::Composite => "Dir",
+                DocNodeKind::Leaf => "File",
+                DocNodeKind::Section => "Section",
+            };
+
+            let type_match = match &opts.type_filter {
+                Some(t) => t.eq_ignore_ascii_case(kind_str),
+                None => true,
+            };
+            
+            if depth > 0 && match_name && type_match {
+                 results.push(NodeMetadata {
+                     path: node.path.clone(),
+                     kind: kind_str.to_string(),
+                     size: 0, 
+                 });
+            }
+            
+            if opts.recursive || depth == 0 {
+                 for child in &node.children {
+                     traverse(child, glob, opts, depth + 1, results);
+                 }
+            }
+        }
+        
+        traverse(&root, &glob, &options, 0, &mut results);
+        Ok(results)
+    }
+
+    fn grep(&self, pattern: &str, paths: Option<Vec<String>>) -> Result<Vec<GrepMatch>> {
+        let re = Regex::new(pattern).context("Invalid regex pattern")?;
+        
+        let target_paths = match paths {
+            Some(p) => p,
+            None => {
+                let all_nodes = self.find("**/*", FindOptions { 
+                    recursive: true, 
+                    type_filter: Some("File".to_string()), 
+                    ..Default::default() 
+                })?;
+                all_nodes.into_iter().map(|n| n.path).collect()
+            }
+        };
+
+        let mut matches = Vec::new();
+
+        for path in target_paths {
+            if let Ok(content) = self.read_text(&path) {
+                for (i, line) in content.lines().enumerate() {
+                    if re.is_match(line) {
+                        matches.push(GrepMatch {
+                            path: path.clone(),
+                            line_number: i + 1,
+                            content: line.trim().to_string(),
+                        });
+                    }
+                }
+            }
+        }
+        
+        Ok(matches)
+    }
+
+    fn cat(&self, paths: &[String]) -> Result<String> {
+        let mut output = String::new();
+        for path in paths {
+            match self.read_text(path) {
+                Ok(content) => {
+                    output.push_str(&format!("<file path=\"{}\">\n", path));
+                    output.push_str(&content);
+                    if !content.ends_with('\n') {
+                        output.push('\n');
+                    }
+                    output.push_str("</file>\n\n");
+                },
+                Err(e) => {
+                    output.push_str(&format!("<!-- Failed to read {}: {} -->\n", path, e));
+                }
+            }
+        }
+        Ok(output)
+    }
+
+    fn wc(&self, paths: &[String]) -> Result<Vec<FileStats>> {
+        let mut stats = Vec::new();
+        for path in paths {
+            if let Ok(content) = self.read_text(path) {
+                stats.push(FileStats {
+                    path: path.clone(),
+                    lines: content.lines().count(),
+                    bytes: content.len(),
+                });
+            }
+        }
+        Ok(stats)
+    }
+
+    fn patch(&mut self, path: &str, original: &str, replacement: &str) -> Result<()> {
+        let content = self.read_text(path)?;
+        
+        let matches: Vec<_> = content.match_indices(original).collect();
+        
+        match matches.len() {
+            0 => return Err(anyhow!("Original text not found in {}", path)),
+            1 => {
+                let new_content = content.replace(original, replacement);
+                self.write_file(path, &new_content)?;
+                Ok(())
+            },
+            _ => return Err(anyhow!("Ambiguous match: original text found {} times", matches.len())),
+        }
+    }
+}
diff --git a/crates/workflow-context/tests/worker_runtime_tests.rs b/crates/workflow-context/tests/worker_runtime_tests.rs
new file mode 100644
index 0000000..b699d5a
--- /dev/null
+++ b/crates/workflow-context/tests/worker_runtime_tests.rs
@@ -0,0 +1,142 @@
+use workflow_context::{WorkerContext, ContextShell, OutputFormat, FindOptions, Vgcs, ContextStore};
+use tempfile::TempDir;
+
+const ZERO_OID: &str = "0000000000000000000000000000000000000000";
+
+fn setup_env() -> (TempDir, String, String) {
+    let temp_dir = TempDir::new().unwrap();
+    let data_path = temp_dir.path().to_str().unwrap().to_string();
+    let req_id = "req-shell-test".to_string();
+    
+    // Init Repo
+    let vgcs = Vgcs::new(&data_path);
+    vgcs.init_repo(&req_id).unwrap();
+    
+    (temp_dir, data_path, req_id)
+}
+
+#[test]
+fn test_shell_comprehensive() -> anyhow::Result<()> {
+    let (_tmp, data_path, req_id) = setup_env();
+    
+    // 1. Setup Initial Context
+    let mut ctx = WorkerContext::new(&data_path, &req_id, ZERO_OID);
+    
+    ctx.write_file("README.md", "Project Root\n\nIntroduction here.")?;
+    ctx.write_file("src/main.rs", "fn main() {\n    println!(\"Hello\");\n    println!(\"Hello\");\n}")?; // Double Hello for ambiguity test
+    ctx.write_file("src/util.rs", "pub fn util() -> i32 { 42 }")?;
+    ctx.write_file("data/config.json", "{\n  \"key\": \"value\",\n  \"retries\": 3\n}")?;
+    ctx.write_file("文档/说明.txt", "这是一个中文文件。")?; // Unicode Path & Content
+    
+    let commit_1 = ctx.commit("Init")?;
+    let mut ctx = WorkerContext::new(&data_path, &req_id, &commit_1);
+
+    // --- Find Tests ---
+    println!("Testing Find...");
+    
+    // Test: Recursive vs Non-recursive
+    // Note: Includes directories (src, data, 文档) + files (5) = 8
+    let all_nodes = ctx.find("**/*", FindOptions { recursive: true, ..Default::default() })?;
+    assert_eq!(all_nodes.len(), 8);
+    
+    // Test: Only Files
+    let only_files = ctx.find("**/*", FindOptions { 
+        recursive: true, 
+        type_filter: Some("File".to_string()),
+        ..Default::default() 
+    })?;
+    assert_eq!(only_files.len(), 5);
+
+    // Test: Non-recursive (Top level)
+    let root_nodes = ctx.find("*", FindOptions { recursive: false, ..Default::default() })?;
+    // Expect README.md, src(dir), data(dir), 文档(dir)
+    assert!(root_nodes.iter().any(|f| f.path == "README.md"));
+    assert!(root_nodes.iter().any(|f| f.path == "src"));
+    
+    // Test: Type Filter (Dir)
+    let dirs = ctx.find("**/*", FindOptions { 
+        recursive: true, 
+        type_filter: Some("Dir".to_string()), 
+        ..Default::default() 
+    })?;
+    assert!(dirs.iter().any(|d| d.path == "src"));
+    assert!(dirs.iter().any(|d| d.path == "data"));
+    assert!(dirs.iter().any(|d| d.path == "文档"));
+    assert!(!dirs.iter().any(|d| d.path == "README.md"));
+
+    // --- Grep Tests ---
+    println!("Testing Grep...");
+    
+    // Test: Regex Match
+    let matches = ctx.grep(r"fn \w+\(\)", None)?;
+    assert_eq!(matches.len(), 2); // main() and util()
+    
+    // Test: Unicode Content
+    let zh_matches = ctx.grep("中文", None)?;
+    assert_eq!(zh_matches.len(), 1);
+    assert_eq!(zh_matches[0].path, "文档/说明.txt");
+
+    // Test: Invalid Regex
+    let bad_regex = ctx.grep("(", None);
+    assert!(bad_regex.is_err());
+
+    // --- Patch Tests ---
+    println!("Testing Patch...");
+
+    // Test: Ambiguous Match (Safety Check)
+    // src/main.rs has two "println!(\"Hello\");"
+    let res = ctx.patch("src/main.rs", "println!(\"Hello\");", "println!(\"World\");");
+    assert!(res.is_err(), "Should fail on ambiguous match");
+    let err_msg = res.unwrap_err().to_string();
+    assert!(err_msg.contains("Ambiguous match"), "Error message mismatch: {}", err_msg);
+
+    // Test: Unique Match
+    // Patch "Introduction here." to "Intro v2." in README.md
+    ctx.patch("README.md", "Introduction here.", "Intro v2.")?;
+    ctx.commit("Patch 1")?; // Must commit to verify via read (if read uses committed state)
+    
+    // Verify
+    let readme = ctx.read_text("README.md")?;
+    assert!(readme.contains("Intro v2."));
+    
+    // Test: Special Characters (Literal Match)
+    // Let's try to patch JSON which has braces and quotes
+    ctx.patch("data/config.json", "\"retries\": 3", "\"retries\": 5")?;
+    ctx.commit("Patch 2")?;
+    
+    let config = ctx.read_text("data/config.json")?;
+    assert!(config.contains("\"retries\": 5"));
+    
+    // Test: Cross-line Patch
+    // Replace the whole function body in util.rs
+    let old_block = "pub fn util() -> i32 { 42 }";
+    let new_block = "pub fn util() -> i32 {\n    return 100;\n}";
+    ctx.patch("src/util.rs", old_block, new_block)?;
+    ctx.commit("Patch 3")?;
+    
+    let util = ctx.read_text("src/util.rs")?;
+    assert!(util.contains("return 100;"));
+
+    // Test: Patch non-existent file
+    let res = ctx.patch("ghost.txt", "foo", "bar");
+    assert!(res.is_err());
+
+    Ok(())
+}
+
+#[test]
+fn test_tool_schema_validity() {
+    let defs = WorkerContext::get_tool_definitions();
+    assert!(defs.is_array());
+    let arr = defs.as_array().unwrap();
+    
+    // Verify critical fields exist for OpenAI
+    for tool in arr {
+        let obj = tool.as_object().unwrap();
+        assert_eq!(obj["type"], "function");
+        let func = obj["function"].as_object().unwrap();
+        assert!(func.contains_key("name"));
+        assert!(func.contains_key("description"));
+        assert!(func.contains_key("parameters"));
+    }
+}
diff --git a/docs/3_project_management/tasks/pending/20251126_design_3_worker_runtime.md b/docs/3_project_management/tasks/pending/20251126_design_3_worker_runtime.md
new file mode 100644
index 0000000..f29b23c
--- /dev/null
+++ b/docs/3_project_management/tasks/pending/20251126_design_3_worker_runtime.md
@@ -0,0 +1,181 @@
+# 设计方案 3: Worker Runtime (Context Shell & Utilities)
+
+## 1. 定位与目标
+
+Worker Runtime 是连接底层文档系统 (DocOS) 与上层业务逻辑 (LLM Worker) 的桥梁。
+如果说 VGCS 是硬盘，DocOS 是文件系统，那么 Worker Runtime 就是 **Shell (命令行工具集)**。
+
+它的核心任务是：**高效地检索、过滤、组装上下文，为大模型准备输入 (Prompt Context)。**
+
+## 2. 核心设计：Context Shell
+
+我们借鉴 Linux Coreutils 及现代 Rust CLI 工具（如 `fd`, `ripgrep`, `exa`）的理念，提供一组高效、强类型、支持结构化输出的原语。
+
+### 2.1 接口定义 (Rust Trait)
+
+```rust
+pub enum OutputFormat {
+    Text, // 人类/LLM 可读的文本 (如 ASCII Tree)
+    Json, // 程序可读的结构化数据 (类似 jq 输入)
+}
+
+pub trait ContextShell {
+    /// [tree]: 生成目录树视图
+    /// depth: 递归深度
+    /// format: Text (ASCII Tree) | Json (Nested Objects)
+    fn tree(&self, path: &str, depth: Option<usize>, format: OutputFormat) -> Result<String>;
+
+    /// [find]: 基于元数据的快速查找 (不读取内容)
+    /// name_pattern: Glob (如 "*.rs")
+    /// min_size/max_size: 大小过滤
+    fn find(&self, name_pattern: &str, options: FindOptions) -> Result<Vec<NodeMetadata>>;
+
+    /// [grep]: 全文检索 (读取内容)
+    /// pattern: Regex
+    /// paths: 限制搜索的文件列表 (通常由 find 的结果输入)
+    fn grep(&self, pattern: &str, paths: Option<Vec<String>>) -> Result<Vec<GrepMatch>>;
+
+    /// [cat]: 读取并组装内容
+    /// 自动处理 Blob 下载，拼接多个文件，添加 header
+    fn cat(&self, paths: &[&str]) -> Result<String>;
+
+    /// [wc]: 统计元数据 (行数, 字节数)
+    fn wc(&self, paths: &[&str]) -> Result<Vec<FileStats>>;
+
+    /// [patch]: 局部修补 (新增)
+    /// 基于精确文本匹配的替换，避免全量读写。
+    /// original: 必须在文件中唯一存在的原文片段
+    /// replacement: 替换后的新文本
+    fn patch(&self, path: &str, original: &str, replacement: &str) -> Result<()>;
+}
+```
+
+### 2.2 数据结构 (Serializable)
+
+```rust
+#[derive(Serialize, Deserialize)]
+pub struct NodeMetadata {
+    pub path: String,
+    pub kind: NodeKind, // File | Dir
+    pub size: u64,
+    // pub modified: bool, // TODO: Implement diff check against base
+}
+
+#[derive(Serialize, Deserialize)]
+pub struct FindOptions {
+    pub recursive: bool,
+    pub max_depth: Option<usize>,
+    pub type_filter: Option<String>, // "File" or "Dir"
+    pub min_size: Option<u64>,
+    pub max_size: Option<u64>,
+}
+
+#[derive(Debug, Serialize, Deserialize)]
+pub struct GrepMatch {
+    pub path: String,
+    pub line_number: usize,
+    pub content: String,
+}
+
+#[derive(Debug, Serialize, Deserialize)]
+pub struct FileStats {
+    pub path: String,
+    pub lines: usize,
+    pub bytes: usize,
+}
+```
+
+## 3. 详细功能设计
+
+### 3.1 tree (Structure Awareness)
+*   **场景**: LLM 需要先看一眼 "Map" 才能知道去哪找宝藏。
+*   **Text Mode**: 生成经典的 ASCII Tree，直接作为 Prompt 的一部分。
+*   **Json Mode**: 嵌套的 JSON 对象，供代码逻辑分析结构。
+
+### 3.2 find (Fast Filter)
+*   **场景**: 效率优化。不要一开始就 `grep` 全文。
+*   **原理**: 只遍历 Git Tree 对象（元数据），**不解压 Blob**。速度极快。
+
+### 3.3 grep (Content Search)
+*   **场景**: RAG (Retrieval) 环节。
+*   **优化**: 接受 `find` 的输出作为 `paths` 参数，避免全盘扫描。
+*   **并行**: 利用 Rayon 并行读取 Blob 并匹配。
+
+### 3.4 cat (Assemble)
+*   **场景**: 组装 Prompt Context。
+*   **格式**: 使用 XML tags 包裹，明确文件边界。
+
+### 3.5 patch (Atomic Update)
+*   **场景**: 修正笔误或更新局部数据。
+*   **逻辑**:
+    1. 读取文件内容。
+    2. 查找 `original` 字符串。
+    3. **Fail Fast**: 如果找不到，或者找到多处（歧义），直接报错，防止改错位置。
+    4. 执行替换并在内存中生成新 Blob，更新 Index。
+*   **优势**: 相比 `write_file`，它不需要 Worker 回传整个文件内容，节省网络传输和 Token。
+
+## 4. Tool Definition Schema (AI Configuration)
+
+为了让 LLM 能够直接使用这些工具，Runtime 将提供一个方法导出标准的 JSON Schema (OpenAI Compatible)。
+
+```rust
+impl WorkerContext {
+    pub fn get_tool_definitions() -> serde_json::Value {
+        serde_json::json!([
+            {
+                "type": "function",
+                "function": {
+                    "name": "tree",
+                    "description": "List directory structure. Use this first to understand the file layout.",
+                    "parameters": {
+                        "type": "object",
+                        "properties": {
+                            "path": { "type": "string", "description": "Root path to list (default: root)" },
+                            "depth": { "type": "integer", "description": "Recursion depth" }
+                        }
+                    }
+                }
+            },
+            {
+                "type": "function",
+                "function": {
+                    "name": "patch",
+                    "description": "Replace a specific text block in a file. Use this for small corrections.",
+                    "parameters": {
+                        "type": "object",
+                        "required": ["path", "original", "replacement"],
+                        "properties": {
+                            "path": { "type": "string" },
+                            "original": { "type": "string", "description": "Exact text to look for. Must be unique in file." },
+                            "replacement": { "type": "string", "description": "New text to insert." }
+                        }
+                    }
+                }
+            }
+            // ... defined for grep, cat, find, etc.
+        ])
+    }
+}
+```
+
+## 5. 高效工作流示例 (The Funnel Pattern)
+
+采用了类似 "漏斗" 的筛选机制，层层递进，效率最高。
+
+```rust
+// 1. 全局感知 (Tree)
+let structure = ctx.tree("/", Some(2), OutputFormat::Text)?;
+
+// 2. 快速定位 (Find)
+let json_files = ctx.find("**/*.json", FindOptions::default())?;
+
+// 3. 精确检索 (Grep)
+let matches = ctx.grep("NetProfit", Some(json_files.map(|f| f.path)))?;
+
+// 4. 局部修正 (Patch)
+// 假设我们在 matches 中发现了一个拼写错误 "NetProft"
+ctx.patch("data/financials.json", "\"NetProft\":", "\"NetProfit\":")?;
+
+// 5. 组装阅读 (Cat)
+let context = ctx.cat(&distinct_files)?;
+```