code-reader-v2-en

分类: 开发与编程 | 上传者: notlate-cn | 下载: 0 | 版本: v1.0（最新）
Cognitive science-based source code deep understanding assistant (English improved version). Supports 3 analysis modes: Quick, Standard, Deep (auto-parallel for large projects). Combines elaborative interrogation, self-explanation testing, and retrieval practice to help truly understand and master code.
更新日志: Source: GitHub https://github.com/notlate-cn/code-reader-skills
目录结构

当前层级: code-reader-en/
📄 code-reader-en.skill 35.6 KB
📄 SKILL.md 72.4 KB
SKILL.md

---
name: code-reader-v2-en
description: Cognitive science-based source code deep understanding assistant (English improved version). Supports 3 analysis modes: Quick, Standard, Deep (auto-parallel for large projects). Combines elaborative interrogation, self-explanation testing, and retrieval practice to help truly understand and master code.
---

# Source Code Deep Understanding Analyzer v2.3

Professional code analysis tool based on cognitive science research, supporting three analysis depths to ensure true understanding rather than fluency illusion.

## Three Analysis Modes

| User Intent | Recommended Mode | Trigger Examples | Duration |
|-------------|-----------------|------------------|----------|
| Quick browse/code review | Quick Mode | "quick look", "what does this do", "briefly scan" | 5-10 min |
| Learning/technical research | Standard Mode ⭐ | "analyze", "help me understand", "explain" | 15-20 min |
| Deep mastery/large projects | Deep Mode 🚀 | "thoroughly analyze", "completely master", "in-depth research", "interview prep", "project analysis" | 30+ min |

**Default: Standard Mode, system auto-selects optimal mode based on code scale and user intent.**

**🚀 Deep Mode internal smart strategy:**
- Code ≤ 2000 lines: Progressive generation (sequential chapter filling)
- Code > 2000 lines: Auto-enable parallel processing (sub-agents analyze chapters in parallel)
- Code > 10000 lines / files > 20: Layered parallel (module-level scan first, then chapter-level parallel)

---

## Core Philosophy: Understanding First, Memorization Second

**Combat Fluency Illusion**

> "Reading code ≠ Writing code"  
> "Understanding explanations ≠ Independent implementation"  
> "Feeling like you get it ≠ Actually understanding it"

**Core Principles:**
- Understand WHY, not just WHAT
- Force self-explanation to verify true understanding
- Build concept connections, not isolated memories
- Test transfer ability through application variants
- **Write like a friend explaining, not a textbook stacking jargon**

**Research Foundation:**
- [Dunlosky et al.](https://www.aft.org/ae/fall2013/dunlosky) - Elaborative interrogation significantly outperforms passive reading
- [Chi et al.](https://onlinelibrary.wiley.com/doi/10.1207/s15516709cog1803_3) - Self-explainers achieve correct mental models with higher probability
- [Karpicke & Roediger](https://science.sciencemag.org/content/319/5865/966) - Retrieval practice outperforms re-reading by 250%

---

## Pre-Analysis Mandatory Check: Understanding Verification Gate

**Based on selected mode, execute corresponding verification:**

### Quick Mode - Simplified Check
- Quickly identify code type and core functionality
- List key concepts (no deep verification)

### Standard Mode - Standard Verification
- Self-explanation test for core concepts
- Verify ability to articulate WHY

### Deep Mode - Complete Verification
- Full self-explanation test
- Application transfer ability verification

**Output Format (at beginning of analysis document):**

```markdown
## Understanding Verification Status [Standard/Deep Mode Only]

| Core Concept | Self-Explanation | Understanding WHY | Application Transfer | Status |
|--------------|------------------|-------------------|---------------------|---------|
| User Authentication Flow | ✅ | ✅ | ✅ | Understood |
| JWT Token Mechanism | ✅ | ⚠️ | ❌ | ⚠️ Needs Deeper Understanding |
| Password Hashing | ✅ | ✅ | ⚠️ | Basic Understanding |
```

---

## Three Mode Output Structures

### Quick Mode Output Structure (5-10 min)

```markdown
# [Code Name] Quick Analysis

## 1. Quick Overview
- Programming language and version
- Code scale and type
- Core dependencies

## 2. Functionality Description
- What it does (WHAT)
- Brief explanation of WHY needed

## 3. Core Algorithm/Design
- Algorithm complexity (if applicable)
- Design patterns used (if applicable)
- WHY this algorithm/pattern

## 4. Key Code Segments
- 3-5 core code segments
- Brief purpose description for each

## 5. Dependencies
- External library list and purposes

## 6. Quick Usage Example
- Simple runnable example
```

### Standard Mode Output Structure (15-20 min) ⭐ Recommended

```markdown
# [Code Name] Deep Understanding Analysis

## Understanding Verification Status
[Self-explanation test results table]

## 1. Quick Overview
- Language, scale, dependencies

## 2. Background & Motivation (Elaborative Interrogation)
- WHY this code is needed
- WHY this approach
- WHY not other alternatives

## 3. Core Concepts
- List key concepts
- Each concept answers 2-3 WHYs

## 4. Algorithm & Theory
- Complexity analysis
- WHY this algorithm
- References

## 5. Design Patterns
- Patterns identified
- WHY used

## 6. Key Code Deep Analysis
- Line-by-line WHY analysis
- Execution flow examples

## 7. Test Case Analysis (if tests available)
- Test coverage analysis
- Boundary conditions from tests
- Hidden behaviors discovered

## 8. Dependencies & Usage Examples
- Detailed WHY comments
```

### Deep Mode Output Structure (30+ min)

**Deep Mode auto-selects optimal strategy based on code scale, ensuring sufficient depth for each chapter:**

#### Strategy A: Progressive Generation (Code ≤ 2000 lines)

**For medium-small code, generate chapters sequentially:**

```markdown
# [Code Name] Complete Mastery Analysis

[All Standard Mode content, plus:]

## 3+. Concept Network Diagram
- Core concept inventory (3 WHYs each)
- Concept relationship matrix
- Connections to existing knowledge

## 6+. Complete Execution Examples
- Multi-scenario execution flows
- Boundary conditions
- Error-prone points

## 7. Test Case Analysis (if tests available)
- Test coverage analysis
- Key test case interpretations
- Hidden behaviors discovered from tests

## 8. Application Transfer Scenarios (at least 2)
- Scenario 1: Constant principles + modifications + WHY
- Scenario 2: Constant principles + modifications + WHY
- Extract universal patterns

## 10. Quality Verification Checklist
- Understanding depth verification
- Technical accuracy verification
- Practicality verification
- Final "four abilities" test
```

#### Strategy B: Parallel Processing (Code > 2000 lines) 🚀

**For large projects, using sub-agent parallel architecture:**

#### Core Architecture

```
┌─────────────────────────────────────────────────────────────┐
│                    Master Coordinator Agent                  │
│  - Generate analysis outline and directory framework        │
│  - Identify core concepts list (shared with sub-agents)     │
│  - Assign chapter tasks                                     │
│  - Aggregate sub-agent results                              │
│  - Final quality verification                               │
└─────────────────────────────────────────────────────────────┘
                              │
            ┌─────────────────┼─────────────────┐
            ▼                 ▼                 ▼
    ┌─────────────┐   ┌─────────────┐   ┌─────────────┐
    │ Sub-Agent 1 │   │ Sub-Agent 2 │   │ Sub-Agent 3 │
    │ Background  │   │ Core        │   │ Algorithm   │
    │ & Motivation│   │ Concepts    │   │ & Theory    │
    └─────────────┘   └─────────────┘   └─────────────┘
            │                 │                 │
            └─────────────────┼─────────────────┘
                              ▼
    ┌─────────────┐   ┌─────────────┐   ┌─────────────┐
    │ Sub-Agent 4 │   │ Sub-Agent 5 │   │ Sub-Agent 6 │
    │ Design      │   │ Code        │   │ Application │
    │ Patterns    │   │ Analysis    │   │ Transfer    │
    └─────────────┘   └─────────────┘   └─────────────┘
```

#### Parallel Execution Flow

| Phase | Executor | Operation | Output |
|-------|----------|-----------|--------|
| **1. Project Map** | Master Agent | **Enumerate all files**, build complete directory tree and module inventory | `project-map.md` |
| **2. Framework Prep** | Master Agent | Based on project map, generate outline and core concepts | `framework.json` |
| **3. Task Dispatch** | Master Agent | Create independent task descriptions for each chapter, **with explicit file path lists** | Task list |
| **4. Parallel Processing** | Sub-Agents | Each sub-agent reads assigned files, generates depth analysis | `chapter-N.md` |
| **5. Coverage Check** | Master Agent | **Compare against project map, identify uncovered files/modules** | Coverage report |
| **6. Result Aggregation** | Master Agent | Merge all chapters, unify format | `complete-analysis.md` |
| **7. Quality Verification** | Master Agent | Check depth standards, supplement weak sections | Final document |

#### Chapter Task Definition (Template for Sub-Agents)

```markdown
# Sub-Agent Task: [Chapter Name]

## Context Information
- **Code Name:** [Project/Code Name]
- **Programming Language:** [Language]
- **Code Scale:** [Total Lines] / [File Count]
- **Core Concepts:** [Concept list passed from master agent]
- **Your Assigned Files:** [Explicit list of file paths, e.g.: src/auth.py, src/utils/crypto.py]
- **Other Module Summary:** [Brief description of other modules' responsibilities, to avoid duplication]

## Your Task
You are the analysis expert specializing in "**[Chapter Name]**". **You MUST use the Read tool to read each assigned file above** before analyzing — base all content on actual code, not memory or assumptions.

## Mandatory Read Steps
1. Use Read tool to read every file listed in "Your Assigned Files"
2. Only begin analysis after confirming file contents
3. If a file doesn't exist, explain why and analyze the nearest alternative

## Output Requirements
1. **Content Depth:** This chapter must be at least [X] words
2. **WHY Analysis:** Each key point must answer 3 WHYs
3. **Code Comments:** Use Scenario/Step + WHY style
4. **Source References:** Provide authoritative reference links
5. **Independence:** Generate complete independent chapter content, no need to reference other chapters
6. **Coverage Completeness:** Every public function/class in your assigned files must be mentioned

## Output Format
Output Markdown format chapter content directly, starting with `## [Chapter Name]`.

## Depth Standards
- [ ] All assigned files have been read (no files skipped)
- [ ] All subsections covered (no "skipped" or "same as above")
- [ ] Each WHY has at least 2-3 sentences of explanation
- [ ] Code examples have complete comments
- [ ] Execution flows have concrete data tracking

Begin analysis:
```

#### Master Agent Aggregation Logic

```markdown
# Parallel Deep Mode Aggregation Specification

## Aggregation Steps

1. **Read All Sub-chapters**
   ```
   chapter_1_background.md
   chapter_2_concepts.md
   chapter_3_algorithm.md
   chapter_4_patterns.md
   chapter_5_code_analysis.md
   chapter_6_test_analysis.md
   chapter_7_transfer.md
   chapter_8_dependencies.md
   chapter_9_verification.md
   ```

2. **Coverage Check (Critical step to prevent information loss)**

   Compare the complete file list in `project-map.md` against what was analyzed:

   ```markdown
   ## Coverage Check Report

   ### File Coverage
   | File Path | Analyzed | Chapter | Notes |
   |-----------|----------|---------|-------|
   | src/auth.py | ✅ | chapter_5 | Core auth logic |
   | src/utils/crypto.py | ✅ | chapter_5 | Crypto utilities |
   | src/models/user.py | ❌ | - | Not covered, needs supplement |
   | tests/test_auth.py | ✅ | chapter_6 | Test analysis |

   ### Module Coverage Rate
   - Core modules: X/Y covered
   - Utility modules: X/Y covered
   - Test files: X/Y covered

   ### Uncovered Content Handling
   [List all uncovered files, supplement brief analysis or explain why skipped]
   ```

   **Handling uncovered content:**
   - Important files (core business logic): Supplement analysis immediately
   - Secondary files (config, utils): Briefly mention in dependencies chapter
   - Test files: Confirm already covered in test case analysis chapter

3. **Merge Order**
   ```markdown
   # [Code Name] Complete Mastery Analysis (Parallel Deep Edition)

   ## Understanding Verification Status
   [Generated from master agent's preliminary analysis]

   ## Coverage Summary
   [Total project files, analyzed file count, coverage percentage]

   [Insert each chapter content in order]
   ```

4. **Cross-Check**
   - Core concepts have consistent definitions across chapters
   - WHY explanations have no contradictions
   - Referenced code examples are consistent

4. **Depth Verification**
   - Each chapter meets word count requirements
   - WHY analysis is thorough
   - Execution examples are complete
```

#### Implementation Pseudocode

```
Function: ParallelDeepMode(projectPath, workDirectory):

  // ========== Phase 0: Build Complete Project Map (Key step to prevent info loss) ==========
  projectMap = {
    "all_files": RecursiveEnumerate(projectPath),   // Absolute path list of ALL source files
    "directory_tree": GenerateTree(projectPath),
    "file_stats": {
      "total_files": len(all_files),
      "total_lines": CountAllLines(all_files),
      "by_language": ClassifyByLanguage(all_files),
      "by_directory": ClassifyByDirectory(all_files)
    },
    "entry_files": IdentifyEntryFiles(all_files),   // main, index, __init__, etc.
    "core_modules": IdentifyCoreModules(all_files),
    "test_files": FilterTestFiles(all_files),
    "config_files": FilterConfigFiles(all_files)
  }

  WriteFile(f"{workDirectory}/project-map.md", FormatProjectMap(projectMap))

  // ========== Phase 1: Framework Preparation ==========
  // Note: Based on complete project map — never analyze just partial files
  framework = {
    "code_name": ExtractName(projectPath),
    "language": IdentifyLanguage(projectMap),
    "code_scale": projectMap["file_stats"]["total_lines"],
    "core_concepts": ExtractCoreConcepts(projectMap["core_modules"]),
    "module_responsibilities": GenerateOneLineSummaryForEachModule(),  // Shared with sub-agents
    "chapter_file_mapping": {  // Explicitly assign files to each chapter
      "Background & Motivation": [entry_files, README, main_config],
      "Core Concepts": [core_module_list],
      "Algorithm & Theory": [files_containing_algorithms],
      "Design Patterns": [core_module_list],
      "Key Code Analysis": [top_5_to_10_most_important_files],
      "Test Case Analysis": [test_file_list],
      "Application Transfer": [core_module_list],
      "Dependencies": [dependency_config_files + all_files_list],
      "Quality Verification": []  // Handled by master agent
    }
  }

  WriteFile(f"{workDirectory}/00-framework.json", framework)

  // ========== Phase 2: Create Sub-Tasks (with explicit file paths) ==========
  subTaskList = []

  for each chapter in framework["chapter_file_mapping"]:
    assignedFiles = framework["chapter_file_mapping"][chapter]
    taskDescription = GenerateTaskTemplate(
      chapter, framework,
      assignedFiles=assignedFiles,                    // Tell sub-agent which files to read
      moduleResponsibilities=framework["module_responsibilities"]  // Avoid duplication
    )
    taskFile = f"{workDirectory}/tasks/{chapter}-task.md"
    WriteFile(taskFile, taskDescription)
    subTaskList.append(taskFile)

  // ========== Phase 3: Execute Sub-Agents in Parallel ==========
  chapterFileList = []

  for each taskFile in subTaskList:
    subAgent = CreateAgent(
      name: f"Analyze-{chapter}",
      task: ReadFile(taskFile),
      outputFile: f"{workDirectory}/chapters/{chapter}.md"
    )
    subAgent.start(parallel=True)
    chapterFileList.append(subAgent.outputFile)

  WaitAll(chapterFileList)

  // ========== Phase 4: Coverage Check ==========
  analyzedFiles = ExtractReferencedFilePaths(chapterFileList)
  uncoveredFiles = projectMap["core_modules"] - analyzedFiles

  if len(uncoveredFiles) > 0:
    // Supplement analysis for important uncovered files
    for each file in uncoveredFiles:
      if file.IsCoreModule():
        supplement = CreateAgent(task=f"Analyze {file}, write to coverage supplement chapter")
        chapterFileList.append(supplement.outputFile)

  // ========== Phase 5: Result Aggregation ==========
  completeDoc = "# {framework['code_name']} Complete Mastery Analysis\n\n"
  completeDoc += "## Coverage Summary\n"
  completeDoc += f"- Total files: {projectMap['file_stats']['total_files']}\n"
  completeDoc += f"- Core modules covered: {len(analyzedFiles)}/{len(projectMap['core_modules'])}\n"
  completeDoc += "## Understanding Verification Status\n\n"
  completeDoc += GenerateVerificationTable(framework) + "\n\n"

  for each chapterFile in chapterFileList:
    chapterContent = ReadFile(chapterFile)
    completeDoc += chapterContent + "\n\n"

  // ========== Phase 6: Quality Verification ==========
  if not PassDepthCheck(completeDoc):
    weakChapters = IdentifyWeakParts(completeDoc)
    for each chapter in weakChapters:
      ReExecute(chapter)
      completeDoc = UpdateChapter(completeDoc, chapter)

  // ========== Final Output ==========
  finalFile = f"{workDirectory}/{framework['code_name']}-complete-mastery-analysis.md"
  WriteFile(finalFile, completeDoc)

  return finalFile
```

---

## Analysis Workflow (Research-Driven)

### Step 1: Quick Overview

**Goal:** Build overall mental model

**Must identify:**
- Programming language and version
- File/project scale
- Core dependencies
- Code type (algorithm, business logic, framework code, etc.)

**For projects with more than 5 files, additionally required:**

```markdown
## Project Complete Map (Prevent Information Loss)

### Full Directory Tree
[Use tools to enumerate all files, generate tree structure]

### File Inventory (Categorized)
| Category | File Path | Lines | Responsibility |
|----------|-----------|-------|---------------|
| Core Logic | src/auth.py | 350 | User authentication & authorization |
| Core Logic | src/db.py | 180 | Database operations wrapper |
| Utilities | src/utils/crypto.py | 90 | Encryption/hashing tools |
| Tests | tests/test_auth.py | 210 | Auth functional tests |
| Config | config/settings.py | 60 | Application configuration |

### Entry Point
- Main entry: [file path] - WHY start here

### Core Call Chain
[Entry → Module A → Module B → ...]
```

**Note: This step must be completed before analyzing any specific code. It is the foundation for preventing information loss.**

---

### Step 2: Elaborative Interrogation - Background & Motivation

**Core Questions (Must Answer):**

1. **WHY is this code needed?**
   - What real problem does it solve?
   - What happens without this code?

2. **WHY choose this technical approach?**
   - What are alternative solutions?
   - WHY not choose other solutions?
   - What are the trade-offs?

3. **WHY is it needed at this time/scenario?**
   - In which business process is it used?
   - What are the preconditions and postconditions?

**Output Format:**

```markdown
## Background & Motivation Analysis

### Problem Essence
**Problem to Solve:** [One sentence description]

**WHY it needs solving:** [Consequences if not solved]

### Solution Choice
**Chosen Solution:** [Current implementation approach]

**WHY choose this solution:**
- Advantages: [List 2-3 key advantages]
- Disadvantages: [List 1-2 known limitations]
- Trade-offs: [Explain what was traded off]

**Alternative Solutions Comparison:**
- Solution A: [Brief description] - WHY not chosen: [Reason]
- Solution B: [Brief description] - WHY not chosen: [Reason]

### Application Scenarios
**Applicable Scenarios:** [Specific scenario description]

**WHY applicable:** [Explain why this scenario is suitable]

**Inapplicable Scenarios:** [List boundary conditions]

**WHY inapplicable:** [Explain why certain scenarios are unsuitable]
```

---

### Step 3: Concept Network Construction

**Goal:** Build connections between concepts, not isolated memories

**Must include:**

1. **Core Concept Extraction**
   - Identify all key concepts (classes, functions, algorithms, data structures)
   - Each concept must answer 3 WHYs

2. **Concept Relationship Mapping**
   - Dependency: A depends on B - WHY?
   - Comparison: A vs B - WHY choose A?
   - Composition: A + B → C - WHY combine this way?

3. **Knowledge Connections**
   - Connect to known concepts
   - Connect to design patterns
   - Connect to theoretical foundations

**Output Format:**

```markdown
## Concept Network Diagram

### Core Concept Inventory

**Concept 1: User Authentication**
- **What it is:** Process of verifying user identity
- **WHY needed:** Protect system resources from unauthorized access
- **WHY implemented this way:** Use JWT for stateless authentication, reducing server load
- **WHY not other approaches:** Session requires server storage, not conducive to horizontal scaling

**Concept 2: Password Hashing**
- **What it is:** Converting plaintext passwords to irreversible hash values
- **WHY needed:** Even if database is leaked, attackers cannot obtain original passwords
- **WHY use bcrypt:** Built-in salt, adjustable computation cost to resist brute force
- **WHY not MD5/SHA1:** Computation too fast, easily brute-forced

### Concept Relationship Matrix

| Relationship Type | Concept A | Concept B | WHY Related This Way |
|-------------------|-----------|-----------|---------------------|
| Dependency | User Auth | Password Hash | Auth process needs password verification, must hash before comparison |
| Sequence | Password Hash | Token Generation | Token generated only after password verification passes |
| Comparison | JWT | Session | JWT stateless, suitable for distributed; Session stateful, more server pressure |

### Connections to Existing Knowledge

- **Design Patterns:** [Detailed below]
- **Algorithm Theory:** [Detailed below]
- **Security Principles:** Principle of least privilege, defense in depth
```

---

### Step 4: Algorithm & Theory Deep Analysis

**Mandatory Requirements:** All algorithms and core theories must:
1. Annotate time/space complexity
2. Explain "WHY this complexity is acceptable"
3. Provide authoritative references
4. Explain when it degrades

**Output Format:**

```markdown
## Algorithm & Theory Analysis

### Algorithm: Quick Sort

**Basic Information:**
- **Time Complexity:** Average O(n log n), Worst O(n²)
- **Space Complexity:** O(log n)

**Elaborative Interrogation:**

**WHY choose Quick Sort?**
- Excellent average performance, typically fastest in practice
- In-place sorting, high space efficiency
- Cache-friendly, good locality of reference

**WHY is worst O(n²) acceptable?**
- Worst case probability extremely low (avoidable through randomization)
- Real data typically not perfectly sorted/reversed
- Can optimize with median-of-three

**WHY not other sorting algorithms?**
- Merge Sort: Requires O(n) extra space, unsuitable for memory-constrained scenarios
- Heap Sort: Though stable O(n log n), poor cache performance, slower than Quick Sort in practice
- Insertion Sort: Excellent for small datasets, but O(n²) unsuitable for large-scale data

**When does it degrade?**
- Input already sorted or reversed (solvable with randomization)
- Poor pivot selection (solvable with median-of-three)
- Many duplicate elements (optimizable with 3-way Quick Sort)

**References:**
- [Quick Sort - Wikipedia](https://en.wikipedia.org/wiki/Quicksort)
- [Quick Sort Analysis - Princeton](https://algs4.cs.princeton.edu/23quicksort/)
- [Why is QuickSort better than MergeSort?](https://stackoverflow.com/questions/70402/why-is-quicksort-better-than-other-sorting-algorithms-in-practice)

### Theoretical Foundation: JWT (JSON Web Token)

**WHY use JWT?**
- Stateless authentication, server doesn't need to store sessions
- Self-contained, token carries all necessary information
- Cross-domain friendly, suitable for microservices architecture

**WHY is JWT secure?**
- Uses signature to verify integrity
- Cannot be forged (unless private key is leaked)
- Can set expiration time (exp)

**WHY does JWT have limitations?**
- Cannot actively invalidate (unless maintaining blacklist, breaking stateless advantage)
- Token size larger (Base64 encoding increases size by ~33%)
- Sensitive info needs encryption, signature alone doesn't provide confidentiality

**References:**
- [JWT.io - Introduction](https://jwt.io/introduction)
- [RFC 7519 - JWT Specification](https://tools.ietf.org/html/rfc7519)
```

---

### Step 5: Design Pattern Recognition & Interrogation

**Mandatory Check:** Each design pattern used in code must:
1. Clearly annotate pattern name
2. Explain WHY use this pattern
3. Explain what happens without this pattern
4. Provide standard reference

**Output Format:**

```markdown
## Design Pattern Analysis

### Pattern 1: Singleton Pattern

**Application Location:** `DatabaseConnection` class

**WHY use Singleton?**
- Database connection overhead high, reusing single instance saves resources
- Avoids connection pool chaos, unified connection lifecycle management
- Global unique access point, convenient concurrency control

**WHY not using Singleton would be problematic?**
- Every operation creates new connection, resource exhaustion
- Multiple connection instances may cause transaction inconsistency
- Difficult to control concurrent access

**Implementation Details:**
```python
class DatabaseConnection:
    _instance = None
    
    def __new__(cls):
        if cls._instance is None:
            cls._instance = super().__new__(cls)
            # WHY initialize in __new__:
            # Ensure singleton before object creation, thread-safe
        return cls._instance
```

**WHY implement this way?**
- Use `__new__` instead of `__init__`: Control instance creation, not initialization
- Class variable `_instance`: Store unique instance
- Lazy Loading: Create only on first use

**Potential Issues:**
- ⚠️ Not thread-safe (multi-threading needs locking)
- ⚠️ Unit testing difficult (global state hard to isolate)
- ⚠️ Violates single responsibility principle (class manages its own instance)

**Better Alternatives:**
- Dependency Injection: More flexible, easier to test
- Module-level variable: Python modules are naturally singleton

**References:**
- [Singleton Pattern - Refactoring Guru](https://refactoring.guru/design-patterns/singleton)
- [Singleton Pattern in Python - Real Python](https://realpython.com/factory-method-python/)
```

---

### Step 6: Line-by-Line Deep Analysis (Key Code Segments)

**Core Principle:**
- Select 3-5 most critical code segments
- Each line must explain "what it does" + "WHY it does this"
- Provide execution flow examples with concrete data
- Mark error-prone points and boundary conditions

**Output Format:**

```markdown
## Key Code Deep Analysis

### Code Segment 1: User Authentication Function

**Overall Purpose:** Verify username and password, return JWT Token or None

**WHY this function is needed:** Authentication is the first line of system security, must be reliable and efficient

**Original Code:**
```python
def authenticate_user(username, password):
    user = db.find_user(username)
    if not user:
        return None
    if verify_password(password, user.password_hash):
        return generate_token(user.id)
    return None
```

**Line-by-Line Detailed Analysis (Recommended Comment Style: Scenario-Based + Execution Flow Tracking)**

> **Comment Style Guide:**
> - **`# Scenario N: [description]` / `// Scenario N: [description]`** - Label different execution paths in conditional branches (if/else, switch, match, etc.)
> - **`# Step N: [description]` / `// Step N: [description]`** - Label sequential execution flow (initialization order, function call sequence, etc.)
> - Comment symbols match language: Python uses `#`, C++/Java use `//`
> - Track execution flow with concrete variable values (`# At this point: xxx` / `// At this point: xxx`)
> - Annotate loop/recursion iteration states
> - Mark key data transformation trajectories

```python
def authenticate_user(username, password):
    # Step 1: Query user
    user = db.find_user(username)
    # WHY query user first: Avoid password hashing for non-existent usernames (save computation)

    # Scenario 1: If user doesn't exist, return None immediately
    if not user:
        return None
        # WHY return None not exception: Auth failure is normal flow, not exceptional
        # WHY not distinguish "user doesn't exist" from "wrong password": Prevent username enumeration

    # Scenario 2: If password verification passes, generate and return Token
    if verify_password(password, user.password_hash):
        # verify_password internal flow:
        #   1. Extract salt from password_hash
        #   2. Hash plaintext password with same salt
        #   3. Constant-time compare hashes (prevent timing attack)
        return generate_token(user.id)
        # At this point: user.id = 42 (assumed)
        # generate_token(42) → "eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9..."

    # Scenario 3: Wrong password, return None
    return None
    # WHY same return value as "user doesn't exist": Prevent attackers from distinguishing failures
```

**Complete Execution Flow Example (Multi-Scenario Tracking):**

```cpp
// Example: Function to trace tensor producer (compiler code typical style)

Value getProducerOfTensor(Value tensor) {
  Value opResult;

  while (true) {
    // Scenario 1: If tensor defined by LinalgOp, return directly
    if (auto linalgOp = tensor.getDefiningOp<LinalgOp>()) {
      opResult = cast<OpResult>(tensor);
      // while loop executes only 1 iteration
      return;
    }

    // Per this example, on first call: tensor = %2_tile
    // Scenario 2: If tensor linked through ExtractSliceOp, trace source
    if (auto sliceOp = tensor.getDefiningOp<tensor::ExtractSliceOp>()) {
      tensor = sliceOp.getSource();
      // At this point: tensor = %2, defined by linalg.matmul
      // Second while loop iteration will enter Scenario 1 (linalg.matmul is a LinalgOp)
      continue;
    }

    // Scenario 3: Through scf.for iteration argument
    // Example IR:
    // %1 = linalg.generic ins(%A) outs(%init) { ... }
    // %2 = scf.for %i = 0 to 10 iter_args(%arg = %1) {
    //   %3 = linalg.generic ins(%arg) outs(%init2) { ... }
    //   scf.yield %3
    // }
    // getProducerOfTensor(%arg)
    if (auto blockArg = dyn_cast<BlockArgument>(tensor)) {
      // First while iteration: tensor = %arg, is a BlockArgument
      if (auto forOp = blockArg.getDefiningOp<scf::ForOp>()) {
        // %arg defined by scf.for, get loop initial value: %1
        // blockArg.getArgNumber() = 0 (%arg is 0th iteration argument)
        // forOp.getInitArgs()[0] = %1
        tensor = forOp.getInitArgs()[blockArg.getArgNumber()];
        // At this point: tensor = %1, defined by linalg.generic
        // Second while loop iteration will enter Scenario 1
        continue;
      }
    }

    return;  // Not found (might be function parameter)
  }
}
```

**Execution Flow Examples (Recommended Style):**

**Scenario 1: Successful Authentication**
```
# Initial state
Input: username="alice", password="Secret123!"

# Execution path
Step 1: db.find_user("alice")
   → Query database
   → Return User(id=42, username="alice", password_hash="$2b$12$KIX...")
   # At this point: user exists, skip Scenario 1's return None

Step 2: Enter Scenario 2 branch (password verification)
   → verify_password("Secret123!", "$2b$12$KIX...")
   → Extract salt: $2b$12$KIX...
   → Hash "Secret123!" with salt
   → Constant-time compare hashes
   → Return True

Step 3: generate_token(42)
   → Create payload: {"user_id": 42, "exp": 1643723400}
   → Sign with private key
   → Return "eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJ1c2VyX2lkIjo0Miwi..."
   # Final return: Token string

# Performance analysis
Duration: ~100ms (mainly bcrypt computation)
```

**Scenario 2: User Doesn't Exist**
```
# Initial state
Input: username="bob", password="anything"

# Execution path
Step 1: db.find_user("bob")
   → Query database
   → Return None
   # At this point: user = None, enter Scenario 1 branch

Step 2: if not user: # true
   → Return None immediately
   # Scenarios 2, 3 not executed

# Performance analysis
Duration: ~5ms (database query only)
⚠️ Note: Much faster than successful auth, may leak user existence
# Security recommendation: Add fixed delay or fake hash to make durations similar
```

**Scenario 3: Wrong Password**
```
# Initial state
Input: username="alice", password="WrongPass"

# Execution path
Step 1: db.find_user("alice")
   → Return User(id=42, ...)
   # At this point: user exists, skip Scenario 1's return None

Step 2: Enter Scenario 2 branch (password verification)
   → verify_password("WrongPass", "$2b$12$KIX...")
   → Hash "WrongPass"
   → Compare hashes
   → Return False

Step 3: Password verification failed, don't execute generate_token
   → Continue to final return None
   # Scenario 3: Password verification failed, return None

# Performance analysis
Duration: ~100ms (similar to successful auth)
✅ Benefit: Cannot determine password correctness by response time
```

**Key Takeaways:**

1. **Security Considerations:**
   - ✅ Plaintext password exists briefly in memory, immediately hashed for verification
   - ✅ Failure reasons not leaked (prevent username enumeration)
   - ✅ Constant-time comparison (prevent timing attack)
   - ⚠️ Potential issue: Faster response when user doesn't exist (needs optimization)

2. **Performance Optimization:**
   - ✅ Fast return when user doesn't exist, no wasted hash computation
   - ⚠️ But this causes timing leakage, need to balance security vs performance

3. **Error Handling:**
   - ✅ Use None for failure, clear and Pythonic
   - ⚠️ Caller must check return value, or risk misusing None

4. **Improvements Possible:**
   - Add logging for failed attempts (detect brute force)
   - Add rate limiting
   - Unify failure scenario response times
```

---

### Step 6.5: Test Case Reverse Understanding (If Tests Available)

**Goal:** Verify and deepen understanding of code functionality through test cases

**Why it's important:**
- Test cases reflect the **expected behavior** of code, serving as the most accurate "user manual"
- Tests typically cover **boundary conditions** and **exception scenarios** that are easily overlooked in main code
- Tests can **verify whether understanding is correct**, avoiding incorrect assumptions

**When test files are detected in code, this step is mandatory.**

#### 6.5.1 Test File Identification

**Common test file patterns:**

| Language | Test File Pattern | Test Directory Structure |
|----------|-------------------|-------------------------|
| **Python** | `test_*.py`, `*_test.py` | `tests/`, `test/` |
| **JavaScript/TypeScript** | `*.test.ts`, `*.test.js` | `__tests__/`, `tests/` |
| **Go** | `*_test.go` | Same directory as source, `*_test.go` |
| **Java** | `*Test.java`, `*Tests.java` | `src/test/java/` |
| **C++** | `*.cpp` (with tests), gtest | `test/`, `tests/`, `unittest/` |
| **Rust** | `*_test.rs`, `tests/*.rs` | `tests/` |
| **MLIR/LLVM** | `*.mlir` (test files) | `test/Dialect/*/` |

**Large project test directory structure examples:**

```bash
# MLIR style (separate test directory)
mlir/test/Dialect/Linalg/
├── ops.mlir           # Linalg dialect operation tests
├── transformation.mlir # Transformation tests
├── interfaces.mlir    # Interface tests
└── invalid.mlir       # Error handling tests

# Traditional C++ project style
project/test/
├── unittest/          # Unit tests
├── integration/       # Integration tests
└── benchmark/         # Performance tests
```

#### 6.5.2 Test Coverage Analysis

**Analyze functionality points covered by tests:**

```markdown
## Test Case Coverage Analysis

### Test File Inventory
| Test File/Directory | Module Tested | Test Case Count |
|---------------------|--------------|----------------|
| `test/Dialect/Linalg/ops.mlir` | Linalg Ops | 156 |
| `test/Dialect/Linalg/invalid.mlir` | Error Handling | 43 |
| `unittest/test_auth.cpp` | `authenticate_user()` | 12 |

### Functionality Coverage Matrix
| Core Function | Main Code Location | Test Coverage | Coverage Assessment |
|--------------|-------------------|--------------|---------------------|
| linalg.matmul operation | `Dialect/Linalg/Ops/*` | ✅ Has tests | Normal + boundary covered |
| linalg.generic interface | `Interfaces/*` | ✅ Has tests | Complete coverage |
| Tile transformation | `Transforms/Tiling.cpp` | ⚠️ Insufficient tests | Missing nested scenarios |
```

#### 6.5.3 Understanding Boundary Conditions from Tests

**Extract key boundary conditions from tests:**

```markdown
## Boundary Conditions Discovered from Tests

### MLIR Example: Understanding linalg.generic Region Constraints

#### Test File: test/Dialect/Linalg/invalid.mlir
```mlir
// Test: generic's region must have exactly one block
func.func @invalid_generic_empty_region(%arg0: tensor<10xf32>) -> tensor<10xf32> {
  %0 = linalg.generic {indexing_maps = [affine_map<(d0) -> (d0)>],
                     iterator_types = ["parallel"]}
    outs(%arg0) {
    // Empty region - should error
  } -> tensor<10xf32>
  return %0 : tensor<10xf32>
}
```
**WHY this test is important:**
- Reveals `linalg.generic`'s **structural constraint**: Must have a block
- Through **negative testing** (invalid test) clarifies error conditions
- Boundary condition: region's block count must = 1

#### Test File: test/Dialect/Linalg/ops.mlir
```mlir
// Test: Input and output counts must match indexing_maps
func.func @generic_mismatched_maps(%a: tensor<10xf32>, %b: tensor<10xf32>) -> tensor<10xf32> {
  %0 = linalg.generic {
    indexing_maps = [
      affine_map<(d0) -> (d0)>,  // 1 input map
      affine_map<(d0) -> (d0)>   // 1 output map
    ],
    iterator_types = ["parallel"]
  } ins(%a, %b : tensor<10xf32>, tensor<10xf32>)  // But 2 inputs
  outs(%0 : tensor<10xf32>) {
  ^bb0(%in: f32, %in_2: f32, %out: f32):
    linalg.yield %in : f32
  } -> tensor<10xf32>
  return %0 : tensor<10xf32>
}
```
**WHY handled this way:**
- Validates **type system constraint**: Input/output count must match maps
- Tests **static verification** logic, catching errors at compile time
- Illustrates MLIR's **static strong typing** characteristic

### C++ Example: Understanding Thread Safety through Tests

#### Test File: unittest/concurrent_map_test.cpp
```cpp
// Test: Concurrent insert of same key
TEST(ConcurrentMapTest, ConcurrentInsertSameKey) {
  ConcurrentMap<int, int> map;
  const int num_threads = 10;
  const int key = 42;

  std::vector<std::thread> threads;
  for (int i = 0; i < num_threads; ++i) {
    threads.emplace_back([&map, key, i]() {
      map.Insert(key, i);  // All threads insert same key
    });
  }

  for (auto& t : threads) t.join();

  // Verify: Only one insert succeeded
  EXPECT_EQ(map.Size(), 1);
  EXPECT_TRUE(map.Contains(key));
}
```
**WHY this test exists:**
- Verifies **thread safety**: Multi-threaded concurrent access won't crash
- Illustrates **conflict handling strategy**: Later inserts overwrite earlier (or vice versa)
- Tests **consistency guarantee**: Final state meets expectations
```

#### 6.5.4 Test-Driven Understanding Example

**Complete example: Understanding `linalg.tile` transformation through MLIR tests**

```markdown
## Test Case Reverse Understanding: linalg.tile Transformation

### Question: Can we understand all tile behavior from documentation alone?

**Documentation description (simplified):**
> `linalg.tile` decomposes linalg operations into smaller fragments

**Potentially missed details:**
1. How is tile size determined?
2. Which operations support tile?
3. What's the loop order after tile?
4. How are remaining elements handled?

### Answers Discovered from Tests

#### Test 1: test/tile-mlir.mlir - Basic tile behavior
```mlir
// Original operation
%0 = linalg.matmul ins(%A: tensor<128x128xf32>, %B: tensor<128x128xf32>)
                     outs(%C: tensor<128x128xf32>)

// Tile size 32x32
%1 = linalg.tile %0 tile_sizes[32, 32]
```
**Discovery:** Tile size directly specified, output contains nested loop structure

#### Test 2: test/tile-mlir.mlir - Remaining element handling
```mlir
// 127x127 matrix, tile size 32x32
%0 = linalg.matmul ins(%A: tensor<127x127xf32>, ...)
%1 = linalg.tile %0 tile_sizes[32, 32]
```
**Discovery:** Auto-generates boundary checks for uneven remainders

#### Test 3: test/tile-mlir.mlir - Non-tileable operations
```mlir
// Try tiling unsupported operation
%0 = linalg.generic ...
%1 = linalg.tile %0 tile_sizes[16]
// Expected: Compile error or runtime failure
```
**Discovery:** Not all operations support tile, has clear limitation conditions

### Understanding Comparison Before vs After Tests

| Question | Documentation Only | After Tests |
|----------|-------------------|-------------|
| How to specify tile size? | ⚠️ Unclear | ✅ Direct parameter |
| How are remainders handled? | ❓ Not mentioned | ✅ Auto boundary check |
| Which operations supported? | ❓ Incomplete list | ✅ Tests cover all supported ops |
| What's loop order? | ⚠️ Vague description | ✅ Visible in test IR |

**Conclusion:** Test cases supplement about **50%** of implementation details!
```

#### 6.5.5 Language-Specific Test File Parsing Points

**Key points for each language's tests:**

```markdown
## Language-Specific Test File Parsing Points

### Python (pytest/unittest)
- Look for `test_*.py` or `*_test.py`
- Note `@pytest.mark.parametrize` parameterized tests
- Focus on `pytest.raises` exception tests
- Find fixtures (`conftest.py`) for test context

### C++ (gtest/gtest)
- Look for `*_test.cpp` or `test/*.cpp`
- `TEST_F` indicates fixture test with preconditions
- `EXPECT_*` vs `ASSERT_*`: Whether execution continues on failure
- `TEST_P` indicates parameterized test

### MLIR/LLVM
- Test files typically `.mlir` or `.td`
- `RUN:` command specifies how to execute test
- `// EXPECTED:` marks expected output
- `// ERROR:` marks expected compilation errors
- FileCheck directives: `CHECK-`, `CHECK-NOT:`, `CHECK-DAG:`

### JavaScript/TypeScript (Jest)
- `*.test.ts`, `*.spec.ts`
- `describe/it` nested structure
- `expect(...).toThrow()` exception tests
- `beforeEach/afterEach` hook functions

### Go
- Tests co-located with source: `*_test.go`
- `TestXxx(t *testing.T)` basic tests
- `TableDrivenTests` table-driven tests
- `TestMain` test entry point

### Rust
- `*_test.rs` inline tests
- `tests/` directory integration tests
- `#[should_panic]` exception tests
- `#[ignore]` skipped tests
```

#### 6.5.6 Test Quality Assessment

**Assess whether tests are adequate:**

```markdown
## Test Quality Assessment

### Covered Functionality Points
- ✅ Normal flow
- ✅ Boundary inputs
- ✅ Exception inputs
- ⚠️ Concurrent scenarios
- ❌ Performance tests

### MLIR-Specific Assessment
- ✅ Positive tests (valid.mlir)
- ✅ Negative tests (invalid.mlir)
- ⚠️ Performance regression tests
- ❌ Cross-dialect interaction tests

### Test Coverage Warning
> ⚠️ **Warning: Module has insufficient test coverage**
> - Uncovered scenarios: [List specific items]
> - Suggested additions: [Specific recommendations]
```

#### 6.5.7 Test Case Analysis Output Template

```markdown
## Test Case Analysis

### Test File Structure
[List test files/directories and their corresponding source modules]

### Key Test Case Interpretations
[Select 3-5 most valuable test cases]

### Hidden Behaviors Discovered from Tests
[List details easily missed when reading code only]

### Test Coverage Assessment
- Core functionality coverage: X%
- Boundary condition coverage: [Adequate/Insufficient]

### Test Quality Recommendations
[If tests inadequate, provide improvement suggestions]
```

---

### Step 9: Application Transfer Test (Verify True Understanding)

**Goal:** Test if concepts can apply to different scenarios

**Must include:**
- At least 2 different domain application scenarios
- Explain how to adjust code for new scenarios
- Mark what principles remain constant, what needs modification

**Output Format:**

```markdown
## Application Transfer Scenarios

### Scenario 1: Apply User Authentication to API Key Validation

**Original Scenario:** Web user login authentication  
**New Scenario:** Third-party API key validation

**Unchanged Principles:**
- Core flow of verifying caller identity
- Hash storage of credentials (API keys should also be hashed)
- Access token generation mechanism

**Parts Needing Modification:**

```python
# Original: username + password
def authenticate_user(username, password):
    user = db.find_user(username)
    if not user:
        return None
    if verify_password(password, user.password_hash):
        return generate_token(user.id)
    return None

# Transfer: API key
def authenticate_api_key(api_key):
    # WHY only one parameter: API key itself is identity + credential
    
    app = db.find_app_by_key_prefix(api_key[:8])
    # WHY query by prefix: Avoid full table scan, API key prefix as index
    
    if not app:
        return None
    
    if verify_api_key(api_key, app.key_hash):
        # WHY also hash: Prevent key leakage from database breach
        
        return generate_token(app.id, scope=app.permissions)
        # WHY add scope: API keys typically have different permission levels
        
    return None
```

**WHY transfer this way:**
- Preserve core security principles (hash storage, constant-time comparison)
- Adjust business logic (single parameter, permission scope)
- Optimize query performance (prefix indexing)

**Learned Universal Pattern:**
- Any "who's calling" verification scenario can use similar structure
- Core: Find entity → Verify credential → Generate token
- Variation: Credential form, query method, token content

### Scenario 2: Apply Quick Sort to Log Analysis

**Original Scenario:** Sort user list by ID  
**New Scenario:** Sort millions of log entries by timestamp

**Unchanged Principles:**
- Divide and conquer approach: Recursive problem decomposition
- Pivot selection: Key to performance
- In-place sorting: Save space

**Parts Needing Adjustment:**

```python
# Original: Simple quick sort
def quicksort(arr):
    if len(arr) <= 1:
        return arr
    pivot = arr[len(arr) // 2]
    left = [x for x in arr if x < pivot]
    middle = [x for x in arr if x == pivot]
    right = [x for x in arr if x > pivot]
    return quicksort(left) + middle + quicksort(right)

# Transfer: Log sorting (external sort + optimization)
def quicksort_logs(log_file, output_file, memory_limit):
    # WHY external sort: Data volume exceeds memory, cannot load at once
    
    # 1. Sort chunks
    chunks = split_file_into_chunks(log_file, memory_limit)
    # WHY chunking: Each chunk can fit in memory for individual sorting
    
    for chunk in chunks:
        logs = load_chunk(chunk)
        
        # WHY use timsort instead of quicksort:
        # - Logs typically partially ordered (appended by time)
        # - timsort optimized to O(n) for partially ordered data
        # - Python's built-in sorted() is timsort
        logs.sort(key=lambda log: log.timestamp)
        
        save_sorted_chunk(chunk, logs)
    
    # 2. Merge sorted chunks
    merge_sorted_chunks(chunks, output_file)
    # WHY merge: Combine multiple sorted sequences into one
    
    return output_file
```

**WHY not use quick sort directly:**
- Data volume exceeds memory: Need external sort
- Logs partially ordered: timsort more optimal
- Need stable sort: Maintain order of logs with same timestamp

**Learned Universal Pattern:**
- Algorithm choice depends on data characteristics (scale, ordering, stability requirements)
- Basic principles transferable (divide-and-conquer, comparison), but implementation needs adjustment
- Super large data needs external algorithms (chunking + merging)
```

---

### Step 10: Dependencies & Usage Examples

(Similar to original but with WHY explanations)

```markdown
## Dependency Analysis

### External Libraries

**bcrypt (v5.1.0)**
- **Purpose:** Password Hashing
- **WHY choose bcrypt:**
  - Built-in salt, no manual management
  - Adjustable computation cost (cost factor)
  - Resist GPU/ASIC acceleration attacks
- **WHY not SHA256:** Computation too fast, easily brute-forced
- **WHY not scrypt/argon2:** bcrypt more mature, better compatibility

**jsonwebtoken (v9.0.0)**
- **Purpose:** JWT token generation and verification
- **WHY choose JWT:** Stateless authentication, suitable for distributed systems
- **WHY not Session:** Session requires server storage, not scalable

### Internal Module Dependencies

**database.js → auth.js**
- **Dependency reason:** Authentication needs to query user data
- **WHY this design:** Separate data access and business logic (single responsibility principle)

**utils/crypto.js → auth.js**
- **Dependency reason:** Authentication needs password hashing and verification
- **WHY encapsulate utility module:** Cryptographic logic complex, centralized management more secure

## Complete Usage Examples

(With detailed WHY comments)

### Example 1: Standard User Login Flow

```javascript
// 1. Import authentication module
const auth = require('./auth');

// 2. Receive user input (from login form)
const username = req.body.username;  // e.g., "alice"
const password = req.body.password;  // e.g., "Secret123!"

// WHY not hash password on client:
// - After client-side hashing, the hash itself becomes the "password"
// - Attacker with hash can login directly
// - Must hash with salt on server, client always sends plaintext

// 3. Call authentication function
const token = await auth.authenticate_user(username, password);

// 4. Respond based on result
if (token) {
    // Authentication successful
    res.json({
        success: true,
        token: token,
        // WHY return token: Client needs to carry it in subsequent requests
        message: 'Login successful'
    });
    
    // WHY set HTTP-only Cookie (optional):
    // res.cookie('auth_token', token, {
    //     httpOnly: true,    // WHY: Prevent XSS attack from reading
    //     secure: true,      // WHY: HTTPS transmission only
    //     sameSite: 'strict' // WHY: Prevent CSRF attack
    // });
} else {
    // Authentication failed (user doesn't exist or wrong password)
    
    // WHY not distinguish failure reason: Prevent username enumeration
    res.status(401).json({
        success: false,
        message: 'Invalid username or password'  // Vague error message
    });
    
    // WHY return 401 not 403:
    // 401 = Unauthenticated (need to provide credentials)
    // 403 = Authenticated but no permission
}
```

**Execution Result Analysis:**

**Success Path:**
```
Client request → Server verify → Return Token
Time: ~100ms
Token example: "eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9..."
```

**Failure Path:**
```
Client request → Server verify → Return 401 error
Time: ~100ms (similar to success, prevent timing attack)
```
```

---

### Step 11: Self-Assessment Checklist

**After analysis completion, mandatory verification of following items:**

```markdown
## Quality Verification Checklist

### Understanding Depth Verification

- [ ] **Each core concept answers 3 WHYs**
  - WHY this concept is needed
  - WHY implemented this way
  - WHY not other approaches

- [ ] **Self-explanation test passed**
  - [ ] Can explain each core concept without looking at code
  - [ ] Can articulate WHY not just WHAT
  - [ ] Can apply in different scenarios (transfer test)

- [ ] **Concept connections established**
  - [ ] Annotated dependency/comparison/composition relationships between concepts
  - [ ] Connected to existing knowledge (design patterns, algorithm theory)
  - [ ] Explained WHY for each connection

### Technical Accuracy Verification

- [ ] **Algorithm analysis complete**
  - [ ] Time/space complexity annotated
  - [ ] WHY choose this algorithm
  - [ ] WHY complexity is acceptable
  - [ ] Provided authoritative references

- [ ] **Design pattern recognition**
  - [ ] All patterns annotated
  - [ ] WHY use this pattern
  - [ ] What happens without it
  - [ ] Provided standard reference

- [ ] **Code analysis detailed**
  - [ ] Key code segments have line-by-line analysis
  - [ ] Each line includes "what it does" + "WHY it does this"
  - [ ] Provided execution examples with concrete data
  - [ ] Marked error-prone points and boundary conditions

### Practicality Verification

- [ ] **Application transfer test**
  - [ ] At least 2 different scenario transfer examples
  - [ ] Explained what stays constant, what needs change
  - [ ] Extracted universal patterns

- [ ] **Usage examples runnable**
  - [ ] Example code complete
  - [ ] Contains detailed WHY comments
  - [ ] Explained execution results

- [ ] **Issues & improvement suggestions**
  - [ ] Pointed out potential issues
  - [ ] WHY they are issues
  - [ ] Provided improvement solutions
  - [ ] WHY improvement solutions are better

### Final Verification Questions

**If not looking at original code, based on this analysis document:**

1. ✅ Can you understand the design rationale?
2. ✅ Can you independently implement similar functionality?
3. ✅ Can you apply to different scenarios?
4. ✅ Can you clearly explain to others?

If any answer is "no," the analysis is insufficient and needs supplementation.
```

---

## Output Format Summary

**Complete Analysis Document Structure:**

```markdown
# [Code Name] Deep Understanding Analysis

## Understanding Verification Status
[Self-explanation test results table]

## 1. Quick Overview
- Programming Language:
- Code Scale:
- Core Dependencies:

## 2. Background & Motivation Analysis (Elaborative Interrogation)
- Problem Essence (WHY needed)
- Solution Choice (WHY chosen + WHY not others)
- Application Scenarios (WHY applicable + WHY not)

## 3. Concept Network Diagram
- Core Concept Inventory (3 WHYs per concept)
- Concept Relationship Matrix
- Connections to Existing Knowledge

## 4. Algorithm & Theory Deep Analysis
- Each algorithm: Complexity + WHY chosen + WHY acceptable + References
- Each theory: WHY used + WHY effective + WHY limited

## 5. Design Pattern Analysis
- Each pattern: WHY used + WHY not using is problematic + Implementation details + References

## 6. Key Code Deep Analysis
- Each segment: Line-by-line analysis (what + WHY) + Execution examples + Key takeaways

## 7. Test Case Analysis (if tests available)
- Test file structure + Key test interpretations + Hidden behaviors discovered

## 8. Application Transfer Scenarios (at least 2)
- Each scenario: Unchanged principles + Parts needing modification + WHY transfer this way

## 9. Dependencies & Usage Examples
- Each dependency: WHY chosen + WHY not others
- Examples contain detailed WHY comments

## 10. Quality Verification Checklist
[Check all verification items]
```

---

## Special Scenario Handling

### Multi-File Projects

1. **Overall Architecture Analysis**
   - Project structure tree + WHY organized this way
   - Entry files + WHY start here
   - Module division + WHY divided this way

2. **Inter-Module Relationships**
   - Dependency graph + WHY depend this way
   - Data flow diagram + WHY flow this way
   - Call chain + WHY call this way

3. **Module-by-Module Analysis**
   - Each core module analyzed by standard workflow
   - Emphasize WHY relationships between modules

### Complex Algorithms

1. **Layered Explanation**
   - First describe approach in natural language
   - Then show structure with pseudocode
   - Finally line-by-line analysis of implementation

2. **WHY Throughout**
   - WHY choose this algorithm
   - WHY each step done this way
   - WHY complexity is what it is

3. **Visualization Aid**
   - Show execution process with concrete data
   - Explain WHY at each step

### Unfamiliar Tech Stacks

1. **Technical Background Explanation**
   - What this tech stack is
   - WHY this tech stack exists
   - WHY project chose it

2. **Key Concept Explanation**
   - Tech stack-specific concepts
   - WHY designed this way
   - Comparison with other tech stacks

3. **Learning Resources**
   - Official documentation links
   - WHY recommend these resources
   - Learning path suggestions

---

## Pre-Analysis Final Check

Before starting analysis, confirm:

- [ ] Understood user's true needs (learning? review? interview prep?)
- [ ] Identified code's language, framework, scale
- [ ] Determined analysis focus (comprehensive understanding vs specific aspect)
- [ ] Ready to always ask WHY
- [ ] Ready to conduct self-explanation test
- [ ] Ready to find concept connections
- [ ] Ready to think about application transfer

**Remember: Goal is not "finish reading code" but "truly understand code."**

---

## ✍️ Human-Friendly Writing Standards

**The analysis document should read like an experienced engineer explaining things to you, not a textbook mechanically reciting definitions.**

### Core Requirements

**1. Open with a conversational lead-in, not a bare definition**

❌ Stiff:
> The `authenticate_user` function is responsible for executing the user identity verification process, accepting username and password parameters...

✅ Natural:
> Think of this function as the bouncer at the door. Every time a user tries to log in, they have to go through it. The logic is simple: first check if the user exists, then verify the password — only then do they get in.

**2. When explaining WHY, lead with the conclusion, then expand**

❌ Stiff:
> Due to bcrypt's adaptive hashing function characteristics, its computational cost can be dynamically adjusted via the cost factor parameter, hence its selection.

✅ Natural:
> The reason to use bcrypt instead of MD5 comes down to speed — MD5 is too fast. A regular computer can compute billions of MD5 hashes per second, making brute force trivial. bcrypt is deliberately slow, and you can tune just how slow it is as hardware improves.

**3. Analogies are your friend**

- Locks → "A read-write lock is like a library reading room: many people can read at once, but when someone needs to write, everyone else has to wait"
- Caching → "A cache is like keeping frequently used things on your desk instead of fetching from the warehouse every time"
- Recursion → "Like looking up a word in a dictionary, only to find you need to look up another word in its definition"

**4. Code comments should speak plain language**

❌ Stiff comment:
```python
# Execute password hash verification operation
if verify_password(password, user.password_hash):
```

✅ Natural comment:
```python
# Check the password: re-hash what the user typed and see if it matches what's stored
# (We can't compare plain text because the database never stored it in plain text)
if verify_password(password, user.password_hash):
```

**5. For complex concepts, explain in layers**

First give the intuition in one sentence, then add technical detail:
> JWT is essentially a "stamped pass." When you log in successfully, the server hands you a pass. On every subsequent request you bring it along, the server checks that the stamp is genuine, and if so, lets you through. Technically it's three Base64-encoded segments joined by dots: header, payload, and signature.

**6. Flag "this is where people get tripped up"**

Don't just describe the happy path — call out common mistakes:
> ⚠️ **Common misconception:** Many people assume JWT is encrypted. It's not — it's only signed. Anyone can Base64-decode the payload and read it. Never put passwords, credit card numbers, or other sensitive data inside.

### Writing Patterns to Avoid

| Pattern | Why it's a problem |
|---------|-------------------|
| Wall of jargon with no explanation | Reader can't see how the terms relate |
| "This function implements X functionality" | States the obvious — reader already knows that |
| Every paragraph opens with the same sentence structure | Feels machine-generated, causes fatigue |
| WHY explanation in a single sentence | Doesn't actually answer why |
| Copy-pasting official documentation wording | The docs already exist; parroting adds no value |

### Self-Check Questions

After writing, ask yourself:
- Could a junior engineer just starting out follow this?
- Did I explain *why*, or only *what*?
- Does the reader get an "aha, now I get it!" moment anywhere?

---

## 📤 Output Requirements (Token-Optimized)

**After analysis completion, you MUST generate a standalone Markdown document!**

### Three Mode Document Generation Strategies

| Mode | Generation Method | File Count | Typical Use Case |
|------|------------------|------------|------------------|
| **Quick** | Single Write | 1 | Quick code review |
| **Standard** | Single Write | 1 | Learning to understand code |
| **Deep** | Auto-select strategy based on scale | 1-2 | Interview prep, complete mastery |
| → Code ≤ 2000 lines | Progressive Write | 1-2 | Medium-small code |
| → Code > 2000 lines | Multi-agent parallel + aggregate | Multiple chapters → 1 final doc | Large projects, complex codebases |

### ⚡ Token-Saving Strategies

**Core Principle: Avoid duplicate output, write directly to file**

1. **Never output full analysis in conversation**
   - Write complete analysis directly to file, don't output to conversation
   - In conversation, output only: analysis summary + file path

2. **Chunk large projects**
   - Single file analysis: Generate single document
   - Multi-file projects: Generate multiple documents by module
   - Extra-long analysis: Split into `overview.md` + `[module]-detailed-analysis.md`

3. **Progressive generation** (for Deep Mode)
   - First generate framework document (TOC + summary)
   - Fill sections incrementally, update file with each Write call

### Document Generation Rules

1. **File Naming Format**
   - Single file: `[code-name]-deep-analysis.md`
   - Multi-file project: `[project-name]-overview.md` + `[module-name]-analysis.md`
   - Examples: `jwt-auth-deep-analysis.md`, `quicksort-deep-analysis.md`

2. **Generation Method (Token-Optimized Flow)**

   **Method 1: Direct write (Recommended)**
   ```
   User: Deeply analyze this code

   1. [Complete analysis process, don't output full content]

   2. Use Write tool to generate document directly:
      File path: [code-name]-deep-analysis.md
      Content: [Complete analysis content]

   3. Output brief summary in conversation:
      - Analysis mode: Standard/Deep
      - Key findings: 3-5 bullet points
      - File path: [code-name]-deep-analysis.md
   ```

   **Method 2: Multi-file project chunking**
   ```
   1. [Complete overall analysis]

   2. Generate overview document:
      Write: [project-name]-overview.md
      Content: Overall architecture, module relationship diagram, analysis framework

   3. Generate detailed documents per module:
      Write: [moduleA]-analysis.md
      Write: [moduleB]-analysis.md
      Write: [moduleC]-analysis.md

   4. Output summary:
      - Generated 4 documents
      - List all file paths
   ```

   **Method 3: Deep Mode (auto-select based on code scale)**
   ```
   Deep Mode will auto-select optimal generation strategy:

   【Strategy A: Progressive Generation】When code ≤ 2000 lines
   - First generate framework document (TOC + summary)
   - Fill sections incrementally, update file with each Write call
   - See "Deep Mode Output Structure - Strategy A" section above

   【Strategy B: Parallel Processing】When code > 2000 lines
   Flow:
   1. Master agent generates framework and task assignments
   2. Use Task tool to create multiple parallel sub-agents
   3. Each sub-agent focuses on one chapter, generates independent file
   4. Master agent aggregates all chapters, generates final document

   File structure:
   work/
   ├── 00-framework.json       # Framework generated by master agent
   ├── tasks/                  # Sub-task description directory
   │   ├── background-task.md
   │   ├── concepts-task.md
   │   └── ...
   ├── chapters/               # Chapters generated by sub-agents
   │   ├── background-chapter.md
   │   ├── concepts-chapter.md
   │   └── ...
   └── [project-name]-complete-mastery-analysis.md  # Final aggregated document

   Example Task call:
   Task(
     description: "Analyze Background & Motivation chapter",
     prompt: "You are a background & motivation analysis expert. Please deeply analyze the following code's background and motivation...[specific instructions]",
     subagent_type: "general-purpose"
   )
   ```

3. **Conversation Output Format (Brief)**

   ```markdown
   ## Analysis Complete

   **Mode:** Standard Mode

   **Key Findings:**
   - Code implements [core functionality]
   - Uses [algorithm/pattern] to solve [problem]
   - Key optimizations: [opt1], [opt2]
   - Potential issues: [issue1], [issue2]

   **Full Document:** `[code-name]-deep-analysis.md`
   ```

### Output Flow Comparison

**❌ High Token Consumption (Avoid):**
```
1. Output 5000 token complete analysis in conversation
2. Write 5000 token to file again
→ Total: 10000+ token output
```

**✅ Token-Optimized (Recommended):**
```
1. Write 5000 token directly to file
2. Output 200 token summary in conversation
→ Total: 5200 token output (~50% saved)
```

### Large Project Chunking Guide

| Project Scale | Recommended Mode | Generation Strategy | File Structure | Info Loss Prevention |
|--------------|-----------------|---------------------|----------------|---------------------|
| < 500 lines | Quick/Standard | Single document | `[name]-analysis.md` | Not needed |
| 500-2000 lines | Standard | Single document (may be long) | `[name]-analysis.md` | Build file inventory |
| 2000-10000 lines / files ≤ 20 | Deep (Strategy B) | Project map + parallel chapters | project-map.md + chapters → 1 final doc | Project map + coverage check |
| > 10000 lines / files > 20 | Deep (Strategy C) | Layered parallel | Module summaries + chapters → 1 final doc | Module scan + chapter coverage check |

**Important: Don't output complete analysis in conversation - write directly to file, only output summary!**

---

### 🚀 Deep Mode Auto Implementation Guide (Specific Instructions for Claude)

Deep Mode will auto-select optimal strategy. When parallel processing is needed:

#### Step 1: Identify if parallel processing is needed
```
Auto-trigger conditions (any match):
- Code files > 10
- Total code lines > 2000
- User explicitly says "large project", "complete project", "project analysis"
- User uses depth triggers like "thoroughly", "completely master", "in-depth research" with large code scale
```

#### Step 2: Select processing strategy
```
if code_lines <= 2000:
    Use Strategy A: Progressive Generation (sequential processing)
elif code_lines <= 10000 and file_count <= 20:
    Use Strategy B: Parallel Processing (chapter-level parallel)
else:
    Use Strategy C: Layered Parallel (module-level + chapter-level)
```

#### Step 3: Parallel processing preparation (Strategy B/C)
```bash
# Step 1: Enumerate ALL project files (critical for preventing info loss)
# Use tools to list all source files and generate complete inventory

# Create working directory
mkdir -p code-analysis/{tasks,chapters}

# Generate project map (must include all file paths)
cat > code-analysis/project-map.md << 'EOF'
# Project File Map

## Complete File Inventory
| File Path | Category | Lines | Responsibility |
|-----------|----------|-------|---------------|
| [List every file here] | | | |

## Core Module List
[List core business logic files]

## Test File List
[List all test files]
EOF

# Generate framework file (with chapter-to-file mapping)
cat > code-analysis/00-framework.json << 'EOF'
{
  "project_name": "[Project Name]",
  "language": "[Language]",
  "total_files": [File Count],
  "total_lines": [Line Count],
  "core_concepts": [Concept List],
  "chapter_file_mapping": {
    "Background & Motivation": ["path/to/main.py", "README.md"],
    "Core Concepts": ["path/to/core1.py", "path/to/core2.py"],
    "Algorithm & Theory": ["path/to/algo.py"],
    "Design Patterns": ["path/to/core1.py", "path/to/core2.py"],
    "Key Code Analysis": ["path/to/most_important.py", "..."],
    "Test Case Analysis": ["tests/test_a.py", "tests/test_b.py"],
    "Application Transfer": ["path/to/core1.py"],
    "Dependencies": ["requirements.txt", "...all files list"]
  }
}
EOF
```

#### Step 4: Create parallel sub-agents (with explicit file paths)
```
For each chapter, use Task tool to create independent sub-agent:

Task(
  description: "Deep analyze [Chapter Name] chapter",
  prompt: """
  You are a [Chapter Name] analysis expert.

  ## Context
  - Project: {project_name}
  - Language: {language}
  - Core concepts: {core_concepts}
  - Module responsibilities: {module_responsibilities}  # To avoid duplication

  ## Files You Must Read (in priority order)
  {chapter_file_list}  # From chapter_file_mapping for this chapter

  ## Mandatory Steps
  1. Use Read tool to read each file listed above
  2. Only begin analysis after reading all assigned files
  3. Analysis must reference actual code lines, not memory

  ## Task
  Deeply analyze the [Chapter Name] part of the code, generate detailed chapter content (at least {min_words} words).

  ## Requirements
  - Use Scenario/Step + WHY style comments
  - Answer 3 WHYs for each key point
  - Provide concrete execution examples
  - Cite authoritative sources
  - Every public function/class in assigned files must be mentioned

  ## Output
  Write complete chapter content to file:
  code-analysis/chapters/{chapter_name}.md
  """,
  subagent_type: "general-purpose"
)
```

#### Step 5: Coverage check and aggregate results
```
After all sub-agents complete:

1. Read code-analysis/project-map.md for the complete file list
2. Scan code-analysis/chapters/*.md, extract all referenced file paths
3. Generate coverage report:
   - Core module coverage target: 100%
   - Overall coverage target: ≥ 80%
4. For uncovered core modules, create supplemental analysis tasks
6. Read code-analysis/chapters/*.md (in order)
7. Merge into final document with coverage summary header
8. Write to {project_name}-complete-mastery-analysis.md
```

---

### 📋 Chapter Depth Self-Check Standards (Ensure Quality)

**For Deep Mode progressive generation, each chapter must pass these checks:**

```markdown
## Chapter Depth Self-Check Checklist

### 1. Content Completeness (Required)
- [ ] All chapter subsections covered (no "skipped", "see above", "same as above")
- [ ] Every WHY has specific explanation (at least 2-3 sentences, not just one)
- [ ] Code examples have complete comments (use Scenario/Step + WHY style)
- [ ] References have source links (algorithms/patterns/theories)

### 2. Analysis Depth (by Chapter Type)

**Concept chapters (Chapter 3):**
- [ ] Each core concept has 3 WHYs
  - WHY this concept is needed
  - WHY implemented this way
  - WHY not other approaches

**Algorithm chapters (Chapter 4):**
- [ ] Has time/space complexity annotation
- [ ] Has WHY choose this algorithm
- [ ] Has WHY complexity is acceptable
- [ ] Has degradation scenarios

**Design pattern chapters (Chapter 5):**
- [ ] Has pattern name and standard reference
- [ ] Has WHY use this pattern
- [ ] Has what happens without it

**Code analysis chapters (Chapter 6):**
- [ ] Has line-by-line analysis (what + WHY)
- [ ] Has execution examples with concrete data
- [ ] Has multi-scenario tracking (at least 2 scenarios)
- [ ] Has error-prone points and boundary conditions

### 3. Practicality (Application Value)
- [ ] Error-prone points marked
- [ ] Boundary conditions explained
- [ ] At least 2 application transfer scenarios
- [ ] Improvement suggestions have WHY

### 4. Format Standards
- [ ] Use Markdown format
- [ ] Code blocks have language tags
- [ ] Tables aligned correctly
- [ ] List hierarchy clear

### Handling Unqualified Chapters

**Case A: Too little content (< 300 words)**
→ Append details: Add more explanations, examples, comparisons

**Case B: Insufficient WHY analysis**
→ Supplement WHY: Ask "why" for each key point

**Case C: Incomplete code comments**
→ Add detailed comments: Use Scenario/Step + WHY style

**Case D: Missing execution flows**
→ Add concrete data examples: Track variable change trajectories
```

**Quick Depth Assessment Standards:**

| Chapter | Min Words | Required Elements |
|---------|-----------|-------------------|
| 1. Quick Overview | 200 | Language, scale, dependencies, type |
| 2. Background & Motivation | 400 | Problem essence, solution choice, scenarios |
| 3. Core Concepts | 600 | Each concept 3 WHYs, relationship matrix |
| 4. Algorithm & Theory | 500 | Complexity, WHY, references |
| 5. Design Patterns | 400 | Pattern name, WHY, standard reference |
| 6. Key Code Analysis | 800 | Line-by-line, execution examples, scenarios |
| 7. Test Case Analysis | 400 | Test structure, key tests, hidden behaviors (if tests available) |
| 8. Application Transfer | 500 | ≥ 2 scenarios, constant principles, modifications |
| 9. Dependencies | 300 | WHY per dependency, usage examples |
| 10. Quality Verification | 200 | Checklist, four abilities test |

**Total: Deep Mode document should be ≥ 4000 words**
登录后下载/点赞/收藏 ❤ 9 | ★ 0
code-reader-v2-en

目录结构

SKILL.md

举报内容

提示
