Metadata

name: auto-tagger
description: Automatically detect and add [TAG-AICODE] or [TAG-AICHECK] annotations to AI-generated code blocks. Use when you need to identify AI-generated code in existing files and add proper compliance tags according to nl-coder-reviwer standards.

Auto Tagger

Automatically detects AI-generated code blocks and adds compliant [TAG-AICODE] or [TAG-AICHECK] annotations following nl-coder-reviwer standards.

When to Use This Skill

Use this skill when:

Analyzing existing codebases to identify AI-generated portions
Preparing code for compliance review with nl-coder-reviwer
Migrating legacy code to meet tagging requirements
Batch processing multiple files for AI code audit
Establishing AI code traceability in projects

Supported Tag Formats

The skill supports two tag formats from nl-coder-reviwer:

[TAG-AICODE] - General AI-generated code annotation
- Format: AI GENERATED CODE START [TAG-AICODE]
- Multiple blocks allowed per file
- No timestamp required
[TAG-AICHECK] - Formal code review annotation
- Format: AI CHECK START YYYY-MM-DD HH:MM:SS TAG-AICHECK
- Single pair per file (multiple = ERROR)
- Timestamp mandatory and must match START/END

Core Capabilities

1. AI Detection Engine

Uses multiple detection strategies in parallel:

AI Model Analysis (primary): LLM-based code analysis with confidence scoring
Pattern Detection: Code pattern matching for common AI characteristics
Git History Analysis: Commit message and author analysis
Heuristic Rules: Code complexity and formatting analysis

2. Confidence Fusion

Combines multiple detector results using weighted scoring:

AI Model: 50% weight
Pattern Detection: 20% weight
Git Analysis: 15% weight
Heuristic Rules: 15% weight

3. Smart Tagging

Automatically selects appropriate tag type:

TAG-AICHECK: For reviewed, production-critical code
TAG-AICODE: For general AI-generated code blocks

4. Language Support

Supports multiple programming languages with proper comment syntax:

Python (#)
JavaScript/TypeScript (//)
SQL (--)
Java (//, /* */)
Go (//)
And more

Quick Start

Basic Usage

# Scan single file (preview mode)
auto-tagger scan file.py --preview

# Scan single file and modify in-place
auto-tagger scan file.py --inplace

# Scan directory recursively
auto-tagger scan src/ --recursive --inplace

# Generate diff instead of modifying
auto-tagger scan file.py --diff

Configuration

Create .auto-tagger.yml in project root:

detection:
  threshold: 0.7          # Confidence threshold (0.0-1.0)
  min_block_size: 5         # Minimum lines for AI block
  skip_test_files: true
  skip_vendor: true

detectors:
  ai_model:
    enabled: true
    model: "gpt-4"
    timeout: 30

  pattern:
    enabled: true

  git:
    enabled: true
    max_commits: 50

  heuristic:
    enabled: true

tagging:
  aicode:
    enabled: true
    add_metadata: true

  aicheck:
    enabled: true
    timestamp_format: "YYYY-MM-DD HH:MM:SS"

output:
  format: "inplace"      # inplace | preview | diff
  backup: true
  verbose: true

Workflow

Step 1: Analysis

When invoked with scan, the skill:

Reads target file(s)
Parses code structure (functions, classes, blocks)
Runs all enabled detectors in parallel
Collects detection results

Step 2: Confidence Scoring

For each code block:

Combines detector results
Calculates weighted confidence score
Determines if confidence >= threshold
Selects appropriate tag type

Step 3: Tag Insertion

For blocks meeting threshold:

Generates proper tag format
Inserts START tag before block
Inserts END tag after block
Preserves code structure and formatting

Step 4: Validation

After tagging:

Validates tag format compliance
Checks for nesting violations
Verifies tag pairs match
Reports any issues

Task-Based Operations

Scan Files

Detect and tag AI code in files:

# Preview changes
auto-tagger scan path/to/file.py --preview

# Apply changes with backup
auto-tagger scan path/to/file.py --inplace --backup

# Generate unified diff
auto-tagger scan path/to/file.py --diff > changes.diff

Output format:

============================================================
AI DETECTION RESULTS: file.py
============================================================

Block 1: Lines 5-15
  Confidence: 0.85
  Tag Type: TAG-AICODE
  Reason: High-confidence AI model detection, generic variable names

📝 Proposed Tag:
  AI GENERATED CODE START [TAG-AICODE]
  <code block>
  AI GENERATED CODE END [TAG-AICODE]

Batch Processing

Process multiple files or directories:

# Recursive directory scan
auto-tagger batch scan src/ --recursive --inplace

# Specific file types only
auto-tagger batch scan src/ --extensions py,js,ts

# Parallel processing (faster)
auto-tagger batch scan src/ --jobs 4 --inplace

Validate Existing Tags

Check existing tags for compliance:

# Validate single file
auto-tagger validate file.py

# Validate directory
auto-tagger validate src/ --recursive

# Strict mode (treat warnings as errors)
auto-tagger validate src/ --strict

Remove Tags

Remove all AI tags from files:

# Remove tags with preview
auto-tagger remove file.py --preview

# Remove tags permanently
auto-tagger remove file.py --inplace

# Remove specific tag type
auto-tagger remove file.py --tag-type aicode --inplace

Detection Strategies

AI Model Detection

Uses LLM to analyze code for AI generation patterns:

Strengths:

Highest accuracy for complex logic
Context-aware analysis
Detailed reasoning

Configuration:

detectors:
  ai_model:
    enabled: true
    model: "gpt-4"           # or local model
    temperature: 0.0             # Deterministic output
    timeout: 30

Pattern Detection

Identifies common AI code patterns:

Detected patterns:

Generic variable names (item1, item2, data, result)
Standardized comment style
Perfect code formatting
Missing human-typical errors
Over-optimized algorithms
Consistent indentation

Configuration:

detectors:
  pattern:
    enabled: true
    rules:
      - generic_names
      - perfect_formatting
      - missing_errors
      - over_optimization

Git History Analysis

Analyzes commit history for AI indicators:

Detected indicators:

Commit keywords: "generated", "copilot", "ai", "assistant"
Automated commit authors
Unusual commit patterns
Large code dumps in single commit

Configuration:

detectors:
  git:
    enabled: true
    max_commits: 50
    keywords:
      - "generated"
      - "copilot"
      - "ai"
      - "assistant"
      - "gpt"

Heuristic Rules

Code quality and complexity analysis:

Rules:

Cyclomatic complexity thresholds
Comment-to-code ratio
Function length analysis
Import pattern analysis

Configuration:

detectors:
  heuristic:
    enabled: true
    complexity_threshold: 8
    max_function_length: 50
    min_comment_ratio: 0.1

Tag Type Selection

The skill automatically chooses between TAG-AICODE and TAG-AICHECK based on:

TAG-AICHECK Selection Criteria

Code is in production/backend systems
High-availability or security-critical code
Has been reviewed/validated
Formal compliance required

TAG-AICODE Selection Criteria

General AI-generated code
Development/experimental code
Multiple AI blocks in same file
No formal review required

Manual Override

User can force specific tag type:

# Force AICODE
auto-tagger scan file.py --tag-type aicode

# Force AICHECK
auto-tagger scan file.py --tag-type aicheck

Advanced Features

Custom Detector Scripts

Add custom detection logic:

# scripts/custom_detector.py
from detectors.base import BaseDetector

class CustomDetector(BaseDetector):
    def detect(self, code: str, context: dict) -> DetectionResult:
        # Your custom logic
        pass

Reference from config:

detectors:
  custom:
    enabled: true
    script: "scripts/custom_detector.py"
    weight: 0.3

Interactive Mode

Confirm each tagging decision:

auto-tagger scan file.py --interactive

# For each detected block:
# Block at lines 5-15, confidence 0.85
# Reason: High-confidence AI model detection
# Add TAG-AICODE? [y/N] y

Confidence Tuning

Adjust detection sensitivity:

# Conservative (high threshold, fewer tags)
auto-tagger scan file.py --threshold 0.85

# Aggressive (low threshold, more tags)
auto-tagger scan file.py --threshold 0.55

Output Examples

Preview Mode

$ auto-tagger scan process_data.py --preview

============================================================
AI DETECTION RESULTS: process_data.py
============================================================

Block 1: Lines 8-18
  Confidence: 0.82
  Tag Type: TAG-AICODE
  Reason: Generic variable names, perfect formatting

--- PREVIEW ---
AI GENERATED CODE START [TAG-AICODE]
# Generated by: Auto Tagger AI Model
# Detection confidence: 0.82
# Date: 2026-02-10
def process_data(items):
    result = []
    for item in items:
        transformed = item.upper().replace('_', ' ')
        result.append(transformed)
    return result
AI GENERATED CODE END [TAG-AICODE]
---

Total blocks: 1
AI code lines: 11 / 25 (44%)
Estimated accuracy: 82%

Apply changes? [y/N]

Diff Mode

$ auto-tagger scan process_data.py --diff

diff --git a/process_data.py b/process_data.py
index abc123..def456 100644
--- a/process_data.py
+++ b/process_data.py
@@ -5,0 +5,12 @@
+AI GENERATED CODE START [TAG-AICODE]
+# Generated by: Auto Tagger AI Model
+# Detection confidence: 0.82
+# Date: 2026-02-10
 def process_data(items):
     result = []
     for item in items:
         transformed = item.upper().replace('_', ' ')
         result.append(transformed)
     return result
+AI GENERATED CODE END [TAG-AICODE]

Troubleshooting

Low Confidence Scores

If tags aren't being added:

Lower threshold:

auto-tagger scan file.py --threshold 0.6

Enable more detectors:

detectors:
  pattern:
    enabled: true
  heuristic:
    enabled: true

Check AI model access:
- Verify API key is configured
- Check network connectivity
- Try with alternative model

False Positives

If human code is tagged as AI:

Use preview mode first:
```
auto-tagger scan file.py --preview
```

Increase threshold:

auto-tagger scan file.py --threshold 0.85

Manual review with interactive mode:
```
auto-tagger scan file.py --interactive
```

Performance Issues

For large codebases:

Use parallel processing:
```
auto-tagger batch scan src/ --jobs 4
```

Limit file types:

auto-tagger batch scan src/ --extensions py,js

Disable heavy detectors:

detectors:
  ai_model:
    enabled: false  # Skip LLM analysis
  pattern:
    enabled: true

Integration with nl-coder-reviwer

After tagging, validate with nl-coder-reviwer:

# Tag code with auto-tagger
auto-tagger scan src/ --recursive --inplace

# Validate tags
cd /path/to/nl-coder-reviwer
python3 nlcr.py validate ../src/ --directory --recursive

References

Tag Standards: See tag_standards.md for complete tag format specifications
Detection Patterns: See detection_patterns.md for detection algorithm details
Usage Guide: See usage_guide.md for comprehensive usage examples

Notes

This skill is designed to assist, not replace, human judgment
Always review detected blocks before applying changes
Use --preview or --interactive mode for critical code
Backup files are automatically created when using --inplace
Git-aware: Skips .git/ directory by default

Auto Tagger

安装技能

Auto Tagger

When to Use This Skill

Supported Tag Formats

Core Capabilities

1. AI Detection Engine

2. Confidence Fusion

3. Smart Tagging

4. Language Support

Quick Start

Basic Usage

Configuration

Workflow

Step 1: Analysis

Step 2: Confidence Scoring

Step 3: Tag Insertion

Step 4: Validation

Task-Based Operations

Scan Files

Batch Processing

Validate Existing Tags

Remove Tags

Detection Strategies

AI Model Detection

Pattern Detection

Git History Analysis

Heuristic Rules

Tag Type Selection

TAG-AICHECK Selection Criteria

TAG-AICODE Selection Criteria

Manual Override

Advanced Features

Custom Detector Scripts

Interactive Mode

Confidence Tuning

Output Examples

Preview Mode

Diff Mode

Troubleshooting

Low Confidence Scores

False Positives

Performance Issues

Integration with nl-coder-reviwer

References

Notes