anti-slop（防坡度）

Anti-Slop Skill

Detect and eliminate generic AI-generated patterns ("slop") across natural language, code, and design.

What is AI Slop?

AI slop refers to telltale patterns that signal low-quality, generic AI-generated content:

• Text: Overused phrases like "delve into," excessive buzzwords, meta-commentary
• Code: Generic variable names, obvious comments, unnecessary abstraction
• Design: Cookie-cutter layouts, generic gradients, overused visual patterns

This skill helps identify and remove these patterns to create authentic, high-quality content.

When to Use This Skill

Apply anti-slop techniques when:

• Reviewing AI-generated content before delivery
• Creating original content and want to avoid generic patterns
• Cleaning up existing content that feels generic
• Establishing quality standards for a project
• User explicitly requests slop detection or cleanup
• Content has telltale signs of generic AI generation

Core Workflow

1. Detect Slop

For text files:

python scripts/detect_slop.py <file> [--verbose]

This analyzes text and provides:

• Slop score (0-100, higher is worse)
• Specific pattern findings
• Actionable recommendations

Manual detection: Read the appropriate reference file for detailed patterns:

• references/text-patterns.md - Natural language slop patterns
• references/code-patterns.md - Programming slop patterns
• references/design-patterns.md - Visual/UX design slop patterns

2. Clean Slop

Automated cleanup (text only):

# Preview changes
python scripts/clean_slop.py <file>

# Apply changes (creates backup)
python scripts/clean_slop.py <file> --save

# Aggressive mode (may slightly change meaning)
python scripts/clean_slop.py <file> --save --aggressive

Manual cleanup: Apply strategies from the reference files based on detected patterns.

Text Slop Detection & Cleanup

High-Priority Targets

Remove immediately:

• "delve into" → delete or replace with "examine"
• "navigate the complexities" → "handle" or delete
• "in today's fast-paced world" → delete
• "it's important to note that" → delete
• Meta-commentary about the document itself

Simplify wordy phrases:

• "in order to" → "to"
• "due to the fact that" → "because"
• "has the ability to" → "can"

Replace buzzwords:

• "leverage" → "use"
• "synergistic" → "cooperative"
• "paradigm shift" → "major change"

Quality Principles

Be direct:

• Skip preambles and meta-commentary
• Lead with the actual point
• Cut transition words that don't add meaning

Be specific:

• Replace generic terms with concrete examples
• Name specific things instead of "items," "things," "data"
• Use precise verbs instead of vague action words

Be authentic:

• Vary sentence structure and length
• Use active voice predominantly
• Write in a voice appropriate to context, not corporate-generic

Code Slop Detection & Cleanup

High-Priority Targets

Rename generic variables:

• data → name what data it represents
• result → name what the result contains
• temp → name what you're temporarily storing
• item → name what kind of item

Remove obvious comments:

# Bad
# Create a user
user = User()

# Better - let code speak
user = User()

Simplify over-engineered code:

• Remove unnecessary abstraction layers
• Replace design patterns used without purpose
• Simplify complex implementations of simple tasks

Improve function names:

• handleData() → what are you doing with data?
• processItems() → what processing specifically?
• manageUsers() → what management action?

Quality Principles

Clarity over cleverness:

• Write code that's easy to understand
• Optimize only when profiling shows need
• Prefer simple solutions to complex ones

Meaningful names:

• Variable names should describe content
• Function names should describe action + object
• Class names should describe responsibility

Appropriate documentation:

• Document why, not what
• Skip documentation for self-evident code
• Focus documentation on public APIs and complex logic

Design Slop Detection & Cleanup

High-Priority Targets

Visual slop:

• Generic gradient backgrounds (purple/pink/cyan)
• Overuse of glassmorphism or neumorphism
• Floating 3D shapes without purpose
• Every element using same design treatment

Layout slop:

• Template-driven layouts ignoring content needs
• Everything in cards regardless of content type
• Excessive whitespace without hierarchy
• Center-alignment of all elements

Copy slop:

• "Empower your business" type headlines
• Generic CTAs like "Get Started" without context
• Buzzword-heavy descriptions
• Stock photo aesthetics

Quality Principles

Content-first design:

• Design around actual content needs
• Create hierarchy based on importance
• Let content determine layout, not templates

Intentional choices:

• Every design decision should be justifiable
• Use patterns because they serve users, not because they're trendy
• Vary visual treatment based on element importance

Authentic voice:

• Copy should reflect brand personality
• Avoid generic marketing speak
• Be specific about value proposition

Reference Files

Consult these comprehensive guides when working on specific domains:

• text-patterns.md - Complete catalog of natural language slop patterns with detection rules and cleanup strategies
• code-patterns.md - Programming antipatterns across languages with refactoring guidance
• design-patterns.md - Visual and UX design slop patterns with improvement strategies

Each reference includes:

• Pattern definitions and examples
• Detection signals (high/medium confidence)
• Context where patterns are acceptable
• Specific cleanup strategies

Scripts

detect_slop.py

Analyzes text files for AI slop patterns.

Usage:

python scripts/detect_slop.py <file> [--verbose]

Output:

• Overall slop score (0-100)
• Category-specific findings
• Line numbers and examples
• Actionable recommendations

Scoring:

• 0-20: Low slop (authentic writing)
• 20-40: Moderate slop (some patterns)
• 40-60: High slop (many patterns)
• 60+: Severe slop (heavily generic)

clean_slop.py

Automatically removes common slop patterns from text files.

Usage:

# Preview changes
python scripts/clean_slop.py <file>

# Save changes (creates backup)
python scripts/clean_slop.py <file> --save

# Save to different file
python scripts/clean_slop.py <file> --output clean_file.txt

# Aggressive mode
python scripts/clean_slop.py <file> --save --aggressive

What it cleans:

• High-risk phrases
• Wordy constructions
• Meta-commentary
• Excessive hedging
• Buzzwords
• Redundant qualifiers
• Empty intensifiers

Safety:

• Always creates .backup file when overwriting
• Preview mode shows changes before applying
• Preserves content meaning (non-aggressive mode)

Best Practices

Prevention Over Cure

When creating content:

1. Write with specific audience in mind
2. Use concrete examples over abstractions
3. Lead with the point, skip preambles
4. Choose words for precision, not impression
5. Review before considering it complete

Context-Aware Cleanup

Not all patterns are always slop:

Acceptable contexts:

• Academic writing may need more hedging
• Legal documents require specific phrasing
• Internal documentation can use shortcuts
• Technical docs have domain-specific conventions

Always consider:

• Who is the audience?
• What is the purpose?
• Does this pattern serve a function?
• Is there a better alternative?

Iterative Improvement

1. Detect - Run detection scripts or manual review
2. Analyze - Understand which patterns are truly problems
3. Clean - Apply automated cleanup where safe
4. Review - Manually verify changes maintain meaning
5. Refine - Fix remaining issues by hand

Quality Over Automation

The scripts are tools, not replacements for judgment:

• Use automated detection to find candidates
• Apply automated cleanup to obvious patterns
• Manually review anything that changes meaning
• Exercise discretion based on context

Integration Patterns

Code Review

# Check files before committing
python scripts/detect_slop.py src/documentation.md --verbose

# Clean up automatically
python scripts/clean_slop.py src/documentation.md --save

Content Pipeline

1. Create initial content
2. Run slop detection
3. Apply automated cleanup
4. Manual review and refinement
5. Final quality check

Standards Enforcement

Create project-specific thresholds:

• Max acceptable slop score: 30
• Required manual review for scores > 20
• Auto-reject submissions with scores > 50

Limitations

Scripts only handle text:

• Code slop detection is manual (use code-patterns.md)
• Design slop detection is manual (use design-patterns.md)

Context sensitivity:

• Scripts can't understand all contexts
• Some "slop" may be appropriate in certain domains
• Always review automated changes

Language coverage:

• Detection patterns optimized for English
• Code patterns focus on common languages (Python, JS, Java)
• Design patterns are platform-agnostic

Common Scenarios

Scenario 1: Review AI-Generated Content

# User asks: "Can you review this article for AI slop?"
1. Read references/text-patterns.md for patterns to watch
2. Run: python scripts/detect_slop.py article.txt --verbose
3. Review findings and apply manual cleanup
4. Optionally run: python scripts/clean_slop.py article.txt --save
5. Do final manual review of cleaned content

Scenario 2: Clean Up Codebase

# User asks: "Help me clean up generic AI patterns in my code"
1. Read references/code-patterns.md
2. Review code files manually for patterns
3. Create list of generic names to rename
4. Refactor following principles in code-patterns.md
5. Remove obvious comments and over-abstractions

Scenario 3: Design Review

# User asks: "Does this design look too generic?"
1. Read references/design-patterns.md
2. Check against high-confidence slop indicators
3. Identify specific issues (gradients, layouts, copy)
4. Provide specific recommendations from design-patterns.md
5. Suggest concrete alternatives

Scenario 4: Establish Quality Standards

# User asks: "Help me create quality standards for our team"
1. Review all three reference files
2. Identify patterns most relevant to user's domain
3. Create project-specific guidelines
4. Set up detection scripts in development pipeline
5. Document acceptable exceptions

Tips for Success

For text cleanup:

• Run detection first to understand scope
• Use non-aggressive mode for important content
• Always review automated changes
• Focus on high-risk patterns first

For code cleanup:

• Start with renaming generic variables
• Remove obvious comments next
• Refactor over-engineered code last
• Test after each significant change

For design cleanup:

• Audit visual elements against patterns
• Prioritize structural issues over aesthetic ones
• Ensure changes serve user needs
• Maintain brand consistency

General principles:

• Quality > uniformity
• Context > rules
• Clarity > cleverness
• Specificity > generality