Unslop

Comprehensive skill for humanizing AI-generated content. Two-pass system: diagnosis → reconstruction with validation.

When to Use

Apply this skill when:

• User asks to “humanize” or “de-slop” text
• Editing AI-generated drafts, emails, articles, or social posts
• Text contains obvious AI patterns (throat-clearing, binary contrasts, emphasis crutches)
• User says text “sounds like AI” or “sounds robotic”
• Reviewing content before publishing
• User asks to “make it sound more natural” or “like a human wrote it”

Arguments

Voice Presets

Workflow

Pass 1: Diagnosis

Before rewriting, analyze the input:

1. Read reference materials from this skill’s directory:

   • references/taboo-phrases.md – banned phrases and patterns
   • references/rubric.md – 8 scoring criteria
   • references/fact-preservation.md – what must survive
   • references/personality-guide.md – voice and personality guidance
   • references/edit-library.md – before/after transformation examples

2. Extract constraints (facts that must survive):

   python3 <skill-path>/scripts/extract_constraints.py <<< "$INPUT"
   

3. Scan for AI-isms:

   python3 <skill-path>/scripts/banned_phrase_scan.py <<< "$INPUT"
   

4. Read the selected preset from presets/ directory

5. Identify: audience, content type, tone target

Pass 2: Reconstruction

Rewrite following these rules:

1. Eliminate all AI-isms cataloged in taboo-phrases.md
2. Replace all em-dashes with periods, commas, or parentheses (zero em-dashes is the target)
3. Follow preset voice characteristics
4. Preserve all constraints (numbers, names, dates, URLs)
5. Apply rubric criteria:
   • Directness: No hedging, no throat-clearing
   • Natural rhythm: Vary sentence length (8-25 words)
   • Concrete verbs: Specific actions, not abstractions
   • Reader trust: No over-explaining
   • Human authenticity: No performative emphasis
   • Content density: Substance over filler
   • Fact preservation: All facts intact
   • Template avoidance: No AI structural clichés

Pass 3: Validation

After rewriting, validate:

1. Fact preservation:

   python3 <skill-path>/scripts/validate_preservation.py original.txt transformed.txt
   

2. Remaining AI-isms:

   python3 <skill-path>/scripts/banned_phrase_scan.py <<< "$OUTPUT"
   

3. Readability metrics:

   python3 <skill-path>/scripts/readability_metrics.py <<< "$OUTPUT"
   

4. Change percentage (flag if >40%):

   python3 <skill-path>/scripts/diff_check.py original.txt transformed.txt
   

5. Score against rubric (8 criteria × 5 points = 40 max)

Output Format

## Transformed Text

[The humanized version]

## Validation

- Constraints: [X]/[Y] preserved
- AI-isms: [N] remaining
- Readability: Grade [X], variance [Y]
- Change: [X]% from original
- Score: [X]/40

## Changes Made

- [List of major transformations]

Quick Examples

Input:

Here’s the thing: building products is hard. Not because the technology is complex. Because people are complex. Let that sink in.

Output (crisp):

Building products is hard. Not the technology. The people.

Input:

In today’s fast-paced business environment, it’s becoming increasingly important for organizations to leverage their core competencies while navigating the complex landscape of digital transformation.

Output (crisp):

Companies need to use their strengths while going digital.

Common AI Patterns to Eliminate

Em-Dash Overuse (HIGH PRIORITY)

Em-dashes are the most reliable AI punctuation tell. Default to zero.

• Replace em-dashes with periods, commas, or parentheses
• Never allow multiple em-dashes per paragraph
• If one is absolutely necessary, max one per several paragraphs

Throat-Clearing Openers

• “Here’s the thing:” / “Here’s why” / “Let’s dive in” / “Let’s unpack”
• “The uncomfortable truth is” / “It’s no secret that”
• “Let me be clear” / “It turns out”

Emphasis Crutches

• “Full stop.” / “Let that sink in.” / “Make no mistake”
• “Buckle up” / “Food for thought” / “It’s a no-brainer”

Binary Contrasts

• “Not because X. Because Y.” → State Y directly
• “X isn’t the problem. Y is.” → “The problem is Y”

Business Jargon

Significance Inflation

• “stands as a testament to” → state the fact
• “pivotal moment” → be specific about what happened
• “rich tapestry” → never use
• “groundbreaking” → name the specific first
• “speaks volumes” / “sends a clear message” / “raises the bar” → just state the fact

Promotional Language

• “nestled in the heart of” → give the actual address
• “boasts a” → “has a”
• “world-class” / “state-of-the-art” → describe specifically
• “a hidden gem” → cut
• “at the forefront of” → “leading”

Superficial -ing Analyses

• “, highlighting…” / “, showcasing…” / “, underscoring…” → delete or give actual analysis in its own sentence

Communication Artifacts

• “I hope this helps” / “Certainly!” / “Great question!” → cut (chatbot residue)

Filler Setups

• “The key takeaway:” / “The bottom line:” / “It’s clear that” → just state the point
• “At the intersection of” / “In an era of” → cut entirely
• “Pro tip:” / “Hot take:” / “Unpopular opinion:” → cut (meta-commentary)

AI Vocabulary

• delve, garner, interplay, intricate, tapestry, underscore, multifaceted, paramount, burgeoning
• resonates, sheds light, strikes a balance, paints a picture, double-edged sword

See references/taboo-phrases.md for the complete list (~200 phrases across 24 categories).

Scoring Rubric

Passing score: 32/40 (80%)

Reference Files

Located in this skill’s directory:

Maintenance Commands

Wiki Sync (/unslop --wiki-sync)

This command syncs the skill’s pattern rules with Wikipedia’s Signs of AI writing page. Run it periodically to pick up new patterns added by Wikipedia editors.

Workflow — execute these steps in order:

1. Check for updates:

   python3 <skill-path>/scripts/wiki_sync.py check
   

   If exit code is 0, report “No updates” and stop.

2. Get structured diff:

   python3 <skill-path>/scripts/wiki_sync.py diff
   

   Parse the JSON output. Each change has type, section, words.

3. For each change with new words/phrases, apply updates:

   • Read references/taboo-phrases.md — add new phrases to the matching section (use the section mapping in the diff output). Skip phrases that already exist.
   • Read scripts/banned_phrase_scan.py — add corresponding entries to BANNED_PHRASES dict with appropriate category, severity (“hard” for clear AI tells, “soft” for context-dependent), and suggestion.
   • If a change warrants a new before/after example, add it to references/edit-library.md.

4. Verify:

   python3 <skill-path>/scripts/banned_phrase_scan.py < /dev/null
   

   Confirm no syntax errors. Report what was added.

Important: Only add phrases that are genuine AI writing tells applicable to general prose. Skip Wikipedia-specific patterns (broken wikitext, DOI issues, category errors, etc.) that don’t apply outside Wikipedia.

Key Principles

1. Diagnosis before writing – Understand violations before fixing
2. Facts are sacred – Never sacrifice accuracy for style
3. Presets guide, don’t constrain – Adapt to content
4. When in doubt, cut – Shorter is almost always better
5. Validation is mandatory – Run the scripts