slop-check

slop-check is a standalone writing-audit skill for catching AI-slop patterns before publishing.

It does not try to prove whether content was generated by AI. It checks whether the writing feels generic, over-polished, templated, low-signal, or synthetic.

Version

Current version: 1.0.0

What It Checks

• Banned AI-sounding vocabulary and filler phrases
• Weak hooks and context-setting openers
• Fake balance and non-committal takes
• Over-structured bullets and template-like paragraphing
• Generic conclusions and tidy moral endings
• Unsupported precise claims or weak source trails
• Forced persona, fake sass, and synthetic authenticity
• Mechanical rhythm, staccato chains, and fake burstiness
• Low specificity, low originality, and weak human texture

When To Use It

Use this before publishing:

• X/Twitter posts and replies
• LinkedIn posts
• Blog posts and newsletters
• Landing page copy
• Essays and op-eds
• Product announcements
• Founder updates
• AI-assisted drafts that need a human quality pass

Skill Installation

Copy the slop-check folder into your skills directory.

For Codex:

mkdir -p ~/.codex/skills
cp -R slop-check ~/.codex/skills/

For Claude Code-style skill folders:

mkdir -p ~/.agents/skills
cp -R slop-check ~/.agents/skills/

Restart your agent session after installing.

Usage

Ask your agent to run slop-check on a draft:

Use slop-check on this X post:

[paste draft]

Or:

Audit this landing page copy for AI slop:

[paste copy]

Output

The skill returns:

• Tier 1 violations with exact quotes
• Second-generation slop patterns
• Three quality-gate checks
• 8-dimension scorecard
• Top rewrite fixes
• Final verdict: PASS, NEEDS REWRITE, or REWRITE

Example report shape:

SLOP CHECK REPORT

Platform:

TIER 1 VIOLATIONS
- [Exact quote] -> [why it fails]

SECOND-GENERATION PATTERNS
- [Pattern found] -> [rewrite direction]

QUALITY GATES
Gate 1 Adds value: PASS/FAIL
Gate 2 Takes position: PASS/FAIL
Gate 3 Stops scroll: PASS/FAIL

SCORES
original_take:
no_ai_tells:
voice_match:
platform_fit:
specificity:
position:
no_emotional_inflation:
humanness:
overall:

TOP FIXES
1.
2.
3.

VERDICT
PASS / NEEDS REWRITE / REWRITE

Scoring Dimensions

Dimension	What It Measures
original_take	Does the draft say something non-obvious?
no_ai_tells	Does it avoid banned phrases, structures, and formatting?
voice_match	Does it sound like a specific human or brand?
platform_fit	Does it fit the target platform and reader behavior?
specificity	Does it use names, numbers, scenes, or examples naturally?
position	Does it make a claim someone could disagree with?
no_emotional_inflation	Does it avoid fake stakes and generic drama?
humanness	Does it have lived texture, judgment, or constraint?

Verdict Thresholds

• 9-10: PASS
• 8-8.9: PASS with minor edits
• 6-7.9: NEEDS REWRITE
• Below 6: REWRITE

Core Test

The skill is built around one question:

Would a sharp human with lived context, taste, and a real stake say this exact thing?

If the answer is "maybe", "it's fine", or "it sounds professional", the draft is not done.

Fine is slop.

License

MIT

Support

If this skill saves you time or helps you publish better work, donations are welcome.

Solana wallet for SOL and ZEC-as-a-Solana-token: