Slop Radar

Slop Radar

Scan text for AI-generated writing patterns before finalizing output.

When to Use

• After generating documentation, READMEs, blog posts, or any prose
• When reviewing or editing written content
• Before committing markdown files
• When user asks to "check for slop", "check quality", or "make it sound human"
• Automatically after creating any substantial text output (500+ words)

How It Works

1. Buzzword scan: Check against 245 English and 127 German AI phrases
2. Structural patterns: Detect em-dash abuse, "Let me" starters, bullet overload, passive voice density, triple-adjective stacking, hedge phrases
3. Score: 0-100 scale (100 = fully human, 0 = pure AI slop)

Instructions

When triggered, analyze the text (from $ARGUMENTS file path or the most recently generated prose):

Step 1: Identify the text to check

If $ARGUMENTS is a file path, read that file. Otherwise, check the most recently generated text output in the conversation.

Step 2: Run the slop scan

Check the text against these pattern categories:

Buzzwords (deduct 2 points each): "dive deep", "transformative", "journey", "landscape", "leverage", "cutting-edge", "holistic", "empower", "stakeholders", "synergy", "unprecedented", "robust", "streamline", "innovative", "paradigm", "ecosystem", "scalable", "game-changer", "best-in-class", "actionable", "at the end of the day", "in today's fast-paced", "it's worth noting", "moreover", "furthermore", "crucial", "pivotal", "seamless", "comprehensive", "elevate", "foster", "harness", "spearhead", "drive", "unlock", "reimagine", "navigate the complexities", "delve into", "underscore"

Structural Patterns (deduct 3-5 points each):

• Sentences starting with "Let me..." or "Here's the thing..."
• Em-dash abuse (more than 1 per 200 words)
• Triple bullet point lists where a paragraph would work
• Passive voice density above 30%
• Paragraphs ending with punchy one-liners
• "Not X -- it's Y" contrast structures
• Rhetorical question followed by immediate answer

Bonuses (add points):

• +5 for natural sentence length variation
• +5 for concrete numbers, names, or specific examples
• +3 for conversational tone without being forced

Step 3: Calculate and report score

Score: [X]/100  [RATING]

Rating scale:
  90-100  HUMAN          Clean, natural writing
  70-89   MOSTLY CLEAN   Minor AI signals
  50-69   SUSPICIOUS     Multiple AI patterns
  30-49   LIKELY AI      Strong AI writing signals
  0-29    PURE SLOP      Heavy buzzword and pattern use

Step 4: Show flagged items with replacements

For each flagged buzzword or pattern, suggest a concrete replacement:

| Found | Replacement | |-------|-------------| | "leverage" | "use" | | "cutting-edge" | (name the specific technology) | | "stakeholders" | (name the actual people: "customers", "engineers") | | "it's worth noting that" | (delete -- just state the thing) | | "dive deep into" | "look at" or "examine" | | "transformative" | (describe the actual change) | | "comprehensive" | (be specific about what it covers) | | "seamless" | (describe the actual experience) | | "robust" | (state what makes it reliable) | | "innovative" | (describe what is actually new) |

Step 5: Offer to rewrite

If score is below 70, offer to rewrite the flagged sections with concrete, specific language. Apply these principles:

• Replace vague adjectives with specific facts
• Replace buzzwords with plain words
• Break formulaic structures
• Add concrete examples where generalities exist
• Use active voice with named subjects

Example Output

Slop Radar Results
------------------
Score: 42/100  LIKELY AI

Buzzwords (8 found):
  "transformative", "leverage", "cutting-edge", "holistic",
  "stakeholders", "ecosystem", "seamless", "robust"

Patterns (3 found):
  - Let-me starter (line 1)
  - Em-dash overuse (4 in 200 words)
  - Punchy one-liner ending (line 12)

Suggested rewrites provided for 11 items.
Rewrite flagged sections? [y/n]

Integration

For automated checking, slop-radar is available as a CLI tool:

npx slop-radar check     # Full analysis
npx slop-radar score     # Score only
npx slop-radar json      # Machine-readable output

License

MIT

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: renefichtmueller
• Source: renefichtmueller/slop-radar
• License: MIT
• Homepage: https://slop-radar-demo.pages.dev

Install and usage instructions live in the source repository linked above.