copy-humanizer
Skill1 starsChanged 2 days agoMITView on GitHub
Remove AI-generated patterns and restore natural human voice to marketing copy, blog posts, and email content. Triggers on: humanize, copy humanizer, de-ai, remove ai patterns, make it human, sounds like ai, too robotic, ai voice.
We split this skill into the rules it gives an agent
9 rules. About 0 are common, 9 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈9
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 11
Core principle: Don't change the message. Strip the AI tells.
- Distinctive: line 89
Don't change the core message or argument
- Distinctive: line 90
Don't remove technical accuracy
- Distinctive: line 91
Don't dumb down -- make it natural, not simplistic
- Distinctive: line 92
Don't restructure the piece (that's a structural edit, not humanising)
- Distinctive: line 93
Don't add new content or examples
- Distinctive: line 94
Don't apply to code blocks, quotes, or technical specifications
- Distinctive: line 101
Create a backup copy with "-original" suffix
- Distinctive: line 128
Use "you" and "your team" -- talk TO the reader
≈ “Common” is an estimate: at least 1 in 5 of the 308,232 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Copy Humanizer
Strip AI fingerprints from marketing copy, blog posts, and email content. Make it read like a real person on your team wrote it.
Core principle: Don't change the message. Strip the AI tells.
Prevention beats cleanup: when drafting from scratch, instruct the model to "Use ASD-STE100" (the Simplified Technical English aircraft-manual standard: short words, short sentences, active voice) plus Zinsser's four principles (simplicity, brevity, clarity, humanity). This removes most tells before they are written. Do NOT apply STE to creative or emotive copy - it flattens anything with a heart. Keep your own banned-tells list (a plain checklist file in your project); check drafts against it and add new tells as they appear.
What Gets Removed
1. Overused Transitions
- "Moreover," "Furthermore," "Additionally," "Nevertheless" as sentence starters
- Excessive "However" usage
- "While X, Y" constructions (restructure as two sentences or use "but")
2. AI Cliches
- "In today's fast-paced world", "In the ever-evolving landscape of"
- "Let's dive deep", "Unlock your potential", "Harness the power of"
- "Game-changer", "Revolutionary", "Cutting-edge"
- "Seamless", "Robust", "Holistic"
- "Navigate the complexities of"
3. Hedging Language
- "It's important to note", "It's worth mentioning", "It's worth noting"
- "At the risk of being hyperbolic"
- Vague quantifiers: "various," "numerous," "myriad," "a plethora of"
4. Corporate Buzzwords
- "utilize" -> "use"
- "facilitate" -> "help"
- "optimize" -> "improve"
- "leverage" (non-financial) -> "use"
- "streamline" -> "simplify" or "speed up"
- "empower" -> "help" or "let"
- "elevate" -> "improve"
5. Structural AI Tells
- Em-dashes used as inline bullet structures:
term -- definition, term -- definition - Bold text for random emphasis mid-paragraph (not headers or callouts)
- Exactly three parallel examples or bullet points in a row (classic AI filler pattern)
- Rhetorical question immediately answered by the next sentence
- "Not only does it X, but it also Y" -> Just say both things directly
- "That's not X. That's Y." (two-sentence pivot) -> keep only the second half: "They're stalling."
- Stacked two-word fragments: "Fast. Simple." / "No fluff. Just answers." -> keep one fragment, cut its twin
- Triplet adjectives in prose: "faster, cheaper, smarter" -> real reasons come in twos or fives, never always three
- Double-metaphor with no advice: "less a hammer, more a scalpel" -> say what to do instead
- Vague ranges: "5 to 10 minutes" -> one real, measured number ("7 minutes"); never invent one
- Lists that feel like padding rather than adding value
6. Announcement Language
- "Here's why:", "Here's what I learned:", "Let me explain:"
- Warm-up openers: "Here's the thing", "Let me be clear", "The truth is" -> start one sentence later
- "Enter stage left:", theatrical introductions
- Standalone transition sentences that just introduce the next section ("That's where X comes in.")
- Self-applause sentences: "And that matters." / "That's the part everyone misses." / "Which is exactly the point." -> delete them all; nothing is lost
- Recap endings: "In short", "At the end of the day", any close that restates the piece -> stop typing; end on the high note
7. Em Dash Overuse
- Replace most em dashes with commas, periods, colons, or parentheses
- An em dash for an aside -> commas or parentheses
- An em dash introducing a conclusion -> colon or period
- A dramatic pause em dash -> period and new sentence
- Two em dashes bracketing a clause -> commas or parentheses
- LLMs massively overuse em dashes. Most sentences read better without them.
What Gets Added
- Varied sentence lengths (short punchy + longer explanatory)
- Direct statements instead of hedged ones
- Specific examples instead of vague claims
- Conversational connectors that real people use ("Look,", "The thing is,", "Here's the deal:")
- Your brand voice: knowledgeable but approachable, human, not corporate (see the voice guide below)
What NOT to Do
- Don't change the core message or argument
- Don't remove technical accuracy
- Don't dumb down -- make it natural, not simplistic
- Don't restructure the piece (that's a structural edit, not humanising)
- Don't add new content or examples
- Don't apply to code blocks, quotes, or technical specifications
Process
Mode 1: File Edit (when given a file path)
- Read the file
- Create a backup copy with "-original" suffix
- Make all humanising edits in-place
- Present a summary: "Made X changes across Y categories. Key changes: ..."
- Offer to show a diff
Mode 2: Interactive Review (when user says "review" or "interactive")
- Read the file
- Build numbered issue list with category tags
- Present issues one at a time: original -> suggested fix
- Accept/skip/quit loop (a/s/q)
- Apply accepted changes
- Summary of accepted vs skipped
Mode 3: Inline Text (when text is pasted or no file path)
- Take the provided text
- Apply humanising edits
- Output the cleaned text directly
- List what was changed
Your Brand Voice (customize this)
Fill this in for your own brand. The items below are a starter template - swap the examples for your own product names, spelling conventions, and stances:
- Name things specifically: say the actual product (e.g. "Google Workspace") not "the platform"
- Be direct: "This saves you 2 hours a week" not "This can potentially help optimise your workflow"
- Use "you" and "your team" -- talk TO the reader
- Contractions are fine (we're, it's, don't)
- Pick a spelling convention and hold it (e.g. UK/AU: organisation, optimise, colour; or US: organization, optimize, color) based on your audience
- OK to be opinionated -- have a point of view on the choices in your space
- Short paragraphs. Web readers scan, they don't read walls of text.
Example Transformations
Before (AI): "In today's rapidly evolving digital landscape, it's crucial to understand that leveraging Google Workspace effectively isn't just about utilizing cutting-edge technology -- it's about harnessing its transformative potential to unlock unprecedented productivity gains for your organisation."
After (human): "Google Workspace works best when you use it for specific things. Focus on what it actually does well: shared drives, collaborative docs, and admin controls that save your IT team hours every week."
Before (AI): "Moreover, it's worth noting that our comprehensive suite of managed IT services facilitates seamless integration between your existing infrastructure and cloud-based solutions, thereby empowering your team to achieve optimal operational efficiency."
After (human): "Our managed IT service connects your existing setup to the cloud. Your team keeps working the way they do now, just faster and with fewer headaches."
$ARGUMENTS
Other versions1 project uses a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 1
- shahrulestar/skills★ 1same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository36 other files it gives its agents.
- ai-debateSkill
- claude-vm-opsSkill
- cleanup-gitSkill
- codex-reviewSkill
- deploy-safelySkill
- dev-auto-buildSkill
- dev-brainstormingSkill
- dev-debuggingSkill
- dev-deploy-safelySkill
- dev-qa-verifySkill
- dev-quick-buildSkill
- dev-update-envSkill
- dev-writing-plansSkill
- document-pdfSkill
- exec-summarySkill
- fable-reviewSkill
- gemini-reviewSkill
- machine-moveSkill
- morning-briefSkill
- plugin-fresh-install-testSkill
- quad-reviewSkill
- repo-cleanupSkill
- repo-context-cleanupSkill
- repo-health-sweepSkill
- review-debateSkill
- review-swarmSkill
- review-with-claudeSkill
- review-with-codexSkill
- review-with-geminiSkill
- review-with-glmSkill
- session-dspSkill
- session-recoverySkill
- session-status-sweepSkill
- session-wrapSkill
- triple-reviewSkill
- zai-reviewSkill
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
Reviews can't be read right now.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

