Humanize Text

Humanize AI-generated text by removing telltale AI writing patterns. Use when text needs to sound natural and human-written — removing em-dashes, AI filler p...

Security Scan

The skill's code and instructions align with its stated purpose (removing AI writing patterns); it has no network or credential demands and no install step, so there are no obvious security mismatches.

Purpose & Capability

Name, description, SKILL.md, and the included scripts all describe the same functionality (mechanical pattern removal and light stylistic fixes). There are no unrelated environment variables, binaries, or capabilities requested.

Instruction Scope

Runtime instructions only read the target text, apply deterministic transformations, show diffs, and write results; they do not instruct the agent to read other system files, access credentials, or send data externally. The file-writing behavior is expected for a file-processing utility.

Install Mechanism

No install spec is provided (instruction-only) and the repository includes a small Python script. No downloads, package installs, or external installers are invoked.

Credentials

The skill requires no environment variables, no credentials, and no config paths. The script operates locally on input text files only.

Persistence & Privilege

always is false and the skill does not request persistent or elevated privileges or attempt to modify other skills or system-wide settings.

Assessment

This skill appears coherent and low-risk from a security perspective: it runs local, deterministic regex/string transformations and does not exfiltrate data or require credentials. Before using it, (1) review and back up any files you will process — the script will rewrite files if you direct it to, (2) test on copies in 'light' or 'preserve' mode to confirm the substitutions don't break technical terms or domain-specific phrasing, (3) be aware regex-based replacements can be overbroad (check capitalization fixes and phrase removals), and (4) if you need stricter safety, inspect or run the included scripts in a sandboxed environment. If you want higher assurance, request a code review of scripts/humanize.py for any domain-specific exceptions or further safeguards (whitelists for technical vocabulary).

Strip AI writing patterns from text to make it read like a human wrote it. Not a rewriter — a pattern remover and style fixer.

Quick Start

For inline text in chat, apply the rules below directly. For files, read the file, apply transformations, write it back.

Core Transformations

Apply ALL of these in order:

1. Punctuation Fixes

• Replace ALL em-dashes (—) with commas, periods, colons, or parentheses as context requires
• Replace en-dashes (–) used as em-dashes the same way
• Remove excessive exclamation marks (max 1 per paragraph)
• Kill semicolons in casual writing — rewrite as two sentences
• Remove colons from bullet point headers (write "Thing does X" not "Thing: does X")

2. Dead Phrases (Delete or Rewrite)

These phrases scream AI. Kill on sight:

Filler/hedging: delve into, it's important to note, it's worth noting, it bears mentioning, let's dive in, take a dive into, at its core, in today's [anything], in the world of, in the realm of, when it comes to, at the end of the day, the bottom line is, as we navigate, first and foremost

Puffery/significance: a testament to, game-changer, groundbreaking, revolutionary, cutting-edge, landscape (as metaphor), tapestry, bustling, embark, beacon of, cornerstone of, paradigm shift, pivotal, multifaceted, comprehensive, robust, leverage (as verb), utilize (use "use"), facilitate, foster, cultivate, spearheading, unparalleled, myriad

Transition filler: in conclusion, in summary, to summarize, moreover, furthermore, additionally, subsequently, consequently, nevertheless, nonetheless, henceforth, thus, hence, thereby, overall, ultimately

Sycophantic openers: great question, that's a great point, absolutely, certainly, indeed, of course, definitely, I'd be happy to, I'm glad you asked, what a fascinating, excellent question, good thinking

Fake empathy: I understand your concern, I completely understand, I hear you, that must be frustrating, I can see why you'd, I appreciate you sharing

Meta-commentary: let me explain, let me break this down, here's the thing, here's what you need to know, the short answer is, to put it simply, in other words, simply put, think of it this way, imagine this scenario

3. Structural Tells

Sentence patterns to break:

• "Not X, but Y" / "It's not about X, it's about Y" — overused AI rhetorical structure, rewrite naturally
• "[Word]. [Word]. [Word]." — staccato three-word fragments used for false drama
• "The result? [Answer]." / "And the X? Y." — fake Q&A structure
• Starting 3+ consecutive sentences the same way
• Paragraphs that all follow: claim → evidence → significance

Structural fixes:

• Vary sentence length (AI defaults to medium-length everything)
• Mix simple and compound sentences (AI over-compounds with commas)
• Start some sentences with "And", "But", "So" — AI avoids this
• Use contractions (don't, won't, can't, it's) — AI under-uses them
• Occasionally use fragments. Like this. AI hates fragments.
• Remove the triple-structure pattern: "X, Y, and Z" appearing repeatedly

4. Formatting Tells

• Remove excessive bold (every other word bolded)
• Don't start every bullet point with a bolded label
• Remove "Key takeaways:" sections
• Kill numbered lists when bullets or prose work better
• Remove emoji used as bullet point decoration (✅, 🔑, 💡, 🎯)
• Headers shouldn't all be questions

5. Vocabulary Swaps

AI Word	Human Word
utilize	use
leverage	use / take advantage of
facilitate	help / enable
implement	build / set up / do
comprehensive	full / complete / thorough
robust	strong / solid
streamline	simplify / speed up
optimize	improve
innovative	new / clever
seamless	smooth
endeavor	try / effort
subsequently	then / after
commence	start / begin
numerous	many / a lot of
sufficient	enough
prior to	before
in order to	to
due to the fact that	because
at this point in time	now
a significant number of	many

6. Tone Calibration

• If the original has personality, keep it. Don't flatten voice into "professional."
• If it's too formal, loosen it. Contractions. Shorter sentences. Direct address.
• If every paragraph sounds equally important, it's AI. Vary emphasis.
• Remove hedging when the author clearly means something definitive.
• Don't add "I think" or "in my opinion" to everything — just state it.

Usage Modes

Mode: Clean (default)

Apply all transformations. Maximum de-AI-ing.

Mode: Light

Only fix punctuation (em-dashes, semicolons) and kill dead phrases. Keep structure.

Mode: Preserve

Keep the author's structure and word choices. Only fix the most egregious tells (em-dashes, "delve into", "it's important to note", sycophantic openers).

For File Processing

1. Read the target file
2. Apply transformations based on mode
3. Show a diff summary of what changed
4. Write the cleaned version (or show it for approval)

Important

• Don't rewrite content. Fix patterns, don't change meaning.
• Don't inject personality that isn't there. Remove AI voice, don't replace it with a different fake voice.
• Preserve technical accuracy. Never change technical terms, proper nouns, or domain-specific language.
• Context matters. "Robust" in a technical spec about software testing is fine. "Robust" describing a blog post strategy is AI slop.