agentleFS
fortunto2fortunto2/solo-factory/skills/humanize/SKILL.md

solo-humanize

Skill18 starsChanged 4 months agoMITView on GitHub

Use when "humanize this", "make it sound human", "strip AI patterns", "clean up the copy", or text reads like AI-generated output with em dashes and stock phrases.

The file

We split this skill into the rules it gives an agent

7 rules. About 0 are common, 6 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈6

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    After /content-gen, /landing-gen, /video-promo — polish the output

    line 24
  2. Distinctive:

    Before publishing any user-facing prose (blog posts, landing pages, emails)

    line 25
  3. Distinctive:

    When editing CLAUDE.md or docs that will be read by humans

    line 26
  4. Distinctive:

    Rule: If you find 3+ triplet lists in one document, break at least half of them.

    line 88
  5. Distinctive:

    Portability test: If a sentence would read identically about another person, company, or product, it is filler.

    line 137
  6. Distinctive:

    Protect the specific fact: Never smooth a detail into generic importance.

    line 144
  7. Distinctive:

    If file path: write the cleaned version back.

    line 176

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

/humanize

Strip AI writing patterns from user-facing text. Takes a file or pasted text and rewrites it to read like a human wrote it, without losing meaning or structure.

Why this exists

LLM output has recognizable tells — em dashes, stock phrases, promotional inflation, performed authenticity. Readers (and Google) notice. This skill catches those patterns and rewrites them.

When to use

  • After /content-gen, /landing-gen, /video-promo — polish the output
  • Before publishing any user-facing prose (blog posts, landing pages, emails)
  • When editing CLAUDE.md or docs that will be read by humans
  • Standalone: /humanize path/to/file.md

Input

  • File path from $ARGUMENTS — reads and rewrites in place
  • No argument — asks to paste text, outputs cleaned version
  • Works on .md, .txt, and text content in .tsx/.html (string literals only)

Pattern Catalog

1. Em Dash Overuse (—)

The most obvious AI tell. Replace with commas, periods, colons, or restructure the sentence.

BeforeAfter
"The tool — which is free — works great""The tool (which is free) works great"
"Three features — speed, security, simplicity""Three features: speed, security, simplicity"
"We built this — and it changed everything""We built this. It changed everything."

Rule: Max 1 em dash per 500 words. Zero is better.

2. Stock Phrases

Phrases that signal "AI wrote this." Remove or replace with specific language.

Filler phrases (delete entirely):

  • "it's worth noting that" → (just state the thing)
  • "at the end of the day" → (cut)
  • "in today's world" / "in the modern landscape" → (cut)
  • "without further ado" → (cut)
  • "let's dive in" / "let's explore" → (cut)

Promotional inflation (replace with specifics):

  • "game-changer" → what specifically changed?
  • "revolutionary" → what's actually new?
  • "cutting-edge" → describe the technology
  • "seamless" → "works without configuration" (or whatever it actually does)
  • "leverage" → "use"
  • "robust" → "handles X edge cases" (specific)
  • "streamline" → "cut steps from N to M"
  • "empower" → what can the user now do?
  • "unlock" → what's the actual capability?

Performed authenticity (rewrite):

  • "to be honest" → (if you need to say this, the rest wasn't honest?)
  • "let me be frank" → (just be frank)
  • "I have to say" → (just say it)
  • "honestly" → (cut)
  • "the truth is" → (cut, state the truth directly)

3. Rule of Three

AI loves triplets: "fast, secure, and scalable." Real writing varies list length.

BeforeAfter
"Fast, secure, and scalable""Fast and secure" (if scalable isn't proven)
"Build, deploy, and iterate""Build and ship" (if that's what you mean)
Three bullet points that all say the same thingOne clear bullet

Rule: If you find 3+ triplet lists in one document, break at least half of them.

4. Structural Patterns

Every section has the same shape: AI tends to write: heading → one-sentence intro → 3 bullets → transition sentence. Real writing varies section length and structure.

Hedging sandwich: "While X has limitations, it offers Y, making it Z." → Pick a side. State it.

False balance: "On one hand X, on the other hand Y." → If one side is clearly better, say so.

5. Sycophantic Openers

  • "Great question!" → (cut)
  • "That's a fantastic idea!" → (cut, or say what's specifically good about it)
  • "Absolutely!" → (cut if not genuine agreement)
  • "I'd be happy to help!" → (just help)

6. Passive Voice / Weak Verbs

  • "It should be noted that" → (cut, just note it)
  • "There are several factors that" → name the factors
  • "It is important to" → say why
  • "This can be achieved by" → "Do X"

7. Rhetorical Patterns

Word-level fixes miss these. They survive shortening: a 300-character message can still carry three of them, which is why manually trimming text does not make it sound human. Cut them, don't rewrite them into better versions.

PatternBeforeAfter
Binary contrast"It's not a docs problem. It's a coverage problem.""6 questions fail because 3 devices have no docs."
Negative listing"Not a wrapper. Not a chatbot. A compiler.""It's a compiler."
Colon reveal"The catch: it needs your docs.""It needs 3 docs nobody has written."
Faux-insight setup"The part everyone misses is distribution.""Distribution is the moat."
Throat-clearing"Here's the thing." / "Let me be clear."(cut, state the point)
Rhetorical setup"What if I told you...", "Think about it:"(cut, make the claim)
-ing pseudo-analysis"…shipped v2, highlighting their focus on speed.""…shipped v2, which cut load time to 400ms."
Importance puffery"marks a pivotal moment", "a testament to"state the fact, let the reader judge
Metadiscourse"This matters more than it sounds."(cut — the fact carries it)
Weasel attribution"experts agree", "studies show"name the source or drop the claim
Synonym cycling"the agent… the assistant… the tool…"one name, repeated
Profound kicker"The future isn't coming. It's already here."delete it, end on the last concrete line
Summary recap"In conclusion…", "Overall…"end on the takeaway or next action

Portability test. If a sentence would read identically about another person, company, or product, it is filler. Replace it with a fact, mechanism, number or consequence specific to this subject, or cut it.

Show, don't label. Cut commentary that tells the reader a point is important, surprising or subtle. If the prose already shows it, delete the label.

Protect the specific fact. Never smooth a detail into generic importance. "significantly improves productivity" → "cut review time from 30 minutes to 8."

Process

  1. Read the input — file path or pasted text.
  2. Scan for patterns — check each category above. Count violations per category.
  3. Rewrite — fix each violation while preserving:
    • Technical accuracy (don't change code, commands, or technical terms)
    • Structure (headings, lists, code blocks stay)
    • Tone intent (if the original was casual, keep it casual)
    • Length (aim for same or shorter, never longer)
  4. Report what changed:
    Humanized: {file or "pasted text"}
    
    Changes:
      Em dashes:  {N} removed
      Stock phrases: {N} replaced
      Inflation: {N} deflated
      Triplets: {N} broken
      Sycophancy: {N} cut
      Rhetorical patterns: {N} cut
      Total: {N} patterns fixed
    
    Before: {word count}
    After:  {word count}
  5. If file path: write the cleaned version back. Show a diff summary. If pasted text: output the cleaned version directly.

Detect mode

Asked "is this AI slop?" or handed someone else's draft to audit: name each pattern from this skill that appears, quote the line, give the fix in a few words. Do not rewrite, do not score the text, and never claim an AI wrote it — detectors guess, a named pattern is evidence the writer can check. Offer to edit afterwards.

What NOT to change

  • Code blocks and inline code
  • Technical terms, library names, CLI commands
  • Quotes from other people (attributed quotes stay verbatim)
  • Numbers, dates, URLs
  • Headings structure (don't merge or split sections)
  • Content meaning — only rephrase, never add or remove ideas

Edge Cases

  • Short text (<50 words): just apply stock phrase filter, skip structural analysis
  • Already clean: report "No AI patterns found. Text looks human."
  • Code-heavy docs: skip code blocks entirely, only process prose sections
  • Non-English text: apply em dash and structural rules (they're universal), skip English stock phrases

Credits

The rhetorical-pattern taxonomy in section 7 and the detect-mode contract are adapted from no-ai-slop by Peter Yang (MIT).

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

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.