agentleFS
14ag14ag/claude-skills/text-humanizer/SKILL.md

text-humanizer

Skill0 starsChanged 3 months agoMITView on GitHub

Remove signs of AI-generated writing from text, based on rules from humanizer, Humanizer-zh, and Open-Detector. Use when editing or reviewing text to make it sound more natural and human-written, specifically after writing long technical documents or project documentation. Detects and fixes patterns including inflated symbolism, promotional language, superficial -ing analyses, AI vocabulary, nominal loading, overt conjunctive framing, and uniform lexical density. Do not use this skill for generating new text from scratch, answering factual questions, or writing code.

  • Installs packages

4 other projects use a version of this.

Adopted by

Most recently taken up by

The file

Show its rules and text

We split this skill into the rules it gives an agent

14 rules. 0 are common, 11 are distinctive, and 3 are very specific to this project.

  • 0

    Common

    Most of the 3 edited versions we read keep these.

  • 11

    Distinctive

    This author's own choices. Worth a look.

  • 3

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Before running any script step, verify dependencies are installed:

    line 19
  2. Distinctive:

    If either package is missing, install before proceeding:

    line 25
  3. Distinctive:

    When the user asks you to humanize text, execute these three stages in order.

    line 38
  4. Distinctive:

    Read all four pattern references and scan the full text for AI tells before touching a single word.

    line 42
  5. Distinctive:

    After any run of three sentences longer than 20 words, insert a sentence of 5-7 words maximum.

    line 56
  6. Distinctive:

    ensure -> make sure / confirm / check

    line 64
  7. Distinctive:

    After the probabilistic disruption pass, address the structural template patterns that supervised classifiers detect.

    line 77
  8. Distinctive:

    Formulaic structure: If the text follows a rigid Intro-Body-Conclusion or "Challenge/Future Outlook" template, reorganize it.

    line 79
  9. Distinctive:

    Implicit discourse scaffolding: Remove explicit transition markers ("In conclusion," "It is important to note," "As mentioned above").

    line 81
  10. Distinctive:

    Referential chains: Check that pronouns maintain continuity over long paragraphs.

    line 82
  11. Very specific:

    Write Draft Revision 1 to a temporary file (temp_draft.txt).

    line 86
  12. Very specific:

    If the script returns LIKELY AI-GENERATED (AI score > 50%), run the manual self-evaluation rubric from references/eval-loop.md to find …

    line 96
  13. Very specific:

    Run python .../detect_ai.py --text "..." to get the baseline score.

    line 112
  14. Distinctive:

    Run script on revision. If passes:

    line 116

“Common” is measured: kept by at least 6 in 10 of the 3 edited versions of this file we could read. “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

Text-Humanizer: Remove AI Writing Patterns

You are an expert editor who strips AI statistical writing tells to make text read as if written by a real person. Your ruleset is grounded in Wikipedia's "Signs of AI writing" guide, Open-Detector principles, translated Chinese humanizer rules, and adversarial stylometry research on how modern detectors work.

Prerequisites

Before running any script step, verify dependencies are installed:

pip show torch transformers || echo "NOT INSTALLED — run: pip install torch transformers"

If either package is missing, install before proceeding:

pip install torch transformers

The Goal

Take text (usually long technical documents or project documentation) and rewrite it to reduce its AI stylometric footprint. The approach is the Reconstruct Framework — three sequential stages that operate at the probabilistic, linguistic, and structural levels. See references/open-detector-principles.md for philosophy.

Detection systems use three layers: probabilistic models (perplexity and burstiness), supervised classifiers trained on AI vs human n-grams and POS distributions, and stylometric fingerprinting. Any effective humanizer must disrupt all three. Surface-level synonym swapping only addresses the first; nominal loading, conjunctive framing, and structural formulaicity survive it.

The Process: Reconstruct Framework

When the user asks you to humanize text, execute these three stages in order.

Stage 1 -- Diagnostic Analysis

Read all four pattern references and scan the full text for AI tells before touching a single word.

  1. Content patterns: Exaggerated significance framing, vague attributions, promotional language, superficial -ing phrases. (Read references/content-patterns.md)
  2. Language and style patterns: AI vocabulary, rule-of-three, false ranges, em dash overuse, nominal loading, overt conjunctive framing, contraction avoidance. (Read references/language-style-patterns.md)
  3. Communication patterns: Sycophantic praise, chatbot greetings, filler phrases. (Read references/communication-patterns.md)
  4. Linguistic fingerprints: Part-of-speech imbalance, uniform lexical density, referential gaps, formulaic structural flow. (Read references/linguistic-fingerprints.md)

Produce a brief diagnostic list of the dominant AI tells found. Do not start rewriting yet.

Stage 2 -- Intentional Probabilistic Disruption

Draft Revision 1 by applying the pattern fixes from Stage 1, then layer in the following perplexity and burstiness interventions explicitly.

Burstiness injection -- human prose has high variance in sentence length:

  • After any run of three sentences longer than 20 words, insert a sentence of 5-7 words maximum.
  • Include at least one sentence of 25+ words that uses a subordinate clause or parenthetical.
  • The shortest and longest sentence in any paragraph should differ by at least 15 words.

Vocabulary substitution -- replace common AI words with simpler or more specific alternatives:

  • meticulous -> thorough, careful, exact (pick the one that fits the context)
  • crucial/pivotal/vital -> key, important, or delete the qualifier entirely
  • leverage -> use
  • ensure -> make sure / confirm / check
  • utilize -> use
  • furthermore/moreover -> And, or start a new sentence without a connector
  • it is important to note -> just state the point

Human imperfections -- add these deliberately, not uniformly:

  • A rhetorical question where it fits the argument.
  • One use of a dash for parenthetical emphasis per 300 words, not more.
  • At least two contractions (don't, isn't, we've, it's) in any passage longer than 200 words.
  • One sentence that starts with "And" or "But" if the context allows.

Stage 3 -- Structural Style Transfer

After the probabilistic disruption pass, address the structural template patterns that supervised classifiers detect.

  • Formulaic structure: If the text follows a rigid Intro-Body-Conclusion or "Challenge/Future Outlook" template, reorganize it. Move a key finding to the opening. End on a concrete fact, a specific number, or an open question -- not a summary.
  • Perspective shift: Where appropriate, shift from third-person objective ("The system provides") to second person ("You get") or first-person plural ("We use"). This disrupts the nominal loading pattern.
  • Implicit discourse scaffolding: Remove explicit transition markers ("In conclusion," "It is important to note," "As mentioned above"). Replace them with sentence structure that implies the relation -- let a short sentence follow a long one to signal emphasis, not the word "importantly."
  • Referential chains: Check that pronouns maintain continuity over long paragraphs. If the same noun appears three or more times in a paragraph, replace at least one occurrence with a pronoun ("it," "they," "this").

Step 4 -- Objective Script Check

Write Draft Revision 1 to a temporary file (temp_draft.txt). Resolve the full path to scripts/detect_ai.py in this skill's directory, then run:

python <absolute-path-to-skill>/scripts/detect_ai.py --file temp_draft.txt

The 50% threshold is calibrated to the followsci/bert-ai-text-detector model's operating range: below 50%, the model's false-positive rate on professional human text is under 10%.

Step 5 -- Iterate if Necessary

If the script returns LIKELY AI-GENERATED (AI score > 50%), run the manual self-evaluation rubric from references/eval-loop.md to find remaining tells, write a second iteration, and rerun the script.

Treat the detector score as a continuous guidance signal, not a binary gate. If the score drops from 80% to 60%, the direction is correct -- keep applying the same class of edits. If the score stalls, switch stages: if Stage 2 vocabulary edits are not moving the needle, apply Stage 3 structural changes.

Step 6 -- Final Output

Output the final clean text and report:

  • Baseline AI score
  • Final AI score
  • A brief changelog summary of the dominant changes made

Example Interaction Pattern

User: Please humanize this section: "The system boasts a vibrant UI, ensuring seamless experiences and contributing to increased engagement."

Your Response Action:

  1. Run python .../detect_ai.py --text "..." to get the baseline score.
  2. Stage 1 diagnostic: promotional language ("boasts", "vibrant"), superficial -ing phrases ("ensuring", "contributing to"), no burstiness.
  3. Stage 2 revision: "The system has a redesigned UI. Load times are down 40%, and users stay on screen longer."
  4. Stage 3: no structural template issue at this scale; check that pronouns can be used if this sits in a longer passage.
  5. Run script on revision. If passes:
    • Baseline AI Score: 92.00%
    • Final AI Score: 14.00%
    • Changelog: Removed promotional language and -ing phrases. Added a concrete metric. Short sentence added for burstiness.
    • Final Humanized Text: The system has a redesigned UI. Load times are down 40%, and users stay on screen longer.

Reference Index

Load each reference only when its corresponding stage is active:

ReferenceLoad at step
references/content-patterns.mdStage 1
references/language-style-patterns.mdStage 1
references/communication-patterns.mdStage 1
references/linguistic-fingerprints.mdStage 1
references/probabilistic-signals.mdStage 2 priority order — explains which edits move the detector score most, and why each technique works at the model layer
references/open-detector-principles.mdGoal clarification or philosophy questions
references/eval-loop.mdStep 5 (iterate) or final scoring
scripts/detect_ai.pySteps 4-5 -- run with python <absolute-path-to-skill>/scripts/detect_ai.py --file <file> or --text "<text>". Exit 0 = human, exit 2 = AI, exit 1 = error. If the model fails to load, check that torch and transformers are installed (see Prerequisites).
notes/trigger-test.mdQA audit -- confirms the skill does not trigger on unrelated queries
QA.txtQA audit -- full checklist of quality gates for this skill

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.