humanizing-generated-prose
Skill1 starsChanged 3 months agoMITView on GitHub
Use when supplied prose sounds generic, formulaic, over-polished, or recognizably AI-shaped and needs a more natural human rhythm while preserving facts, meaning, citations, protected text, and any supplied writer voice.
In this repository65 other files it gives its agents.
- analyzing-change-blast-radiusSkill
- analyzing-developer-career-opportunitiesSkill
- art-directing-web-interfacesSkill
- assessing-developer-proficiencySkill
- auditing-dependency-riskSkill
- auditing-pixel-perfect-frontendSkill
- bootstrapping-modern-web-appsSkill
- brag-slimSkill
- building-brand-guidelinesSkill
- building-conversion-product-pagesSkill
- building-developer-career-roadmapsSkill
- building-hybrid-game-assetsSkill
- building-premium-nextjs-interfacesSkill
- building-project-showcasesSkill
- building-regression-testsSkill
- building-reliable-node-api-boundariesSkill
- choosing-application-architectureSkill
- craft-premium-motionSkill
- creating-character-sprite-pipelinesSkill
- defining-brand-strategySkill
- designing-action-combatSkill
- designing-developer-portfolio-evidenceSkill
- designing-relational-data-modelsSkill
- designing-software-boundariesSkill
- designing-test-strategiesSkill
- designing-ui-systemsSkill
- designing-visual-identitiesSkill
- documenting-architecture-decisionsSkill
- editing-for-clarity-and-toneSkill
- engineering-gsap-animationsSkill
- evaluating-developer-proficiencySkill
- evolving-postgres-schemas-safelySkill
- implementing-reference-faithful-uiSkill
- investigating-codebase-semanticallySkill
- managing-implementation-slicesSkill
- mapping-existing-codebase-structureSkill
- naming-brands-and-productsSkill
- optimizing-frontend-motion-performanceSkill
- orchestrating-cinematic-web-motionSkill
- planning-codebase-changes-with-evidenceSkill
- planning-engineering-workSkill
- planning-launch-distributionSkill
- planning-safe-refactorsSkill
- planning-written-communicationSkill
- preparing-project-launchSkill
- profiling-postgres-query-performanceSkill
- reconstructing-images-as-threejsSkill
- reviewing-api-securitySkill
- reviewing-apple-platform-interfacesSkill
- reviewing-pull-requestsSkill
- reviewing-web-securitySkill
- selecting-working-methodsSkill
- shipping-github-vercel-changesSkill
- teaching-developer-conceptsSkill
- testing-integration-boundariesSkill
- testing-playable-gamesSkill
- testing-web-applications-end-to-endSkill
- threat-modeling-applicationsSkill
- tracing-code-execution-pathsSkill
- translating-figma-to-nextjsSkill
- The first 60 shown.
The file
We split this skill into the rules it gives an agent
11 rules. About 0 are common, 9 are distinctive, and 2 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.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 14
Before changing style, identify protected material.
- Distinctive: line 16
If the user supplies a voice sample, infer only observable traits such as sentence density, directness, vocabulary range, formality, …
- Distinctive: line 20
Look for patterns that make the prose feel generated rather than assuming any single marker proves AI authorship:
- Distinctive: line 32
Do not impose arbitrary bans on punctuation or vocabulary.
- Distinctive: line 36
Put the concrete point earlier.
- Distinctive: line 39
Remove redundant transitions and let adjacent ideas connect directly when the relationship is clear.
- Distinctive: line 40
Keep useful asymmetry: not every section needs the same number of sentences, bullets, or rhetorical beats.
- Distinctive: line 47
Do not optimize for AI-detector scores or claim that the result is human-authored.
- Distinctive: line 48
Do not fabricate anecdotes, opinions, sensory details, mistakes, slang, or personal experience to create fake authenticity.
- Very specific: line 49
Use editing-for-clarity-and-tone when the main goal is conventional editorial improvement rather than generated-prose naturalization.
- Very specific: line 50
Use writing-brand-voice-and-messaging when the task is to define a reusable verbal identity rather than adapt one draft.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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
Humanizing Generated Prose
Overview
Rewrite generic generated prose into a deliberate human draft without changing its factual payload or pretending to know who authored it. Humanization is a fidelity-sensitive edit, not detector evasion.
Protect the Source First
Before changing style, identify protected material. Unless the user explicitly puts it in scope, preserve quotations, citations, links, code, identifiers, proper nouns, numbers, dates, product names, legal language, and other exact strings whose mutation could change evidence or behavior.
If the user supplies a voice sample, infer only observable traits such as sentence density, directness, vocabulary range, formality, punctuation habits, and use of first person. Do not caricature a person or invent biographical mannerisms.
Diagnose the Draft
Look for patterns that make the prose feel generated rather than assuming any single marker proves AI authorship:
- generic throat-clearing before the actual point;
- repeated sentence shapes and paragraph lengths;
- stacked abstractions with few concrete nouns or actions;
- inflated adjectives, empty significance claims, or vague praise;
- mechanical transition words between every paragraph;
- symmetrical lists or repeated rule-of-three structures without semantic need;
- unnecessary restatement of the prompt or conclusion;
- excessive hedging or certainty not supported by the source;
- canned openings, summaries, and closing invitations.
Do not impose arbitrary bans on punctuation or vocabulary. A dash, colon, heading, or transition is a problem only when its use is repetitive, unnatural for the target voice, or obscures meaning.
Rewrite
- Put the concrete point earlier.
- Replace abstraction with source-supported actors, actions, objects, and consequences.
- Vary sentence length where the idea benefits from it; do not manufacture randomness.
- Remove redundant transitions and let adjacent ideas connect directly when the relationship is clear.
- Keep useful asymmetry: not every section needs the same number of sentences, bullets, or rhetorical beats.
- Preserve uncertainty at the same level as the source.
- Match the supplied voice when evidence exists; otherwise prefer a neutral, natural, direct register.
- Re-read against the original and restore any fact, qualification, citation, or protected span that drifted.
Boundaries
- Do not optimize for AI-detector scores or claim that the result is human-authored; detector outputs are not a reliable writing-quality target.
- Do not fabricate anecdotes, opinions, sensory details, mistakes, slang, or personal experience to create fake authenticity.
- Use
editing-for-clarity-and-tonewhen the main goal is conventional editorial improvement rather than generated-prose naturalization. - Use
writing-brand-voice-and-messagingwhen the task is to define a reusable verbal identity rather than adapt one draft.
Completion Check
The revision should remain factually equivalent, preserve protected material, sound less templated at sentence and paragraph level, retain any distinctive supplied voice, and avoid replacing one rigid style recipe with another.
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
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.

