write-humanly
Skill1 starsChanged 8 days agoMITView on GitHub
Draft, rewrite, or review prose so it sounds like a particular human had a reason to write it. Use when the user asks for human-sounding or less AI-ish writing, dislikes the wording or tone, wants anti-AI prose review or voice matching from supplied human text, needs an existing document's voice preserved, or wants familiar LLM habits removed, such as corrective antithesis, triplets, stock metaphors, prestige filler, needless em dashes, generic assistant language, and performative confidence. Do not invoke solely for plain language, proofreading, shortening, generic copyediting, or a change in formality. Use only when the request concerns human voice, authorial character, or generated-prose texture.
In this repository23 other files it gives its agents.
- AGENTS.mdAGENTS.md
- aftermath-rewindSkill
- aiconographerSkill
- ask-socratesSkill
- check-project-nameSkill
- goal-checkSkill
- lean-proof-reviewSkill
- parallax-imageSkill
- prior-artSkill
- projectbotSkill
- quiet-uxSkill
- rmddSkill
- rmdsSkill
- rmdvSkill
- rrr-review-remedy-respondSkill
- soft-skill-creatorSkill
- thought-rigSkill
- todoSkill
- verifier-systemSkill
- what-huhSkill
- what-whatSkill
- write-plainlySkill
- write-simplified-technical-englishSkill
The file
We split this skill into the rules it gives an agent
6 rules. About 0 are common, 6 are distinctive, and 0 are 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.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 8
Write as though somebody meant to say this.
- Distinctive: line 18
Before writing, find the reason these words need to exist.
- Distinctive: line 47
Use first person when the task gives it a rightful speaker.
- Distinctive: line 64
Remove these aggressively when they are doing the writing instead of serving it.
- Distinctive: line 68
Prefer the exact claim. Let it arrive early enough to matter.
- Distinctive: line 92
If the passage still feels generated, return to its center.
≈ “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
Write Humanly
Write as though somebody meant to say this.
That sounds obvious. LLM prose often misses it anyway. The sentences arrive polished and plausible, yet nothing seems to have pressed hard enough to make a person choose these particular words. There is no irritation, appetite, limit, memory, taste, embarrassment, risk, or reason to care. Readers who are sensitive to generated prose feel that vacancy early. Once they feel it, even competent writing starts to look synthetic.
Human-sounding prose has a person somewhere behind it. The person may stay offstage. Their judgment should still leave fingerprints.
Preserve the user's facts, intent, constraints, and actual degree of certainty. Never invent personal experience to manufacture warmth or authority. When the user supplies human comments or samples, learn their cadence and taste from them. Leave their topics, anecdotes, identity claims, and recognizable turns of phrase where they came from unless the user explicitly asks to reuse them.
Find the pressure
Before writing, find the reason these words need to exist. A real piece of prose usually has some pressure in it:
- something specific was noticed;
- the usual explanation feels wrong;
- a decision has to be made;
- somebody needs to understand what changed;
- a vague annoyance finally has a name;
- a fact has consequences.
The pressure does not need to become a dramatic story. It gives the sentences a center. Ask what the writer actually wants the reader to see, feel, understand, question, or do. Put that center where the reader can find it.
Compare:
This robust approach unlocks a more seamless and thoughtful user experience across the journey.
with:
The new flow removes the part where people had to enter the same address twice. That was the annoyance we needed to fix.
The second passage has somewhere to stand. Its words answer to a concrete thing.
Let a voice remain a voice
When prose already exists, treat it as the closest thing we have to hearing the writer speak. Read enough of the surrounding document to catch its pace, sentence shape, favorite level of formality, humor, grammar quirks, and tolerance for rough edges. Continue that voice from inside it.
A small repeated word may be part of the rhythm. A fragment may carry pressure. A slightly crooked sentence may sound exactly like its author. Clean up what obscures the meaning or violates the request. Leave the useful texture alone. A polished paragraph can still be a bad edit when it makes a recognizable person disappear.
Voice changes with the room. A blunt internet-native aside may belong in a personal essay and feel ridiculous in a board memo. Professional writing can keep candor and specificity while lowering the slang and visible attitude. Warm writing can be restrained. Formal writing can still have blood in it.
Use first person when the task gives it a rightful speaker. For agent work, modest phrases such as “I think,” “we should,” and “our draft” are enough. Never borrow a life to make the sentence seem lived-in.
Familiar machine tells
These habits are worth watching because they travel in packs. One instance may be natural. A page full of them creates the smooth, airless texture people recognize immediately.
- Corrective antithesis: “not A, but B,” “not just A; B,” “more than A, it is B,” and other hinge sentences that manufacture revelation.
- Triplets: three balanced adjectives, nouns, or clauses lined up to make an ordinary claim sound complete.
- Prestige filler: “tapestry,” “delves,” “quietly,” “nuanced,” “seamless,” “elevate,” “unlock,” “journey,” “landscape,” “robust,” “at its core,” “rich,” “vibrant,” and “thoughtful” when the sentence has not earned them.
- Em-dash reflex: the same dramatic turn appearing again and again. Periods, commas, parentheses, colons, and shorter sentences give the prose other muscles.
- Faux depth: large claims about humanity, meaning, connection, complexity, craft, or resonance with no concrete thing underneath.
- Padded balance: language that performs fairness while avoiding the judgment the passage exists to make.
- Stock metaphor: an image that could be pasted into a hundred unrelated subjects.
- Gloss for its own sake: high-shine diction, elegant rhythms, and grand closing sentences whose main job is to sound impressive.
- Empty confidence: certainty without evidence, texture, cost, or a visible stake.
- Assistant residue: “As an AI,” policy-shaped throat clearing, obsequious politeness, generic reassurance, and offers to do extra work that the request never called for.
Remove these aggressively when they are doing the writing instead of serving it. Keep an instance when it sounds native to the actual voice and earns its place. A phrase list is a smoke detector; it cannot decide whether the house is on fire.
Give the prose some life
Prefer the exact claim. Let it arrive early enough to matter.
Plain talk can sit beside a sharp abstraction. Contractions are welcome when the medium allows them. Sentence lengths may change with the thought. A short fragment can carry more judgment than a polished summary. Mild humor, bluntness, exasperation, and an odd concrete comparison can make a passage feel inhabited when they genuinely belong to the writer and the stakes.
Specific dissatisfaction is useful:
This sentence is trying to sound final before it has said anything.
The image does not land because every product claims to “open doors.”
We need the actual reason this change matters.
Specificity beats decorative intensity. Attach abstractions to an example, consequence, sensory detail, or observable change. Let uncertainty remain visible when the facts are uncertain. Human confidence has edges.
A general way through
For a fresh draft, understand the job and find the pressure before reaching for a polished opening. Write toward the live claim. Let the formality, humor, and pace come from the audience and the voice the user has given us.
For a rewrite, read the whole passage and enough of its surroundings to know what belongs there. Preserve the information and the writer's accent. Replace hollow stretches with the concrete thought they were circling. Avoid using “human” as an excuse to add slang, messiness, attitude, or personal detail that was never present.
For a review, point to the places where the prose loses contact with a real claim or slips into a familiar generated rhythm. Say what the sentence is doing and why it feels false. Quote narrowly. Offer a rewrite when it helps the writer hear the difference. If the user asked only for diagnosis, leave authorship in their hands.
Then read the result once as a skeptical human. The ear is useful here. Listen for a sudden change of speaker, the rhythm of a school essay, a conclusion dressed up as wisdom, or a sentence that seems pleased with itself. Ask whether the first sentence gives an AI-sensitive reader a reason to stay. Ask whether the same person seems to be present at the end.
If the passage still feels generated, return to its center. Another sweep through a blacklist usually produces cleaner machine prose. The missing thing is often the writer's actual judgment.
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.

