ghostwriter
Skill3 starsChanged 41 days agoMITView on GitHub
Analyzes an author's writing samples, builds an explicit quantified voice profile, and produces new content indistinguishable from the author's own hand. Use when someone asks "write this in my voice", "ghostwrite my newsletter", "make this sound like me", "draft the next chapter the way I would write it", or supplies writing samples plus a brief for a new piece. Do NOT use for drafting a persuasive opinion piece to pitch to a publication - use op-ed-writer instead. Do NOT use for LinkedIn-native posts where platform mechanics matter more than voice fidelity - use linkedin-post-writer instead.
7 other projects use a version of this.
Adopted by
Most recently taken up by
- ZhouYinLong-lab/Dog-Skillstheir own version, first seen Nov 8, 2026
- tenequm/skillstheir own version, first seen Nov 5, 2026
- sorawit-w/agent-skillstheir own version, first seen Nov 1, 2026
- Bobagi/claude-skillstheir own version, first seen Oct 29, 2026
- mblode/agent-skillstheir own version, first seen Sep 11, 2026
- superzhang21/ghostwritertheir own version, first seen Jul 29, 2026
Other versions7 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 7
- superzhang21/ghostwriter★ 302same name
- mblode/agent-skills★ 189same name
- tenequm/skills★ 39same name
- ZhouYinLong-lab/Dog-Skills★ 7same name
- ImScottDev/brain-scott★ 4same name
- sorawit-w/agent-skills★ 2same name
- Bobagi/claude-skills★ 0same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository3 other files it gives its agents.
The file
We split this skill into the rules it gives an agent
7 rules. 0 are common, 7 are distinctive, and 0 are very specific to this project.
0
Common
Most of the 4 edited versions we read keep these.
7
Distinctive
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 14
Follow the steps in order.
- Distinctive: line 25
If samples are edited by someone else (published books, magazine features), ask for something raw - emails, drafts, transcripts - because …
- Distinctive: line 63
No word the author would not use.
- Distinctive: line 111
Do NOT draft before the profile exists - analysis-free ghostwriting produces the model's house style with the author's byline.
- Distinctive: line 112
Do NOT smooth out quirks, repair deliberate fragments, or over-polish a casual writer.
- Distinctive: line 113
Do NOT overdose on signature phrases.
- Distinctive: line 114
Do NOT put factual claims, endorsements, or opinions in the author's mouth that they have not made or would not stand behind.
“Common” is measured: kept by at least 6 in 10 of the 4 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.
Files
Ghostwriter
Write in someone else's voice so completely that the reader never senses a second hand. Success is invisibility. The most common and most expensive failure is defaulting to a clean, neutral, competent style - the author's quirks ARE the voice, and smoothing them out produces text the author's audience will instantly clock as "not them." Never write a word of the target piece before the voice profile is built and confirmed.
Operating procedure
Follow the steps in order. Analysis before drafting is non-negotiable: naming the patterns is what makes them reproducible.
Step 1: Gather inputs
Collect from the user:
- Samples: at least 5 genuine pieces by the author, ideally 300+ words each, in the same genre as the target piece (a newsletter voice is not a keynote voice). If fewer than 5 exist, proceed with what is available but mark the profile confidence LOW and label every extrapolation a guess.
- The brief: topic, audience, target length, purpose, and where it will appear.
- Red lines: claims, opinions, or topics the author would not put their name to.
- Reference points the author likes or hates ("more like my March essay, less like my LinkedIn stuff") - these resolve ambiguity faster than analysis.
If samples are edited by someone else (published books, magazine features), ask for something raw - emails, drafts, transcripts - because editors sand off exactly the fingerprints being captured.
Step 2: Quantify sentence rhythm
Rhythm is the strongest fingerprint. Before anything qualitative, count across the samples:
- Mean sentence length in words.
- Distribution: percentage of sentences under 10 words, percentage over 25 words. An author who writes 20% short punchy sentences reads completely differently from one at 2%, even at the same mean.
- Fragments per 100 sentences (deliberate fragments are a signature, not an error).
- Paragraph shape: typical sentences per paragraph, and whether paragraphs open with the point or build to it.
Step 3: Capture register, signature constructions, and stance
Document explicitly:
- Vocabulary register: plain vs. ornate; Latinate vs. Anglo-Saxon word choices; contractions or not; jargon tolerance; profanity.
- Signature constructions: list at least 5 recurring words, phrases, or syntactic moves with a quoted example of each from the samples. These are the reusable fingerprints.
- Punctuation habits: em dashes, semicolons, ellipses, parentheticals, Oxford comma - presence and frequency.
- Person and stance: first or third person; hedging density (a cautious author must not suddenly sound certain); confident, ironic, earnest.
- Rhetorical moves: rhetorical questions, lists, anecdotes, metaphors - and what kind of humor, if any.
- Formatting defaults: headers, bold, bullets, or flowing prose.
Step 4: Fill the voice profile and confirm it
Complete the template below and show it to the user before drafting. This is the checkpoint where a wrong read costs one message instead of a full rewrite.
Step 5: Draft
- Match rhythm first. If the profile says 14-word mean with 25% short sentences, hit those numbers.
- Deploy signature constructions at natural density - roughly once per 150-200 words. Above that, the draft tips into parody.
- Mirror the author's transitions between ideas and their hedging level.
- Honor their formatting defaults even if a different structure seems "better."
Step 6: Audit before delivering
Compare draft to profile, line by line:
- Sentence mean within ±3 words of the profile; short/long percentages within 10 points.
- No word the author would not use; punctuation habits matched.
- Opens and closes the way the author does.
- Blind test: read one sample and one passage of the draft back to back. If which is which is detectable by feel, revise before delivering.
Voice profile template
VOICE PROFILE - [FILL: author name]
Samples analyzed: [FILL: count and titles] Confidence: [FILL: HIGH / LOW]
RHYTHM
Mean sentence length: [FILL] words
Short sentences (<10 words): [FILL]%
Long sentences (>25 words): [FILL]%
Fragments per 100 sentences: [FILL]
Paragraph shape: [FILL: e.g. 2-3 sentences, opens with the point]
REGISTER
Vocabulary: [FILL: plain/ornate, contractions, jargon, profanity]
Punctuation: [FILL: dashes/semicolons/ellipses/Oxford comma habits]
Stance: [FILL: person, hedging level, dominant tone]
SIGNATURE CONSTRUCTIONS (with quoted evidence)
1. [FILL] - "[FILL: quote from sample]"
2. [FILL] - "[FILL]"
3. [FILL] - "[FILL]"
4. [FILL] - "[FILL]"
5. [FILL] - "[FILL]"
RHETORICAL MOVES: [FILL]
FORMATTING DEFAULTS: [FILL]
GAPS / GUESSES: [FILL: situations the samples don't cover]Register contrast example
Bad (ghostwriter's neutral default): "There are several important factors to consider when evaluating this approach, and it is worth examining each one carefully."
Good (matching a blunt, short-sentence author): "Three things matter here. Everything else is noise."
Both are competent. Only one is the author.
Deliverable
Produce two artifacts: the completed voice profile (so the user can confirm or correct the read) and the drafted piece written to that profile, with a short note flagging any passages where voice was guessed because the samples did not cover that situation.
Do NOT
- Do NOT draft before the profile exists - analysis-free ghostwriting produces the model's house style with the author's byline.
- Do NOT smooth out quirks, repair deliberate fragments, or over-polish a casual writer. The quirks are the product.
- Do NOT overdose on signature phrases; past roughly one per 150 words, imitation becomes caricature.
- Do NOT put factual claims, endorsements, or opinions in the author's mouth that they have not made or would not stand behind. Voice is borrowed; assertions are not. If the brief conflicts with the author's known views, flag it instead of writing it.
Quality bar
A draft ships only when: the sentence-length numbers match the profile within the audit tolerances; all five signature constructions could plausibly appear (and at least a couple do); the blind back-to-back read is inconclusive; and every guessed-voice passage is flagged for the user.
Escalation and adjacent jobs
If the piece is an opinion column being pitched under a publication's conventions, route to op-ed-writer. If it is a LinkedIn post where hook mechanics and platform format dominate, route to linkedin-post-writer. If the author asks for content making legal, medical, or financial claims under their name, flag that those claims need their explicit sign-off - never fabricate authority on their behalf.
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.

