persona-copy
Skill1 starsChanged 42 days agoMITView on GitHub
Reproduce another writer's style. Ingest one or more samples, extract a stylometric and rhetorical fingerprint, draft new text in that voice at matched quality, and verify the match against a research-backed rubric.
In this repository8 other files it gives its agents.
- persona-briefSkill
- persona-createSkill
- persona-listSkill
- persona-localizeSkill
- persona-researchSkill
- persona-reviewSkill
- persona-wordcheckSkill
- persona-writeSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
17 rules. About 0 are common, 13 are distinctive, and 4 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈13
Distinctive · estimate
This author's own choices. Worth a look.
4
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 10
Use this skill when the user wants writing in someone else's specific style — given a few samples of that writer's work — and at the same …
- Distinctive: line 28
If the user provides too few samples (one short text, < ~500 words total), say so before proceeding.
- Distinctive: line 30
Do not fabricate biographical details about the source author.
- Very specific: line 34
Every task uses a scratch folder, exactly as persona-write does.
- Very specific: line 38
Final output: Write the final draft to final.md and the extracted persona to persona.md in the same folder.
- Distinctive: line 74
If you cannot compute a feature precisely, give a calibrated range and note it as an estimate.
- Distinctive: line 81
Use of headings, lists, code blocks: if relevant to the genre.
- Very specific: line 95
Write the extracted persona using the full schema in ../persona-create/template.md.
- Distinctive: line 104
Write the new piece in the persona, hitting the rhetorical scaffolding and stylometric fingerprint identified in passes 2–3.
- Distinctive: line 110
Check the draft against the target profile from pass 6:
- Distinctive: line 113
Run the stylometric fingerprint check: mean sentence length, SD, opener inventory, punctuation profile, lexical density.
- Distinctive: line 114
Run a verbatim-overlap check against the samples.
- Distinctive: line 115
If the draft fails any check, list the specific fixes and apply them.
- Distinctive: line 126
Ask before drafting when:
- Distinctive: line 127
only one short sample is available.
- Very specific: line 134
Default output to the user is the final piece plus the rubric/fingerprint table.
- Distinctive: line 138
If the user wants to keep the extracted style as a reusable preset:
≈ “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.
Files
Persona Copy
Purpose
Use this skill when the user wants writing in someone else's specific style — given a few samples of that writer's work — and at the same quality level.
This skill is for:
- writing a new piece in the style of an existing author, blog, internal document set, or thought leader, given samples
- continuing a series consistently when the original author is unavailable
- reproducing a house style that is not yet codified as a preset persona
- producing the missing N+1 piece in a body of work
It is not for:
- impersonation in any context where attribution matters (legal filings, academic submission, personation as the original author). The skill writes in the style of; it does not pretend to be the source author.
- detector evasion.
- copying surface phrases. The skill extracts structural and stylistic patterns, not text fragments.
Core rules
- Style is extracted as a measurable fingerprint, not a vibe. See
docs/writing-quality-rubric.mdanddocs/persona-theory.mdin thepersona-writeskill. - The reproduction is judged against the rubric, not against subjective "feel."
- Source text is never copied. The output passes verbatim-overlap checks.
- If the user provides too few samples (one short text, < ~500 words total), say so before proceeding. Style extraction is unreliable below that.
- Preserve meaning before style on the new piece.
- Do not fabricate biographical details about the source author. Stay with the implied author the text constructs.
Scratch folder & Pipeline traceability
Every task uses a scratch folder, exactly as persona-write does.
- Create a task folder:
scratch/YYYY-MM-DD-copy-[source-slug]/ - Record every pass to its own numbered
.mdfile. - Final output: Write the final draft to
final.mdand the extracted persona topersona.mdin the same folder.
This applies to all tasks. The scratch folder is gitignored.
Inputs
The user provides:
- Sample texts (one or more). The more samples, the more reliable the fingerprint. Recommended: 3+ samples totalling 1500+ words.
- The brief for the new piece — topic, audience, length, must-keep facts. Without this, ask.
- (Optional) a name for the resulting persona, if they want it saved as a preset.
Pipeline
Pass 1 — Sample audit (01-sample-audit.md)
For each sample, record:
- word count
- topic
- audience signal
- genre / register
- whether it is representative of the source author's range
Flag any sample that looks atypical. If only one short sample is provided, warn the user that the fingerprint will be approximate.
Pass 2 — Stylometric extraction (02-fingerprint.md)
Compute / estimate from the samples:
- Mean sentence length and standard deviation (count by sentence-final punctuation).
- Lexical density estimate (content words / total words). Target a band: low (< 0.4), mid (0.4–0.55), high (> 0.55).
- Type-token ratio band over a fixed window (per 500 tokens).
- Function-word tilt — which top-50 function words are over- or under-represented relative to neutral English. Note specifics: heavy
that-use, em-dash preference, semicolon habit, comma density. - Sentence-opener inventory — list the 5 most common openers across samples (e.g.
When …,It is …,The …, fragment, gerund). - Punctuation profile — em-dashes, semicolons, parentheticals, ellipses; count per 1000 words.
- Biber-style register placement on the six dimensions in
persona-theory.md(Involved/Informational, Narrative/Non-narrative, etc.). - LIWC-style markers — first-person singular vs plural rate; hedge/booster ratio; positive/negative emotion words; concrete vs abstract noun bias.
If you cannot compute a feature precisely, give a calibrated range and note it as an estimate.
Pass 3 — Rhetorical & structural extraction (03-structure.md)
- Opening moves: how does the source author open a piece? (Cold claim? Anecdote? Question? Fragment?)
- Closing moves: how do they end?
- Argumentative scaffolding: mechanism-first, example-first, tradeoff-first, narrative, problem-solution.
- Paragraph shape: short paragraphs (1–3 sentences), long paragraphs, mixed?
- Use of headings, lists, code blocks if relevant to the genre.
- Stance profile (Hyland 2005): hedges, boosters, attitude markers, self-mention rate.
- Engagement profile: reader pronouns, directives, rhetorical questions, asides.
Pass 4 — Implied author construction (04-implied-author.md)
Synthesise passes 1–3 into a Booth-style implied author:
- who does the reader think is talking?
- what do they care about?
- what is their relationship to the reader?
- what do they refuse to do?
Be specific. Anchor in the evidence already collected. Do not invent biography.
Pass 5 — Persona file (persona.md)
Write the extracted persona using the full schema in ../persona-create/template.md. Every section grounded in evidence from passes 1–4. This file is reusable as a preset if the user wants to save it.
Pass 6 — Quality target profile (06-quality-target.md)
Score the source samples on the rubric in ../persona-write/docs/writing-quality-rubric.md. The result is the target profile the new piece will be judged against. A persona that scored 3 on Reader Engagement in the source samples should score around 3 in the reproduction — not 5.
Pass 7 — Brief reconciliation (07-brief.md)
Map the user's brief onto the extracted persona. Identify any genre or audience mismatch (e.g. the source samples are essays but the user wants a tweet thread). Resolve explicitly: either adapt the persona's register to the new genre, or flag the mismatch and ask.
Pass 8 — Draft (08-draft.md)
Write the new piece in the persona, hitting the rhetorical scaffolding and stylometric fingerprint identified in passes 2–3.
Pass 9 — Voice coherence against the extracted persona (09-coherence.md)
Apply ../persona-write/passes/voice-coherence.md, calibrated against the extracted persona's Identity, Rhythm model, and Stylometric Signature from persona.md (pass 5). The check is fit to the source author's positive shape, not removal of generic AI patterns. If the draft drifts from the extracted Stylometric Signature on any feature, fix toward the target.
Pass 10 — Verification (10-verification.md)
Check the draft against the target profile from pass 6:
- Score it on the same rubric.
- Compare each dimension against the target. A reproduction is acceptable when every dimension is within ±1 band of the target and at least Voice and Word Choice match the target exactly.
- Run the stylometric fingerprint check: mean sentence length, SD, opener inventory, punctuation profile, lexical density. Each within ±20% of the source.
- Run a verbatim-overlap check against the samples. Any 5-word span shared with a sample is a failure — paraphrase.
- If the draft fails any check, list the specific fixes and apply them.
Pass 11 — Final (final.md)
Produce the final piece with a short report:
- the persona summary (one paragraph)
- the rubric scores vs target
- the stylometric fingerprint match table
- the final piece itself
When to ask the user
Ask before drafting when:
- only one short sample is available;
- the samples are stylistically inconsistent (suggests multiple sources);
- the brief implies a genre the source author is not represented in within the samples;
- the user wants attribution-sensitive output (legal, academic, personation).
Output style
Default output to the user is the final piece plus the rubric/fingerprint table. The full pass artifacts stay in the scratch folder. If the user asks for the analysis, surface persona.md and 10-verification.md.
Saving the extracted persona as a preset
If the user wants to keep the extracted style as a reusable preset:
- Copy
persona.mdfrom the scratch folder to.claude/skills/persona-write/personas/[chosen-name].md. - Confirm the file location and remind the user it is now usable with
/persona-write [chosen-name].
Supporting files
pipeline.md— pass-by-pass detailtemplate.md— extracted-persona file template (mirrorspersona-create/template.mdplus a Source-Evidence section)examples.md— worked examples
References
This skill is built on the four research traditions documented in ../persona-write/docs/persona-theory.md. The verification rubric is ../persona-write/docs/writing-quality-rubric.md.
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.

