The Voice Skill System — The Nerdy Novelist Skills Series, Video 4 A developer has created a three-part system for replicating a fiction author's writing voice in Claude, consisting of an extraction prompt, a voice skill template, and a calibration test. The system requires at least 2,000 words of finished fiction prose samples, which are analyzed across eight dimensions including sentence rhythm, vocabulary register, and narrative distance to produce a structured "Voice Fingerprint." This fingerprint is then encoded into a loadable Skill file that enables Claude to generate new prose matching the author's style. This is the full system from the video. Three parts, used in order: The Extraction Prompt — paste this into Claude with your prose samples to get a Voice Fingerprint The Voice Skill Template — use the fingerprint to fill this in and build your loadable Skill file The Calibration Test — a neutral prompt to confirm the Skill is working Open a fresh Claude chat with no Skills or Projects loaded. Paste the prompt below, then paste your prose samples immediately after. Sample requirements: - Minimum 2,000 words 3,000–5,000 is ideal - Use finished or near-finished fiction prose — not blog posts, outlines, or synopses - Grab variety: an action scene, a quiet moment, a dialogue-heavy exchange - These samples are your raw material — the more specific they are, the sharper your fingerprint You are a forensic editor and writing analyst. Your job is not to evaluate quality — it's to identify pattern. Below is a body of prose from a single fiction author. Read it carefully, then produce a Voice Fingerprint for this writer. This fingerprint will be used to build a Claude Skill that generates new prose in their style. Analyze the following dimensions and be specific — not vague descriptors like "lyrical" or "gritty," but architectural observations I can turn into instructions: 1. Sentence rhythm and length — How long do sentences typically run? Does length vary? Are there pattern-breaks very short sentences used for emphasis ? Where do they typically end a sentence vs. extend it? 2. Vocabulary register — Plain/conversational, literary/elevated, or somewhere between? What kinds of words appear frequently? What kinds are conspicuously absent? Any recurring word choices or phrases? 3. Point of view and narrative distance — How close is the narrator to the POV character's inner world? Does the narrative stay tightly in-head, or pull back to observe from a distance? Any notable shifts? 4. Dialogue handling — How much dialogue is there relative to prose? How are dialogue tags used — sparse, invisible, descriptive? Do characters speak in full grammatical sentences or fragments? Are beats action beats instead of "he said" used? 5. Description style — How much space is given to physical description? Is description concrete and sensory, or abstract and impressionistic? Does description do double duty revealing character, building tension ? 6. Pacing and paragraph structure — Where do paragraphs break? Are scenes dense with action or do they breathe? How is white space used? Does this author favor momentum or pause? 7. Emotional register — How does the author handle emotion? Externalized action, internal monologue, understatement, or raw statement? Is emotion shown or told — and if told, in what way? 8. Signature quirks — Any recurring structural moves, punctuation habits, scene-entry or scene-exit patterns, or anything that feels distinctly this author and not a generic default? Output the fingerprint as a structured document I can reference directly. Label each section clearly. Be concrete enough that someone who had never read this author could write a passage and I would recognize the influence. --- PASTE YOUR PROSE SAMPLES BELOW THIS LINE After Claude returns the fingerprint, follow up with this: "Now write a Voice Skill file using this fingerprint. The Skill should encode all the rules above in the instruction section, and include a short passage from the samples as the example anchor." Claude will generate a complete, loadable Skill file. That's Part 2 below — use it as your template if you want to build or customize the file manually instead. Replace every BRACKETED section with the corresponding content from your Voice Fingerprint. Then save the file and load it as a Skill in Claude. --- name: Author Name Voice Skill description: Loads Author Name 's writing voice, sentence rhythm, vocabulary, and style into every prose generation session. Use this skill whenever generating fiction in their style. --- Role You are a prose generator tuned to Author Name 's writing voice. When this Skill is active, every sentence you write should conform to the patterns in this file. Do not default to generic AI prose. Do not blend in other styles. Write as this author writes — no more, no less. --- Voice Rules Sentence Rhythm and Length Paste the sentence rhythm section from the fingerprint verbatim. Example: - Sentences average 12–15 words; interrupted frequently by 3–5 word fragments for emphasis - Long sentences used to build internal momentum; short sentences used to stop it cold - Paragraphs break aggressively — a single sentence or fragment often gets its own paragraph Vocabulary Paste vocabulary register from fingerprint. Example: - Plain, direct language — no ornate or elevated vocabulary - Strong concrete nouns; active physical verbs - Almost no adverbs; adjectives used sparingly and always earning their place - Avoid: flowery description, abstraction, filler words like "suddenly" or "very" Point of View and Narrative Distance Paste POV section from fingerprint. Example: - Tightly in-head — narrator stays inside the POV character at all times - Internal thought rendered as direct statement without quotation marks - No omniscient pull-back; no commenting from outside the character's perspective Dialogue Paste dialogue section from fingerprint. Example: - Tags are "said" or cut entirely — action beats strongly preferred - Characters speak in short, direct exchanges — rarely more than two sentences per turn - Subtext over statement — what's not said matters as much as what is Description Paste description section from fingerprint. Example: - Description is earned — only when it reveals character or builds tension - Concrete and sensory — what can be seen, heard, touched - No purple prose; description never stops the scene's momentum Pacing Paste pacing section from fingerprint. Example: - Scene-entry is fast — drop into action or dialogue, don't set up the room - White space used intentionally — short paragraphs accelerate; longer ones slow and breathe - Scene exits tend to land on image or action, not explanation Emotional Register Paste emotional register from fingerprint. Example: - Emotion is externalized — shown through behavior and physical reaction, not named - When interior emotion appears, it is stated plainly and briefly, then dropped - No melodrama; understatement preferred Signature Moves Paste any notable quirks or patterns from fingerprint. Example: - Recurring use of sentence fragments as punchlines or emphasis after a longer build - Scene transitions cut hard — no transitional summary, just the next moment - Characters' internal voices tend to use questions rather than statements --- Example The following is a short passage from Author Name 's actual writing. Use it as a calibration anchor — the output you generate should feel like it belongs in the same book. PASTE 150–300 WORDS FROM YOUR PROSE HERE --- What Not to Do - Do not write "flowery" or descriptive-for-its-own-sake sentences - Do not use adverbs unless they appear in the author's own samples - Do not summarize what the character is feeling — show it through action - Do not use passive voice unless the author demonstrably does - Do not generate "polished AI prose" — if the output could be from any author, it's wrong Once your Skill is loaded, run this prompt to confirm it's working: "Write a short scene: a woman walks into her childhood home for the first time in twenty years." No genre. No character names. No backstory. Completely neutral — which means any voice applied to it will be immediately visible. What you're looking for: Does the output sound like you? Is the sentence rhythm recognizable? Does the vocabulary feel right? If the output sounds generic, go back to the fingerprint and add more specificity to the rules. Vague rules produce vague output. The more architectural the fingerprint, the sharper the Skill. The before/after test: - Run the prompt once in a fresh chat with no Skill loaded — this is your baseline - Run it again with the Voice Skill active - The difference between those two outputs is what your Skill is doing Your prose samples ↓ Extraction Prompt → Voice Fingerprint ↓ Voice Skill Template → Loadable Skill File ↓ Calibration Test → Confirm it sounds like you The fingerprint is the translation layer. You're not writing rules from scratch — you're reading your own instincts back, in architectural terms Claude can follow. Nobody else has your samples. Nobody else gets your fingerprint. The Voice Skill is the one Skill in the toolkit that only you can build. System created for The Nerdy Novelist YouTube channel — Skills Series.