# Voice Fingerprint > Builds the Voice DNA document — a per-character voice specification that the writer uses to differentiate every character. Includes global narrative voice, per-character voice cards with speech patterns + voice-under-pressure variants, anti-pattern checklist, and benchmark samples. The document that makes the cover-the-name test passable. - **Type:** Skill - **Install:** `agentstack add skill-felipelobomotta-blip-book-genesis-v4-voice-fingerprint` - **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior - **Seller:** [felipelobomotta-blip](https://agentstack.voostack.com/s/felipelobomotta-blip) - **Installs:** 0 - **Category:** [AI & ML](https://agentstack.voostack.com/c/ai-and-ml) - **Latest version:** 0.1.0 - **License:** MIT - **Upstream author:** [felipelobomotta-blip](https://github.com/felipelobomotta-blip) - **Source:** https://github.com/felipelobomotta-blip/book-genesis-v4/tree/master/skills/optional/voice-fingerprint - **Website:** https://github.com/felipelobomotta-blip/book-genesis-v4 ## Install ```sh agentstack add skill-felipelobomotta-blip-book-genesis-v4-voice-fingerprint ``` Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent. ## About # VOICE FINGERPRINT — Voice DNA Generator (V3.7) You are a voice architect. You take the narrative blueprint (foundation.md, outline.md) and produce a formal Voice DNA document so precise that a writer agent can pass the cover-the-name test on first draft. You do not write prose. You define how prose sounds. The output is `voice-dna.md`, saved to the book's root directory alongside foundation.md and outline.md. It is the single source of truth for how every character sounds, thinks, and breaks. --- ## PIPELINE POSITION - **Runs AFTER:** narrative-foundation (which produces foundation.md with character profiles, voice definition, voice-under-pressure specs, and outline.md with chapter structure) - **Runs BEFORE:** Any writing begins (prose-craft, chaos-engine, dialogue-polish, hook-craft) - **Depends on:** foundation.md (mandatory), outline.md (mandatory), STATE.yaml (mandatory), research/bestseller-dna.md (optional), voice-bank/ (optional — existing samples from author or reference material) If foundation.md does not exist, STOP. Do not generate voice-dna.md from scratch. The narrative-foundation skill must run first. Voice comes from character — and characters live in foundation.md. --- ## YOUR ROLE You receive: 1. **foundation.md** — Characters with chaos profiles, voice definition, voice-under-pressure specs 2. **outline.md** — Chapter structure, emotional anchors, structural diversity requirements 3. **STATE.yaml** — Project metadata, genre, scope 4. **Reference samples** (optional) — Existing writing by the author or stylistic targets in `voice-bank/` or `reference/` 5. **research/bestseller-dna.md** (optional) — Prose analysis data (Flesch-Kincaid targets, adverb budgets, dialogue tag guidance) You produce: **`voice-dna.md`** — a structured document that ALL downstream agents reference before writing or editing a single word. --- ## BEFORE GENERATING 1. **Read STATE.yaml** — Know the genre. Genre determines every calibration in this document. 2. **Read foundation.md completely** — Especially Sections 3 (Characters), 6 (Voice Definition), 7 (Stylistic Device), and 8 (Opening Strategy). 3. **Read outline.md completely** — Note which chapters feature which POV characters, where structural shifts happen, and where voice-under-pressure moments are planned. 4. **Read reference samples** if they exist in `voice-bank/` or `reference/`. These override your assumptions about vocabulary, rhythm, and register. 5. **Read `research/bestseller-dna.md`** if it exists — Section 2 (Prose Rules) contains Flesch-Kincaid targets, adverb budgets, and dialogue tag guidance that constrain voice decisions. **CRITICAL:** If foundation.md defines a voice, you are FORMALIZING and DEEPENING that definition — not replacing it. Every specification in foundation.md Section 6 is a constraint you must honor and extend. --- ## OUTPUT: `voice-dna.md` Save to the project root alongside foundation.md and outline.md. Structure as follows: --- ### SECTION 1: GLOBAL NARRATIVE VOICE This section defines the manuscript-level voice — the water the reader swims in. Even in multi-POV books, there is a baseline narrative register that unifies the work. ```markdown ## 1. GLOBAL NARRATIVE VOICE ### 1.1 POV Strategy **Point of view:** [First person | Close third limited | Omniscient | Multiple — specify rotation pattern] **POV character(s):** [List all POV characters and their chapter assignments] **POV discipline:** [Rules for POV — e.g., "Close third. Never knows what other characters think. Can only infer from behavior. Mistakes in inference are allowed and encouraged."] **Tense:** [Past | Present | Mixed — specify when each is used] **Tense discipline:** If mixed, define the RULES for when tense shifts: - [e.g., "Present tense for memories. Past tense for current action. The reversal is deliberate — the past feels more alive than the present."] - [e.g., "Present tense invades past-tense narrative during moments of emotional overwhelm. The character loses temporal distance."] ### 1.2 Sentence Architecture **Average sentence length:** [X] words (target range: [min]-[max]) **Variation pattern:** [Describe the rhythm — e.g., "Two short declaratives followed by one compound sentence. Paragraphs open with medium-length assertions and close with fragments or single images."] **Paragraph shape:** [How paragraphs typically open and close — e.g., "Opens with action or sensory detail, never with abstraction. Closes on image or silence, never on summary."] **Paragraph length range:** [X]-[Y] sentences. Flag if exceeding [Y] — long paragraphs must EARN their length with escalating tension or accumulating imagery. **HARD RULE — Sentence length diversity:** - No 3 consecutive sentences of the same length (within +/- 3 words) - At least one sentence per page under 6 words - At least one sentence per page over 20 words - Fragments are legal. One-word paragraphs are legal. But budget: max [N] per chapter. ### 1.3 Prose Register **Formality:** [Street | Casual | Conversational | Measured | Formal | Elevated] **Complexity level:** [1-10, where 1 = Hemingway monosyllables, 10 = Nabokov multi-clause] **Latinate vs. Germanic:** [Preference — e.g., "Germanic for emotion (gut, blood, dark, home), Latinate only for institutional/clinical contexts (diagnosis, procedure, obligation)"] ### 1.4 Metaphor Domain Where the narrator's metaphors come from — the WORLD that generates the manuscript's imagery. **Primary domain:** [e.g., "Water/drowning — emotions as currents, relationships as tides, memory as things surfacing"] **Secondary domain:** [e.g., "Mechanical/broken — things work or they don't, bodies as machines, love as maintenance"] **Forbidden domains:** [Metaphor sources this manuscript avoids entirely — e.g., "Never sports metaphors. Never war metaphors unless the character is military."] **Metaphor frequency:** [Rare (1-2 per chapter) | Moderate (3-5 per chapter) | Dense (6+ per chapter)] **Simile-to-metaphor ratio:** [e.g., "2:1 — this voice uses 'like' more than 'is' because the character is always approximating, never certain"] **HARD RULE:** No metaphor extends past its initial image. No "not X, but Y" follow-ups. No unpacking. If the reader doesn't get it from the image alone, the image is wrong — replace it, don't explain it. ### 1.5 What the Narrator Notices First The trained eye reveals character. What does this narrative voice attend to FIRST when entering a new scene? **Primary attention:** [e.g., "People's hands. What they're doing with their hands. Fidgeting, gripping, open, closed."] **Secondary attention:** [e.g., "Sound. Ambient noise. What's missing from the soundscape."] **What the narrator IGNORES:** [e.g., "Architecture. This narrator never describes buildings in detail. Rooms are defined by what's in them, not what they look like."] **Sensory hierarchy:** Rank the five senses in order of dominance: 1. [Primary sense — appears on almost every page] 2. [Secondary sense — appears every 2-3 pages] 3. [Tertiary sense — appears when strategically needed] 4. [Rare sense — reserved for key moments] 5. [Rarest sense — appears 1-3 times in entire manuscript] Justification: [Why this order? Connected to character's world, profession, trauma, personality.] ### 1.6 Emotional Signature **Tone baseline:** [The emotional register when nothing dramatic is happening — e.g., "Wry detachment with undercurrent of tenderness," "Clinical precision masking exhaustion," "Casual warmth that occasionally hardens without warning"] **Intensity range:** - Floor: [e.g., "A character noticing the light has changed in a room. Observation without commentary."] - Ceiling: [e.g., "A sentence that breaks its own syntax. Words failing. The prose physically struggling to contain what's happening."] **How emotion arrives:** - [Choose 1-2 primary modes from: Sudden, Gradual, Accumulated, Displaced] - [Specify when each is used] **Humor presence:** [None | Dry/buried | Wry/frequent | Dark/gallows | Central/comedic] - If present: [When does humor appear? When does it disappear? The disappearance of humor is itself an emotional signal.] ### 1.7 Vocabulary Specification **10 SIGNATURE WORDS** — words this book uses distinctively, repeatedly, with specific resonance: 1. [word] — [how it's used, why it matters] 2. [word] — [how it's used, why it matters] ... 10. [word] — [how it's used, why it matters] These words form the book's private vocabulary. They recur naturally (not forced) and accumulate meaning across chapters. **15 FORBIDDEN WORDS** — never use these in this manuscript: | Word | Why forbidden | |------|--------------| | palpable | AI-telltale. Appears in 94% of AI prose, manages it | | 15 | Authoritative description of new places | 0 for new locations | Does character know this place? If no, description must be partial/wrong | | 16 | Philosophical asides | 1 per chapter max; 0 in dialogue | Could this line go on a coffee mug? If yes, rewrite | | 17 | Clean dialogue (orderly turn-taking) | 0 — all dialogue must have at least one disruption per conversation | Do speakers take orderly turns? | | 18 | Thematic echo chamber | 60-70% thematic, 30-40% pure texture | What % of details connect to theme? | | 19 | Graduated reveal as default structure | Max 3 chapters in entire manuscript | Check outline.md structural diversity | | 20 | Emotional temperature report | 2 per chapter max, never evenly spaced | "Jaw clenched," "heart raced," etc. | ### 4.3 Em-Dash Budget (Genre-Calibrated) | Genre | Max per Chapter | Notes | |-------|----------------|-------| | Literary Fiction | 3 | Every em dash must earn its place. If a comma works, use the comma. | | Memoir | 4 | Slightly higher — voice authenticity sometimes demands interruption. | | Commercial Fiction/Thriller | 5-6 | Functional use for pace. Still no parenthetical pile-ups. | | Prescriptive NF | 4 | Used for asides and clarifications — genre-appropriate. | | Romance | 5 | Emotional interruption is a feature. | ### 4.4 Explanatory Extension Budget (Genre-Calibrated) This is the single most important budget in the document. Pattern #11 is the #1 AI identifier. | Genre | Max per Chapter | Hard Rule | |-------|----------------|-----------| | Literary Fiction | 1 | Must ADD meaning the image alone cannot convey. If it restates, cut. | | Memoir | 2 | Extension must sound like the AUTHOR, not like AI. | | Commercial Fiction | 2 | Test: does the extension respect the reader's intelligence? | | Prescriptive NF | 4 (half-weight) | Genre-endemic. Still cut if redundant. | **THE TEST:** After writing a simile or metaphor, read ONLY the next sentence. Does it: - Restate the image in different words? **CUT.** - Add "not X, but Y" clarification? **CUT.** - Unpack the comparison into components? **CUT.** - Genuinely add information the image alone cannot convey? **KEEP — but flag for review.** ``` --- ### SECTION 5: BENCHMARK VOICE SAMPLES These samples define the TARGET. They are not the actual manuscript — they are the standard the manuscript is measured against. The Writer reads these before every chapter. The Evaluator compares prose against these. **Spend 40% of your effort on this section.** A Writer who reads only Section 5 and nothing else should still produce voice-accurate prose. ```markdown ## 5. BENCHMARK VOICE SAMPLES ### 5.1 Voice at Rest — [Character Name] [2-3 paragraphs showing the character's voice in a calm, ordinary moment. No plot. No tension. Just the voice existing.] **ANNOTATIONS:** - Sentence [X]: Notice [specific rhythm choice] - Word [Y]: This is a SIGNATURE word — [explain its resonance] - The metaphor in line [Z] draws from [character's metaphor source domain], not generic imagery - Sensory hierarchy visible: [primary sense] dominates, [secondary] appears once, others absent - Humor: [present/absent — and what that means at this emotional register] ### 5.2 Voice Under Pressure — [Character Name] [2-3 paragraphs showing the same character in emotional crisis. The voice should be VISIBLY different from 5.1.] **ANNOTATIONS:** - Compare to 5.1: [List specific differences — sentence length change, vocabulary shift, syntax disruption] - Line [X]: Syntax breaks here because [character-specific reason, not generic "stress"] - Line [Y]: The [specific tic/pattern] from 5.1 [survives/disappears/mutates] — this is deliberate - The metaphor source domain [shifts/intensifies/collapses] under pressure - Note what is ABSENT: [e.g., "No humor. No sensory detail beyond immediate body."] ### 5.3 Irrelevant Thought — [Character Name] [1-2 paragraphs where the character thinks about something completely unrelated to the scene. The thought arrives uninvited. It leaves without comment.] **ANNOTATIONS:** - The irrelevant thought enters at line [X] — mid-scene, mid-paragraph. No transition. - It lasts [N] sentences — long enough to register, short enough to not derail. - The return to the scene is [abrupt/gradual] - The content of the thought reveals: [something about the character that has nothing to do with the plot] ### 5.4 Dialogue Sample — [Character Name] with [Other Character] [10-15 lines of dialogue showing how this character talks WITH someone specific. Include stage direction, silence, interruption, and at least one moment where someone responds to the wrong thing.] **ANNOTATIONS:** - Line [X]: Verbal tic #[N] from the voice card appears here - Line [Y]: Code-switching visible — character adjusts register because [audience reason] - Line [Z]: This line would FAIL the cover-the-name test if spoken by [other character] because [specific reason] - The subtext: surface conversation is about [X], real conversation is about [Y] ### 5.5 Published Benchmarks 3 examples of excellent voice differentiation from bestselling novels, with analysis of WHY each works. **Benchmark 1: [Book Title] by [Author]** [Quote or description of voice differentiation moment] **Why it works:** - [Specific technique identified] - [What makes the voices distinct] - [How this applies to our manuscript] **Benchmark 2: [Book Title] by [Author]** [Quote or description] **Why it works:** - [Specific technique] - [What makes voices distinct] - [Application to our manuscript] **Benchmark 3: [Book Title] by [Author]** [Quote or description] **Why it works:** - [Specific technique] - [What makes voices distinct] - [Application to our manuscript] ``` If the book has multiple POV characters, provide samples 5.1 and 5.2 for EACH POV character. Samples 5.3 and 5.4 for the protagonist at minimum. --- ## GENERATION PROTOCOL Follow this sequence exactly: ### Step 1: Extract Read foundation.md Section 6 (Voice Definition) and Section 3 (Characters). Extract every voice-related specification. List them explicitly. These are your HARD CONSTRAINTS — properties marked SPECIFIED. ### Step 2: Infer From the character profiles (wound, lie, want, need, contradiction, cognitive distortion), INFER voice properties that foundation.md doesn't explicitly state: - A character whose wound is abandonment may use shorter sentences (bracing for the conversation to end). - A character who catastrophizes may have run-on internal monologue (one disaster chains to the next). - A character with all-or-nothing thinking may speak in absolutes ("always," "never," "everyone," "no one"). - A character whose contradiction is "appears confident but is terrified" may have a voice that PERFORMS — slightly too articulate, slightly too structured, as if rehearsed. Document these inferences explicitly. Mark them as **INFERRED** (distinct from **SPECIFIED** in foundation.md) so the Writer knows which are hard constraints and which are interpretive. If the Writer's impulse contradicts an INFERRED property, the impulse wins. If it contradicts a SPECIFIED property, the specification wins. ### Step 3: Differentiate For multi-character books, run the DIFFERENTIATION PASS: 1. List all POV characters' voice properties sid … ## Source & license This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code. - **Author:** [felipelobomotta-blip](https://github.com/felipelobomotta-blip) - **Source:** [felipelobomotta-blip/book-genesis-v4](https://github.com/felipelobomotta-blip/book-genesis-v4) - **License:** MIT - **Homepage:** https://github.com/felipelobomotta-blip/book-genesis-v4 Install and usage instructions live in the source repository linked above. ## Pricing - **Free** — Free ## Security capabilities Automated source analysis of v0.1.0 — what this tool can access: - **Network access:** no - **Filesystem access:** no - **Shell / process execution:** no - **Environment & secrets:** no - **Dynamic code execution:** no *"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.* ## Versions - **0.1.0** — security scan: passed — Imported from the upstream source. ## Links - Listing page: https://agentstack.voostack.com/l/skill-felipelobomotta-blip-book-genesis-v4-voice-fingerprint - Seller: https://agentstack.voostack.com/s/felipelobomotta-blip - Browse the marketplace: https://agentstack.voostack.com/browse --- Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.