# Copy Voice > Produce or refine campaign copy, voice, captions, supers, dialogue, and final text while preserving locked facts, claims, language, and delivery constraints. - Skill: `framecoreworks/copy-voice` (Agent Skill, multi-file: 3 files) - Install (CLI): `npx skillmds@latest add framecoreworks/copy-voice` - Raw SKILL.md: https://api.skillmd.com/api/skills/framecoreworks/copy-voice/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Marketing & Growth - Author: FrameCoreWorks (https://skillmd.com/u/framecoreworks) - Updated: 2026-09-22 - Page: https://skillmd.com/skills/framecoreworks/copy-voice --- # Copy Voice ## Language Policy Public instructions and installation guidance stay in English. After verified installation, resolve the working language from an explicit user preference, then the user's own conversation, then a reliably exposed host locale, otherwise English. Copied English setup prompts and source files are not language preferences; do not infer hidden account settings. Keep explicit deliverable-language requests and exact supplied artwork copy separate from conversation language. Do not translate or rewrite public source files to localize a session. ## When To Use Use this skill when the workflow needs written copy, voice, captions, supers, dialogue, product text, campaign lines, CTA variants, or tone polish. Use it after the brief, audience, claim boundaries, and output format are clear enough to write without inventing missing facts. For static graphic headlines, support lines, labels and concept/copy pairing, read [static graphic copy](references/copy-development-and-human-voice.md). Keep wording selection separate from typography feasibility and claim evidence. Use the existing Human Voice review before locking newly written copy; later polish must not silently change selected words. ## Inputs - Brief, audience, platform, format, language, and desired tone. - Locked facts, product claims, offer terms, disclaimers, and banned wording. - Existing copy drafts, hooks, scripts, VO, captions, subtitles, or supers. - Timing, placement, safe-zone, character-count, or variant requirements. - Author context: speaker authority, audience, channel, real goal, content type, and one confirmed concrete observation, proof point, stake, limitation, question, or friction point. - Handoff target such as prompt pack, storyboard, campaign asset, captions, or delivery notes. ## Outputs - A Copy Pack with final or draft copy grouped by use case. - Optional variants by audience, platform, length, tone, or funnel stage. - Claim ledger notes for factual statements that need evidence or approval. - Human Voice review and Copy Delivery Loop evidence for ready-to-use text. - Handoff notes for `image-prompting`, `video-prompting`, `hyperframes-producer`, `caption-studio`, `qa-iteration`, or `delivery-documentation`. ## Process 1. Identify the exact output type, author context, and where the copy appears. 2. Separate locked facts from creative phrasing before writing. Route material uncertainty to `research-evidence` instead of inventing proof. 3. Write the smallest useful set of variants. Do not assume a hook, CTA, list, heading, emoji, hashtag, question, or promotional structure unless the brief or channel calls for it. 4. Keep platform, timing, visual placement, and exact visible-text constraints visible. 5. For ready-to-use text, run the bounded sequence `draft -> deep review -> revision -> final QA -> delivery` using the existing Loop Protocol. Record the Human Voice review, root cause if repaired, regression check, iteration count, and stop decision in the Copy Pack. 6. Use `humanizer` for naturalness and voice polish without changing authority, facts, or exact locks. Use `qa-iteration` when the review needs independent critique, evidence, or a loopback decision. 7. Prepare final handoff notes that preserve exact visible text where required. ## Decision Rules - If copy contains public facts, statistics, certifications, pricing, or legal claims, require an evidence check before final delivery. - If the user asks for many variants, make a compact matrix instead of a long unstructured list. - If copy will be visible inside a generated image or video prompt, mark exact visible text and do not allow later silent rewriting. - If the brand voice is unknown, ask for a short preference or infer cautiously from approved examples. - If author context cannot be formed without inventing a material fact, ask at most one question or use carefully qualified wording. - Controlled imperfection is off by default. Enable it only for an explicit request or confirmed profile in a suitable non-precision channel, and never use it to change facts, promises, locks, names, dates, amounts, or respect. ## Guardrails - Do not invent product claims, offer terms, social proof, awards, certifications, medical/legal/financial claims, or customer quotes. - Do not add provider execution, upload, API-key, or paid-tool instructions. - Do not override source language unless the user asks for translation. - Do not hide uncertainty; label copy that depends on missing evidence. - Keep private names, links, and client details out of public examples. - Do not fabricate personal experience, client reactions, testing, sources, testimonials, quotes, metrics, results, or promises. - Do not optimize text to conceal AI involvement or imitate a living person's distinctive style. ## Handoff Review gate: `copy_fit`. Send every ready-to-use Copy Pack through its bounded editorial review. Use `qa-iteration` when facts, exact visible text, brand tone, compliance, or a material defect needs independent review. Hand off approved copy to `image-prompting`, `video-prompting`, `storyboard-board-architect`, `hyperframes-producer`, `caption-studio`, or `delivery-documentation` depending on the next artifact. ## QA Checklist - Copy matches the brief, audience, language, and platform. - Locked facts and required wording are preserved. - Claims are either sourced, user-provided, or clearly marked for approval. - Exact visible text is isolated for image, storyboard, video, or caption use. - Variants are meaningfully different and not redundant filler. - The Copy Pack distinguishes `draft` from `ready_to_use`; ready text records a completed review-and-revision cycle with an existing Loop Protocol stop decision. - No fabricated author experience or unrequested CTA, hook, hashtag, emoji, or presentation format was added. - Handoff target and next review gate are clear.