# Vc Voiceover Script > Turn a reference video (style you want to copy) plus a Vantage Circle topic/product angle into an original voiceover script and a two-column (voiceover left, screen ideas right) production sheet artifact. Use when the user shares a YouTube link and says "make something like this for Vantage Circle," or asks for a video script, shot list, screen ideas, or production sheet. - Skill: `prawal-bhuyan/vc-voiceover-script` (Agent Skill) - Install (CLI): `npx skillmds@latest add prawal-bhuyan/vc-voiceover-script` - Raw SKILL.md: https://api.skillmd.com/api/skills/prawal-bhuyan/vc-voiceover-script/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Coding & Dev Tools - Author: Prawal-Bhuyan (https://skillmd.com/u/prawal-bhuyan) - Updated: 2026-09-22 - Page: https://skillmd.com/skills/prawal-bhuyan/vc-voiceover-script --- ## What this skill produces Given a reference video (a link the user liked the style of) and a Vantage Circle topic or product angle, produce: 1. An **original** voiceover script that borrows the reference's structural technique and pacing — never its sentences. 2. A **two-column HTML production sheet** (voiceover left / screen ideas right, timecoded) built from the template below. ## Step 1 — Understand the reference, don't just fetch it YouTube pages are JS-rendered shells — `WebFetch` on a `youtube.com/watch` URL only returns nav/footer, never the transcript. Getting real signal takes two passes: - `WebFetch` the oembed endpoint for title/channel: `https://www.youtube.com/oembed?url=&format=json` — gives title + channel name reliably. - `WebSearch` the title + channel for context (series name, sponsor, publish date, general subject). This gets you metadata, not structure. **If the user wants the actual rhetorical structure copied (hook, pacing, central metaphor, sponsor pivot), ask them to paste the transcript** rather than guessing from a title — a prior session guessed from genre conventions and it was wrong until the user pasted the real transcript. ## Step 2 — Extract structure, not sentences Once you have the real transcript, identify: - **The hook** (first 10-15 sec device) - **The reframe** (why the obvious fear/assumption is wrong) - **The central throughline device** (a dichotomy, a metaphor, a recurring visual/verbal callback — whatever organizing idea the writer used) - **The concrete examples** used to prove the point - **The pivot** into product/sponsor — note exactly how it transitions (tone shift, framing move) since this is the hardest part to make feel natural - **The outro** (CTA style, pace change) Do **not** reuse the reference's actual phrases, character names, or parallel sentence structures ("There has never been a better time to be a ___", "picture two types of ___", etc.) with nouns swapped — that reads as a reskin, not an original script. **A prior session did exactly this and the user caught it immediately.** The fix: keep the *technique* (e.g., "one throughline metaphor," "reframe then contrast two archetypes," "specific examples before the abstract point"), invent fresh wording, a fresh central device, and fresh examples grounded in the actual VC topic. ## Step 3 — Write the VC script - Confirm target runtime with the user (default: ask; don't assume 90 sec vs 4 min vs longer). - ~150 words/minute for a natural pace, but pacing devices (holds, pauses) can stretch it — write to the low end of the target and let cut rhythm fill the rest. - Tie the payoff/pivot to a specific Vantage Circle product angle (R&R, engagement, pulse surveys, etc.) — confirm which one if not stated. - **Confirm host vs. voiceover-only before writing screen directions.** VC has produced voiceover + b-roll only (no on-camera host) before, but don't assume it's always the case — ask. ## Step 4 — Break into timed beats Match the reference's cut frequency (ask, or infer from the transcript's pacing — e.g. a "fast" reference cuts every 2-4 sec even mid-sentence). Break the script into one row per beat: timecode, the voiceover phrase spoken during that beat, and the specific visual for it. Flag any deliberate **hold beats** (a pull-quote moment, a tone-shift into the product pivot, an outro slowdown) — these are the moments that should NOT follow the fast-cut rule, and calling them out explicitly stops an editor from cutting through them by default. ## Step 5 — Build the production sheet artifact Use the HTML template below as the base (already validated with a real script in production). Fill in the `` rows; keep the token system, fonts, and layout — they're deliberately chosen (serif italic for voiceover = "spoken word" feel, sans for screen directions = "technical note" feel, monospace tabular timecodes, a warm paper ground with a gold/recognition accent, hold-beat badges in the accent color). Adjust the accent hue only if the topic calls for a different mood. Publish via the `Artifact` tool with `favicon: "🎬"`. ```html {{VIDEO_TITLE}} — Production Sheet

Vantage Circle · Production Sheet

{{VIDEO_TITLE}}

Format: {{FORMAT}} Runtime: {{RUNTIME}} Pace: {{PACE_NOTE}}
TCVoiceoverScreen / B-roll
``` ## Gotchas learned in production - Don't claim to have watched a video you only fetched metadata for — say plainly what you could and couldn't verify. - Don't reskin the reference's actual sentences — extract technique, write fresh content. - A recurring visual motif (a graphic that opens, changes, and pays off again near the end) gives a host-less voiceover+b-roll video the continuity a human host would otherwise provide — reach for one by default when there's no host.