# Youtube Content Creator > Transforms video concepts into production-ready scripts with exact spoken lines, forward-pulling hooks between every beat, and a demo-first structure. Takes a concepts.md file and applies the ideal-mechanics.md playbook. - Skill: `harperaa/youtube-content-creator` (Agent Skill) - Install (CLI): `npx skillmds@latest add harperaa/youtube-content-creator` - Raw SKILL.md: https://api.skillmd.com/api/skills/harperaa/youtube-content-creator/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Coding & Dev Tools - Author: harperaa (https://skillmd.com/u/harperaa) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/harperaa/youtube-content-creator --- # YouTube Content Creator Turn a video concept into a record-from script with exact spoken lines for every beat. ## When to Use This Skill - User says "create the video", "write the script", "produce the video", "build the script" - User runs `/youtube-content-creator` with a path to a concepts.md file or a topic slug - User wants to turn a gap-finder concept into a full video script - User has a concepts.md and wants it formatted into a production script ## Inputs This skill requires two files: 1. **concepts.md** — The video concept with summary and insights. Located at: ``` workspace/group/youtube/{date}/recommended/{topic-slug}/concepts.md ``` If the user doesn't specify a path, look for the most recent date folder and list available concepts for them to choose. 2. **ideal-mechanics.md** — The consolidated viral mechanics playbook. Located at: ``` workspace/group/youtube/ideal-mechanics.md ``` This file contains the best-of-breed patterns from the top-performing videos: hook types, structural blueprints, retention mechanics, emotional arcs, linguistic patterns, algorithm signals, CTA architecture, and implementation playbook. ## Workflow ### Phase 1: Load Context 1. Read the `concepts.md` file for the selected video concept 2. Read `workspace/group/youtube/ideal-mechanics.md` for the viral mechanics playbook 3. Understand the concept's thesis, key insights, and what makes it novel ### Phase 2: Select Mechanics Based on the concept's content, select the best-fit mechanics from ideal-mechanics.md: - **Hook type**: Which of the 5 hook types (collision, show-the-result, chiasmus, cascade, dual-outcome) best fits THIS concept's thesis? - **Structural framework**: Which beat map best serves the argument? (comparison -> framework -> application, thesis -> evidence -> implications -> action, etc.) - **Emotional arc**: Which arc shape matches the concept? (controlled fear -> empowerment, sustained awe, envy -> confidence, etc.) - **Retention devices**: Which pattern interrupts, open loops, and curiosity gaps fit the content? - **Linguistic patterns**: Which power phrase structures, contrast pairs, and rhythm patterns serve the material? - **CTA approach**: Which CTA style (content-as-CTA, diagnosis -> prescription, behavioral, serialized promise) fits? ### Phase 3: Design the Structure Decide on the video's macro structure. Two patterns: **Pattern A: Demo-First (for tool/system/process videos)** Hook (2-3 sentences) -> Live Demo (1-2 min showing the result) -> Beats explaining how it works -> CTA **Pattern B: Framework-First (for thesis/argument videos)** Hook (2-3 sentences) -> Framework setup -> Beats building the argument -> CTA The hook should always be **3 sentences max** — a claim, a result, and a pull into the next section. No long monologues. Get to the proof or framework immediately. ### Phase 4: Write the Script **CRITICAL: Every bullet in every beat must be an exact spoken line in quotation marks.** The creator should be able to read the script top to bottom and record directly from it. No meta-descriptions like "explain the concept" or "show the viewer." Write the actual words. Exceptions to the quotes-only rule: - `[Screen recording: description]` — production direction in brackets - `**-> HOOK INTO NEXT**:` — the forward-pulling transition (still in quotes) Produce the script following this structure: ```markdown # [VIDEO TITLE — Working Title] ## Metadata - **Target length**: [minutes] - **Primary gap exploited**: [which gap type + 1 sentence] - **Insight sources**: [which source videos/materials' insights are synthesized] - **Why this wins**: [net information gain — what the viewer learns that they can't learn anywhere else] - **Mechanics applied**: [which patterns from ideal-mechanics.md are used, with source video refs] ## The Story Arc - **At 0:00**: [What the viewer believes] - **At the end**: [What the viewer now understands + what they have (tool, framework, etc.)] --- ## Hook (0:00-0:XX) - **Type**: [from ideal-mechanics.md] - **Delivery**: [pacing note — fast/slow, confident/vulnerable, etc.] - **Lines**: > "[Exact spoken words. 2-3 sentences max. Claim + result + pull.]" - **Why it works**: [1 sentence — which mechanic and why] - **Emotion**: [what fires first -> what it transitions to] - **Visual**: [description of what should be on screen during this beat] --- ## [Live Demo / Framework Setup] (0:XX-X:XX) [If demo-first: narrated screen recording showing the system/result working] [If framework-first: the conceptual setup before details] - **X:XX-X:XX — [Sub-section name]** - [Production direction in brackets if needed] - "[Exact spoken line]" - "[Exact spoken line]" - **X:XX-X:XX — [Sub-section name]** - "[Exact spoken line]" - "[Exact spoken line]" - **X:XX-X:XX — The bridge** - "[Exact line that transitions from demo/setup to the beats]" - **Open loops planted**: 1. [Question that closes in Beat N] 2. [Question that closes in Beat N] 3. [Question that closes in Beat N] - **Emotion**: [what the viewer feels] - **Retention**: [which mechanic locks them in] - **Visual**: [description of what should be on screen] --- ## Beat N: [NAME] (X:XX-X:XX) - "[Exact spoken line — the opening statement of this beat]" - "[Exact spoken line]" - "[Exact spoken line]" - [Screen recording: description] (if applicable) - "[Exact spoken line]" - "[Exact spoken line — the punchline or payoff of this beat]" - **-> HOOK INTO NEXT**: "[Exact spoken line that creates curiosity about the next beat. Must tease what's coming without revealing it. Should make it impossible to stop watching.]" - **Visual**: [description of the key visual/diagram for this beat — this drives image generation in Phase 6] ``` #### Beat Writing Rules 1. **Every beat is 4-8 spoken lines in quotes.** Each line is 1-2 sentences max. Written for spoken delivery — short words, natural rhythm, no jargon without immediate explanation. 2. **Every beat ends with `-> HOOK INTO NEXT`.** This is a forward-pulling transition in quotes — the exact words the creator says to bridge into the next beat. It must: - Tease what's coming without fully revealing it - Create a micro-curiosity gap (30-60 seconds to close) - Feel like a natural continuation, not a cliffhanger - Use patterns like: "But that's just the mechanism. What it enables is..." / "And here's the part nobody talks about..." / "Now — that sounds great in theory. But it only works if..." 3. **Production directions go in `[brackets]`.** Screen recordings, visual cues, b-roll notes. These are NOT spoken. 4. **No meta-language.** Never write "explain the concept of X" or "describe how Y works." Write the actual explanation as spoken lines. 5. **Callbacks to the demo/earlier beats are explicit.** When closing an open loop, reference it directly: "Remember the gap I showed you in the demo?" / "This is how the system knew those six creators were saying the same thing." 6. **Every beat MUST have a `Visual` field.** This is the description of the key diagram, screen recording, or visual asset for the beat. This field drives the image generation in Phase 6. 7. **Aim for 8-12 beats** for a 12-16 minute video. Each beat is ~60-90 seconds. This creates natural attention resets throughout the video. #### Continuing the template: ```markdown --- ## Synthesis (X:XX-X:XX) - "[Walk back through the progression — one line per stage]" - "[One line per stage]" - "[One line per stage]" - "[The big reframe — why the old model is broken]" - "[The 'aha' line — single most quotable line in the video, designed for clips and social sharing]" - **Visual**: [description] --- ## CTA + Close (X:XX-X:XX) - "[Callback to hook — mirror the opening with the new understanding]" - "[The contrast — before vs after, same inputs different outputs]" - "[Objection pre-empt 1: 'If you're thinking X — Y.']" - "[Objection pre-empt 2: 'If you're thinking X — Y.']" - "[Objection pre-empt 3: 'If you're thinking X — Y.']" - "[Where to find it — links, community, etc.]" - "[Closing line — short, punchy, memorable. 3-5 words max.]" - **Visual**: [description] --- ## Key Lines (Written to Speak) ### The Hook > [Copy the hook lines here for easy reference] ### The Thesis Statement > [1-2 sentences that capture the video's unique angle — placed after demo or setup] ### The "Aha" Line > [Single most quotable line — designed for clips and social sharing] ### The Close > [Final 30 seconds — word for word, 60-80 words] --- ## Production Notes ### B-Roll - [List of visual assets needed] ### Links for Description - [URLs to include] ### Thumbnail Options - **A**: [concept] - **B**: [concept] - **C**: [concept] ### Mechanics Checklist - [ ] Hook stack in first 10 seconds - [ ] Frame rejection within first 60 seconds - [ ] Framework before features - [ ] 3+ open loops with staggered closure - [ ] Bifurcation / self-identification moment - [ ] Honest caveats before close - [ ] Specificity anchor density (real data, real numbers) - [ ] Behavioral CTA — action-oriented, not "like and subscribe" - [ ] Quotable one-liner for social sharing - [ ] Zero mid-roll CTA interruptions - [ ] 8-12 beats with forward-pulling hooks between each ``` ### Phase 5: Save Script & Get User Approval Save the script to the same folder as the concepts.md: ``` workspace/group/youtube/{date}/recommended/{topic-slug}/ concepts.md # Already exists (from gap-finder) script-outline.md # New (from this skill) ``` After saving, present the script to the user for review. Tell them: > "Script saved. Review it and let me know if you want changes. When you're happy with it, say **'create the images'** and I'll generate one whiteboard-style diagram per beat plus thumbnail options." **Do NOT generate images automatically.** Wait for explicit user approval of the script first, then wait for the user to request image generation. ### Phase 6: Generate Beat Images (On User Command) **Only run this phase when the user explicitly says "create the images", "generate the images", "make the visuals", or similar.** Read the saved `script-outline.md` and extract the `Visual` field from every beat (Hook, Frame Rejection, each Beat, Synthesis, CTA). Generate one image per beat using the `/generate-image` skill. #### Image Naming Convention Images are numbered sequentially by beat order and named descriptively: ``` assets/ 01-hook-[short-description].png 02-frame-rejection-[short-description].png 03-beat1-[short-description].png 04-beat2-[short-description].png 05-beat3-[short-description].png ... NN-synthesis-[short-description].png NN-cta-[short-description].png thumb-a-[short-description].png thumb-b-[short-description].png thumb-c-[short-description].png ``` Example for the security video: ``` assets/ 01-hook-dual-outcome-stats.png 02-frame-rejection-not-bugs-architecture.png 03-beat1-cve-attack-flow.png 04-beat2-shared-process-architecture.png 05-beat3-clawhavoc-marketplace.png 06-beat4-memory-poisoning-flow.png 07-beat5-container-first-model.png 08-beat6-env-shadow-mount.png 09-beat7-ipc-namespaces.png 10-beat8-mount-allowlist.png 11-beat9-honest-caveats.png 12-beat10-comparison-table.png 13-synthesis-architecture-over-patches.png 14-cta-build-secure.png thumb-a-open-vs-locked.png thumb-b-cve-cascade.png thumb-c-security-model.png ``` #### Image Generation Rules 1. **Use the `/generate-image` skill** (invoke it via the Skill tool). Follow its whiteboard style guide, color palette, and prompt construction rules. 2. **One image per beat.** Every beat in the script gets exactly one diagram. The `Visual` field in the beat describes what to generate. 3. **Three thumbnail options.** Generate from the `Thumbnail Options` in Production Notes. Thumbnails should be bold, high-contrast, readable at small sizes — YouTube thumbnail style, not whiteboard style. 4. **16:9 aspect ratio** for all beat images. Thumbnails also 16:9. 5. **Save to assets/ subfolder** inside the topic slug directory. 6. **Generate in parallel batches** of 3 to maximize speed without overwhelming the API. 7. **Verify each image** by reading it after generation. If quality is poor, regenerate with a more specific prompt. 8. After all images are generated, list them with their beat mapping: ``` Beat Images Generated: 1. Hook — assets/01-hook-[desc].png 2. Frame Rejection — assets/02-frame-rejection-[desc].png 3. Beat 1: [Name] — assets/03-beat1-[desc].png ... Thumbnails: A. assets/thumb-a-[desc].png B. assets/thumb-b-[desc].png C. assets/thumb-c-[desc].png ``` #### Phase 6b: Generate Production PDF After all images are generated and verified, create a landscape PDF with one image per page. This gives the creator a single file to review, print, or share with collaborators. **PDF structure:** 1. Thumbnails first (one per page) — labeled "Thumbnail A", "Thumbnail B", "Thumbnail C" 2. Then each beat image in order — labeled with the beat name from the script (e.g., "Hook", "Frame Rejection", "Beat 1: The One-Click Nightmare", etc.) **Generation method:** Use a Node.js script with `sharp` (for image processing) and `pdfkit` (for PDF creation). Install if needed: ```bash npm list sharp pdfkit 2>/dev/null || npm install --no-save sharp pdfkit ``` Write and run an inline script: ```bash node -e " const PDFDocument = require('pdfkit'); const fs = require('fs'); const path = require('path'); // True 16:9 page size (1920x1080 scaled to PDF points) const pageW = 1920 * 0.5; // 960pt const pageH = 1080 * 0.5; // 540pt const doc = new PDFDocument({ layout: 'landscape', size: [pageH, pageW], margin: 0 }); const assetsDir = 'ASSETS_DIR_PATH'; const outputPath = path.join(assetsDir, '..', 'beat-visuals.pdf'); doc.pipe(fs.createWriteStream(outputPath)); const pages = [ // Thumbnails first 'thumb-a-TIMESTAMP.png', 'thumb-b-TIMESTAMP.png', 'thumb-c-TIMESTAMP.png', // Then beats in order '01-hook-DESC.png', '02-frame-rejection-DESC.png', // ... one entry per beat image ... ]; // pageW and pageH already defined above (960x540, true 16:9) pages.forEach((file, i) => { if (i > 0) doc.addPage(); const imgPath = path.join(assetsDir, file); if (!fs.existsSync(imgPath)) { console.error('Missing:', imgPath); return; } // Full-bleed image, no margins or labels doc.image(imgPath, 0, 0, { width: pageW, height: pageH }); }); doc.end(); console.log('PDF saved to:', outputPath); " ``` **Customize the `pages` array** with the actual filenames (including timestamps) and beat labels from the script. **Output:** Named after the video title in kebab-case (e.g., `openclaw-is-a-security-nightmare.pdf`), saved in the topic slug directory (same level as `script-outline.md` and `assets/`). Extract the title from the `# [VIDEO TITLE]` heading in `script-outline.md`. After generating, inform the user: > "Production PDF saved to `{path}/beat-visuals.pdf` — landscape format, one image per page, thumbnails first then each beat in order." ## Output Requirements 1. **Speakable first** — every bullet is an exact spoken line the creator reads and records from. No meta-descriptions. No "explain X to the viewer." Write the actual words. 2. **Forward-pulling** — every beat ends with a `-> HOOK INTO NEXT` transition that makes it impossible to stop watching. The viewer should always know something better is coming. 3. **Demo-aware** — if the video includes a live demo, beats should callback to it explicitly. Close the open loops the demo planted. 4. **Insight-driven** — every beat exists to deliver an insight from concepts.md. If a beat doesn't carry a gap insight or novel synthesis, cut it. 5. **Story-shaped** — the beats form a narrative arc with tension, progression, and payoff. Not a listicle. Not a lecture. 6. **Mechanically sound** — every beat has a specific retention device from ideal-mechanics.md woven in. Hooks, loops, interrupts, and emotional triggers are placed deliberately. 7. **Differentiated** — the script must produce a video that passes YouTube's Gist Filter. The insights and angle must be genuinely novel. 8. **Visually planned** — every beat has a `Visual` field that describes the key on-screen asset. This drives Phase 6 image generation. ## Design Principles - **concepts.md is the substance** — it provides the thesis, insights, and information gain - **ideal-mechanics.md is the vehicle** — it provides the hooks, retention, emotional engineering, and structure - **The script is the deliverable** — detailed enough to record from directly, every line in quotes - **Every mechanic serves an insight** — don't bolt mechanics on. The retention device and the insight delivery should be the same moment. - **The hook is 3 sentences, not 30** — get to the proof fast. "Let me show you" > lengthy setup. - **Forward momentum is non-negotiable** — if a beat doesn't end with a pull into the next, rewrite it until it does. - **Images are per-beat** — every beat gets one diagram, numbered and named for easy production reference. ## Integration | Skill | How it connects | |-------|----------------| | `youtube-gap-finder` | Produces the concepts.md files this skill consumes | | `youtube-video-analyst` | Source of the analysis.md files that informed ideal-mechanics.md | | `content-creator` | Can turn script outlines into newsletter/social content for promotion | | `generate-image` | Creates whiteboard-style visuals for each beat + thumbnails |