The AI Content Workflow Playbook
You commented CONTENT, so here is everything I promised: the three skills (editing, strategy and building) and the full playbook for using them. Install them in Claude Code or Codex, run the worked example, then swap in your own material. No call, no email, no catch.
01 · Quick Start
A skill is a folder with a SKILL.md file in it. The file tells your AI agent when to use the skill and how to do the job. Claude Code and Codex both read the same format. They just look for the folder in different places.
What you will make
What you need first
Step 1
Download the three zips
Each zip holds one folder: content-editing, content-strategy or content-building. The links are in the Downloads section below.
Claude Code
Unzip into your Claude skills folder
Personal skills live in ~/.claude/skills/. Start a new claude session and type /content-editing, or just ask for the job.
macOS or Linux terminal
mkdir -p ~/.claude/skills for f in content-editing content-strategy content-building; do unzip -o "$f-skill.zip" -d ~/.claude/skills/; done
Windows PowerShell
foreach ($f in "content-editing","content-strategy","content-building") { Expand-Archive "$f-skill.zip" -DestinationPath "$HOME\.claude\skills" -Force }
Codex
Unzip into your agent skills folder
Codex reads personal skills from ~/.agents/skills/. Start a new Codex session and type $content-editing, or run /skills to see the list.
macOS or Linux terminal
mkdir -p ~/.agents/skills for f in content-editing content-strategy content-building; do unzip -o "$f-skill.zip" -d ~/.agents/skills/; done
Windows PowerShell
foreach ($f in "content-editing","content-strategy","content-building") { Expand-Archive "$f-skill.zip" -DestinationPath "$HOME\.agents\skills" -Force }
Step 2
Run your first useful task
Install Node.js 22+ and ffmpeg if you do not have them. Then open a terminal in the unzipped content-editing/example/hook-card folder, start your agent, and paste the prompt. The first run downloads HyperFrames, so it needs internet. You get a real MP4 back.
Node.js and ffmpeg · Windows
winget install OpenJS.NodeJS.LTS winget install Gyan.FFmpeg
Node.js and ffmpeg · macOS
brew install node ffmpeg
Paste into Claude Code or Codex
Use the content-editing skill. Lint and render this HyperFrames hook card to out/hook-card.mp4, then pull the frame at 4.3 seconds and show it to me.
I downloaded the three zips from this page, unzipped them with the commands above into a project's skills folder for each agent, and asked each agent to load them. Codex CLI 0.144.1 listed all three. Claude Code (in the Claude desktop app) listed all three and loaded one on request. The personal folders in the commands above come from each vendor's documentation; I tested the project folders. Your version may differ, so if a skill does not show up, start a fresh session first, then check the folder path. To use the skills in one project only, unzip into .claude/skills/ (Claude Code) or .agents/skills/ (Codex) inside that project.
02 · Downloads
Every bundle has a SKILL.md, a README.md with install steps, the templates and checklists the skill points to, a tested example, and a PROVENANCE.md that lists every source and licence. Everything inside was written for this playbook. The tools they use install from their official sources.
01 / 03 · ZIP
Editing
Cut dead space, add captions and build motion graphics. Edit brief, editor chooser, output checklist, five prompts and a tested HyperFrames hook card.
Download Editing ↓02 / 03 · ZIP
Strategy
Turn audience comments into scripts in your voice. Research brief, evidence log, voice profile, hook and script prompts, review checklist and a full worked example.
Download Strategy ↓03 / 03 · ZIP
Building
Design and code the page your content points to. Design brief, build checklist, preview and review steps, and a tested one-page resource site.
Download Building ↓PROVENANCE.md.
03 · Editing
Editing combines three open tools, but one job rarely needs all three. The skill reads your edit brief, picks the engine, does the work, and checks the render frame by frame before it shows you anything.
VIDEO-USE · MIT
Raw talking footage
Cuts filler words and dead space on word boundaries and burns captions. Needs Python, ffmpeg and an ElevenLabs API key for transcription. ElevenLabs has a free tier and paid plans.
HYPERFRAMES · APACHE-2.0
Captions and motion graphics
You write HTML, it renders an MP4. Easy for an agent to edit. Free, renders locally. Needs Node.js 22 and ffmpeg. This is what rendered the Reel you came from.
REMOTION · OWN LICENCE
React video at scale
Video built from React components. Free for individuals and companies with up to 3 employees. Bigger for-profit companies need a company licence.
Cut
Remove dead space, keep the meaning
The skill builds a cut list from the transcript and shows it to you before it renders. It never speeds up your voice to hit a length. If a video runs long, it cuts a weaker line or holds the visual longer.
Caption
From the transcript, with your spelling list
Speech-to-text writes "Clawed" for Claude and "hyper frames" for HyperFrames. The brief carries a spelling list, and every caption gets checked against it, inside the phone safe area.
Animate
Motion only where it explains something
Lists, numbers, tool names, before and after. Anything that competes with the speaker gets cut.
Tested
What I ran, and what I did not
The HyperFrames path is tested: the hook card in the zip renders as shipped. The video-use cutting path and the Remotion path were not run for this playbook. Follow their own install guides, and treat the skill's steps for them as a starting point.
Example prompt · Editing
Use the content-editing skill. Here is my edit brief: edit-brief.md. Inventory the footage in ./raw, tell me which engine you will use and why, then wait for my OK before cutting anything.
04 · Strategy
Most AI scripts sound like everyone else because they start from a blank prompt. This skill starts from what your audience already said, logs every source, and writes from real examples of how you talk. If there is no evidence for a claim, the claim does not go in.
What you give it
What you get back
Example prompt · Strategy
Use the content-strategy skill. Log these comments in a source-evidence table, find the strongest idea, and write one 40-second script in my voice profile at voice-profile.md. CTA keyword is INQUIRY, exactly.
Lead Gen Jay publishes a free collection of Claude Code skills at github.com/leadgenjay/claude-skills. For content, start with short-form-script, animated-carousels, conversion-copywriting and social-proof-builder. Read a skill before you install it, and open the one-line installer script before you run it, because it pipes a remote script into your shell.
05 · Building
Every piece of content points somewhere: a resource page, a download, a small tool. This skill builds that page from a design brief, checks it on a phone and a desktop, and tells you plainly whether you built a content page or the start of a product.
A content page. Build it today
A product. Prototype first
Brief
Who lands here and the one thing they should do
The design brief asks four things before any code: the visitor, the one action, the real copy, and your brand.
Build
One self-contained file to start
Inline styles, very little JavaScript, real text instead of screenshots, and copy buttons that work with a keyboard.
Preview
375 px and 1280 px, then the keyboard
Nothing scrolls sideways. Tab reaches every link and button. Every download opens. Then the same three checks again on the live address.
Example prompt · Building
Use the content-building skill. Make a copy of example/inquiry-replies for my audience: [who]. The three things to copy are [x, y, z]. My colours are [colours]. Then run the preview review and tell me what passed.
06 · The Full Playbook
Here is one piece of content run all the way through. The creator is demonstration material: Sam Ortiz, a fictional wedding photographer who teaches newer photographers to book weddings. Every input and output below is in the zips, so you can follow along and then swap in your own.
1 · Idea
Start from a problem your audience keeps naming
Input: Sam's research brief. Audience: photographers in their first three years. Stuck on: inquiries that go quiet. Output wanted: one 40-second Reel. Keyword: INQUIRY.
2 · Research
Log the evidence before you write
Input: five demo comments and DMs plus one inbox note, such as "I send my pricing PDF the second someone inquires and then they ghost." Output: an evidence log (E1 to E6) and the idea: new photographers think a fast price reply is professional, but it gives the couple nothing to reply to. Evidence rated medium, so the script tells it as Sam's experience, with no statistics.
3 · Voice + Script
Five hooks, one script, every claim tagged
Output: hook "Photographers, stop sending your price list first." A 101-word script timed to 40 seconds, ending "Comment INQUIRY and I'll send you the three replies I use." The full table is in content-strategy/examples/worked-example.md.
4 · Build + Edit
The opening card and the page the keyword delivers
Edit output: content-editing/example/hook-card/index.html, a HyperFrames card with the hook and one-word captions. Build output: content-building/example/inquiry-replies/index.html, the three replies with copy buttons.
5 · Render
Local, free, about 20 seconds
Output: a 6.0-second, 1080 x 1920 H.264 MP4 of about 180 KB. Rendered with HyperFrames 0.8.217 on Windows 11 for this playbook.
6 · Review
Run the checklists before anyone sees it
Script: evidence, voice, structure, length. Video: cuts, captions, spelling, safe area, identity and lip-sync for any presenter. Page: 375 px, 1280 px, keyboard, every link and download.
7 · Delivery
Keyword in, link out. Test it before you post
Publish the page, then set the comment-to-DM tool you already use to send its link when someone comments the keyword. That tool is outside these bundles. Comment the keyword yourself from a second account and confirm the link arrives before the Reel goes live.
Run the whole example · One prompt
Use the content-strategy, content-editing and content-building skills on the Sam Ortiz demo. Read content-strategy/examples/worked-example.md, then render content-editing/example/hook-card to MP4 and open content-building/example/inquiry-replies/index.html at 375 px and 1280 px. Run each skill's checklist and tell me what passed, what failed and what you could not check.
07 · Voice
In the Reel I said Codex talks back like those Jarvis videos. That comes from the app you run Codex in, not from these skills. A skill file is text instructions. It cannot listen, speak or control your computer on its own.
CODEX · CHATGPT DESKTOP APP
A spoken conversation that starts and steers tasks
Open a Codex task in the ChatGPT desktop app and choose Start voice chat. You can start a task, check on it, redirect it and hear progress back. OpenAI lists it for Plus, Pro, Business, Edu and Enterprise plans, rolling out by plan and workspace. Their docs describe the macOS app, and Codex on iOS once your phone is paired with a desktop host. If you do not see the button, update the app.
CLAUDE CODE · /VOICE
Dictation into the prompt
Run /voice in the Claude Code CLI, then hold Space and speak. Your words appear in the prompt and Claude replies in text. It needs a Claude.ai sign-in (not an API key) and a local microphone, and it does not work over SSH or in cloud sessions.
Illustrative only. Not a recorded transcript
A task started by voice runs with the same permissions as one you type. Claude Code dictation does not use Claude messages. Plans and platforms change, so check the current docs before you rely on either: OpenAI: ChatGPT Voice and Claude Code: voice dictation, both read October 29, 2026.
08 · Troubleshooting
The eight problems that come up most, and what to do about each. The rule underneath all of them: keep the approved footage, faces and voice as they are, and fix around them.
Install
The skill does not show up
Start a fresh session. Check the path is skills/content-editing/SKILL.md, not skills/content-editing/content-editing/SKILL.md. Claude Code reads ~/.claude/skills. Codex reads ~/.agents/skills.
Media
Missing files or a blank render
Ask the agent to inventory every input with ffprobe before editing. Keep media inside the project folder. The hook card loads GSAP from a CDN, so the first render needs internet.
Captions
Wrong words, wrong timing or covered faces
Add the misheard names to the brief's spelling list and re-run captions only. Keep captions out of the top 250 px and bottom 400 px of a vertical frame.
Render
The render fails
Run npx hyperframes doctor, then npx hyperframes lint. Most failures are a missing ffmpeg, Node older than 22, or a timeline not registered on window.__timelines.
Voice
It does not sound like you
The voice profile is too thin. Paste five more real captions or transcripts, then ask for a voice check that rewrites only the lines that miss.
Identity
An AI presenter's face or outfit drifts
Go back to the approved reference images for every clip. Do not regenerate a face to patch one shot. Cut to a graphic over that moment instead.
Lip-sync
Mouth and audio do not match
Check every clip boundary at full speed. Fix sync on the clip, or cover the moment with a cutaway. Never stretch or squeeze the audio to line it up.
Pace
It sounds rushed
The script is too long, not the voice too slow. Aim for about 2.3 to 2.6 spoken words per second. Cut a line, or hold the visuals longer. Do not speed up the narration.
09 · Sources
Checked on October 29, 2026. Install each tool from its official source, and read its licence before you use it for client work.
video-use
github.com/browser-use/video-use
MIT. Checked at commit 43cfc56 (October 8, 2026). Not run for this playbook because it needs an ElevenLabs key.
HyperFrames
github.com/heygen-com/hyperframes
Apache-2.0. CLI 0.8.217 tested; 0.8.218 is the latest release.
Remotion
github.com/remotion-dev/remotion
Remotion License: free for individuals, companies with up to 3 employees and non-profits; company licence above that. v4.0.539. Agent skills: npx skills add remotion-dev/skills. Not run for this playbook.
Jay's skills
github.com/leadgenjay/claude-skills
README states MIT. Linked, not included. Checked at commit 6f9bff9 (October 18, 2026).
You commented CONTENT on my Reel about rebuilding a content workflow with Codex. This page is the whole delivery: three skills and the playbook. If something here is wrong or out of date, tell me in the comments and I will fix the page.