How to use the VideoEssay agent skill: setup, triggers and examples

What the VideoEssay skill makes the agent do

Given a source and a stance, the agent writes an argument first, breaks it into scripted scenes and visuals, narrates it, renders an mp4 and then reads the render back to check it. Given a blog post, a post, idea or feature plus a length, it can instead produce a short DA-narrated explainer video.

A second mode, voice mode, builds a video of your own blog post. The page scrolls while each sentence is highlighted as your recording, or a voice clone of you, reads it. The skill narrates only the author's own words in this mode.

• Essay mode: argue, script, narrate, render, verify.
• Explainer mode: any source plus a length, narrated by the DA with no opposing quotes.
• Voice mode: your recorded reading or voice clone over the real post page.
• Runs stop in review. The skill does not publish anything; the mp4 is delivered to a folder on your machine.

Install the skill from the download ZIP

Download the VideoEssay ZIP from this page. You can install it in one of two ways, depending on how you use your agent.

For Claude Code, unzip the download into ~/.claude/skills/ to make it available everywhere, or into a project's .claude/skills/ folder to make it available only in that project. For the Claude apps, open Settings, go to Capabilities, then Skills, and upload the ZIP there.

• Personal install: unzip into ~/.claude/skills/.
• Project install: unzip into the project's .claude/skills/ folder.
• Claude apps: Settings → Capabilities → Skills, then upload the ZIP.

Install the requirements before the first run

The skill needs the Bun runtime, the HyperFrames rendering toolchain (shared under LIFEOS/TOOLS/HyperFrames), and the LIFEOS framework with a user Config.json. The config holds your fonts, blog base URL, avatar and runs root. Until these are in place, the tools stop and point you to Setup.md.

To check what is still missing, run the setup check that ships with the skill: bun skills/VideoEssay/Tools/Setup.ts --check. The config file lives at LIFEOS/USER/CUSTOMIZATIONS/SKILLS/VideoEssay/Config.json.

Narration is synthesized on Fish Audio, so you need a FISH_API_KEY. Voice mode also needs recorded audio or a voice clone service; the clone uses the FISH_PRINCIPAL_VOICE_ID and ELEVENLABS_PRINCIPAL_VOICE_ID environment ids.

• Bun runtime
• HyperFrames toolchain (LIFEOS/TOOLS/HyperFrames)
• LIFEOS framework with Config.json (fonts, blog base URL, avatar, runs root)
• FISH_API_KEY for narration
• Optional: recorded audio or a voice clone for voice mode
• Optional: the sibling TightCut skill for tight-cutting recorded narration

Trigger the skill with natural requests

You do not need a special command. Ask in plain language and the agent routes the request to the right intent. Phrases such as video essay, explainer video, respond to this video, argue against this in a video, or turn this post into a video are all recognised triggers.

Say whether you want a review checkpoint. By default the agent shows you the written essay before rendering. If you say straight through or no review, just make it, the checkpoint is skipped.

• Essay: 'Make a video essay arguing against this post, I think it overstates the case.'
• Explainer: 'Make a 2-minute explainer video about my latest post.'
• Response: 'Respond to this video in a video essay.'
• Voice: 'Turn this post into a video from my audio, I recorded myself reading it.'

Use voice mode with your own recording

When you have a recording of your own reading, or a voice clone, ask for a video of the post itself. The agent builds the video from the real post page, scrolls it, and highlights each sentence as it is spoken. Inserts are timed to the paragraph being read, and your portrait appears in the corner badge.

Voice mode narrates only the author's own words. It does not write new narration for you.

• Works with a post plus a recording of you reading it.
• Works with a voice clone of your voice in place of a recording.
• Leave out the recording and the DA reads your exact words instead.

Safety notes: scripts and network access

This skill runs scripts and uses the network. The stage tools run on your machine under Bun, and narration is sent to Fish Audio to be synthesized, so your script text leaves your machine for that step. Review the Config.json and the API keys you provide before the first run.

Nothing is uploaded by this skill. Publishing is a separate decision you make through the Video skill. The finished mp4 and thumbnail are copied to ~/Downloads/<slug>/ only through the Deliver tool, after the render has passed its checks.

• Scripts run locally with Bun.
• Narration text is sent to Fish Audio over the network.
• Runs stop in review; nothing publishes automatically.
• Your camera footage and clips belong to the Video skill, not this one.

Tips for better videos

Give the agent a clear stance. An essay argues a position, so a vague request produces a weak argument. For a response video, paste the material you want answered and say what you disagree with.

Read the essay at the checkpoint. The agent cites verbatim quotes and attributes them to speakers, but you are the one who confirms the argument is fair. Ask for changes there, because text is cheaper to change than a render.

• State your stance in one sentence.
• Name the length for explainers, for example 2 minutes.
• Keep the review checkpoint on unless you are sure of the result.
• Run the setup check again after any config change.