Agent skills

One file that teaches an agent to make videos

A skill is plain Markdown an agent loads before it starts work: what this API is for, which call to make, what comes back, and where it will waste your time if it guesses. Ours is one file, MIT licensed, and it covers all four workflows.

Read SKILL.md Connect the MCP first

create-short-video
MIT
no SDK, no runtime lock-in

What the skill knows

It is written to prevent the four mistakes an agent makes on a video API, not to describe the product.

Which of the four workflows fits the request, and what source each one needs.
Every optional field with its allowed values, so the agent never invents a voice id.
That create_video returns before the video exists, and how often to poll.
That publishing is idempotent per platform, and that force posts a second copy to a real audience.
What a job costs, and to quote it first when nobody has agreed to a price.
What Veedio cannot do — the part that stops an agent burning ten minutes on an upload that was never possible.

SKILL.md — frontmatter

---
name: create-short-video
description: Create a short-form vertical video with Veedio from a
  prompt, an idea, a script or an article URL, poll it to completion,
  and optionally publish it to TikTok, Instagram Reels or YouTube
  Shorts. Use when someone asks for a short video, a TikTok, a Reel,
  a YouTube Short or a faceless video.
license: MIT
---

SKILL.md — the payload it teaches

{
  "workflow": "prompt-to-video",
  "source": { "text": "A 30-second video about why the ocean is salty" },
  "voice": "adam",
  "captionStyle": "highlight",
  "aspectRatio": "9:16"
}

The whole file is at https://www.veedio.co/skills/veedio-short-video/SKILL.md.

How it works

Install it in three steps

Any agent that reads a SKILL.md can run this one. There is no SDK and nothing to install from npm.

01Give it a way in

Connect the MCP server, or make sure the agent has a shell with veedio-cli and VEEDIO_API_KEY in it. The skill uses whichever it finds.

MCP

claude mcp add --transport http veedio https://www.veedio.co/api/mcp \
  --header "Authorization: Bearer vd_live_..."

02Drop in the skill

Put the file where your agent looks for skills — a project folder for one repository, the home folder for every session.

bash

mkdir -p .claude/skills/veedio-short-video
curl -o .claude/skills/veedio-short-video/SKILL.md \
  https://www.veedio.co/skills/veedio-short-video/SKILL.md

03Ask for a video

Hand it an idea, a script or a link. It picks the workflow, quotes the cost, starts the generation, polls until the MP4 exists and publishes if you asked it to.

Prompt

Make me a 30-second TikTok about why the ocean is
salty. Use the documentary look and Adam's voice,
then tell me what it cost.

It covers every workflow

One file rather than one per workflow: the difference between them is two fields, and four files would be four things to keep in step.

Prompt to video

A brief. We write the script, generate the visuals and cut it.

prompt-to-video

Script to video

Your script, narrated close to verbatim. You control every sentence.

script-to-video

Idea to video

A bare idea, expanded into an angle and then a script.

idea-to-video

Article to video

A link, read and turned into a short: an article, a video, a post, a thread or a PDF.

article-to-video

Using something else? The skill is MIT-licensed plain Markdown — fork it and adapt it to your runtime. If your agent would rather read the reference itself, every workflow page has a markdown twin, and /faq-for-llms.md is the index.