How to use the Video Creation Matrix agent skill

What an agent skill is and what this one makes the agent do

An agent skill is a folder containing a SKILL.md file and optional reference files. When the skill is installed, an agent that supports skills can read these instructions and follow them for matching tasks. This skill is named cli-hub-matrix-video-creation.

For a vague idea, the agent first writes a creative brief, script, shot list, timing map and audio plan, and saves a creative_direction.md file before final assembly. It then chooses tools for each step based on your goal, budget and offline needs. Those tools can include CLI-Anything harnesses, ffmpeg-style command-line tools, Python libraries or cloud APIs.

• Planning: brief, script, shot list, timing map and audio plan, saved in creative_direction.md
• Source finding: internet video and music search and download, with source and licence notes
• Editing and captions: caption design, timeline editing in Shotcut or Kdenlive, and rendering
• Review and troubleshooting: frame-based art direction review and render diagnostics

Install the skill in Claude (ZIP or Settings)

Download the skill ZIP from this page. You can then install it in one of two ways. For Claude Code, unzip it into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder to share it with that project. In the Claude apps, you can upload the ZIP under Settings, then Capabilities, then Skills.

After installing, start a new session so the agent can see the skill. The skill folder should contain SKILL.md and the references folder.

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

Install the command-line tools the skill uses

The skill uses the cli-hub command-line tool to install and check its matrix. Install the registered matrix CLIs with the command below. The skill notes that cloud APIs, Python packages, native binaries and third-party public CLIs are also providers, but they should be installed only when a task needs them. Do not hand-write pip install commands for CLI-Anything matrix members; use the matrix install instead.

The skill's requirements include ffmpeg and ffprobe, which preflight checks for, and optional Python libraries such as moviepy, whisper, pydub and yt_dlp.

Check what is available before you start

Preflight reports which providers are available in your environment. It is an availability report, not a choice of provider; the agent still chooses tools based on your request. Run preflight once per session and reuse the result. If you are offline, use the offline option so the agent only considers offline providers.

You can also inspect providers and recipes with the info command. Look at the preflight output to see which tools or API keys are missing before you ask for a video that needs them.

• Full check in JSON form
• Check one capability, for example composite.assemble
• Offline-only check

Trigger the skill with a natural request

You do not need special syntax. Describe the video, its length, audience and the material you have. The skill is designed for a vague concept as well as for a full brief. If you want captions, a particular language or a specific platform, say so in the request, because the skill uses the output language you specify for all viewer-facing content.

For a non-trivial video, expect the agent to write creative_direction.md and show you a plan before final assembly. Read the plan and correct it. The skill notes that its outputs involve many judgement calls and tool choices that should be checked.

• Make a 30-second product promo from my clips, with short captions for social media.
• Plan a 60-second recap of my training project. Show me the shot list and timing before you render.
• Use only my own footage and a royalty-free music track I provide. Work offline with local tools.
• Diagnose why my last render has audio out of sync and fix the edit project.

Safety notes: scripts, network and paid APIs

The skill runs scripts and uses the network. It includes internet video and music search and download steps. Downloaded footage or music can carry licensing restrictions, so record the source, licence and attribution for every file before you use it in a published video.

The skill is designed not to call paid APIs silently. Cloud providers such as RUNWAY_API_KEY, ELEVENLABS_API_KEY, OPENAI_API_KEY and SUNO_API_KEY are used only when you supply credentials or explicitly agree. If a paid provider is needed, the agent should ask you to set the variable and tell you the likely cost and quality before continuing.

• Do not use copyrighted music or footage without rights to it
• Keep paid API keys unset unless you want cloud providers used
• Use offline mode when you must avoid network calls

Tips for better results

Give the agent the material first, then the goal. The source describes sorting candidate footage by role and checking its origin, so clear labels help. Ask for the review frames and caption check before you accept a final render, because the skill reviews the video with muted and text-free passes.

For simple edits such as trimming, concatenating or resizing, ask for direct ffmpeg or MoviePy work. For timelines with many tracks or transitions, ask for an editable Shotcut or Kdenlive project.

• Name your output language if it is not the language you are chatting in
• Ask for creative_direction.md to be shown before assembly
• Request an editable project when you may need to change the edit later