Guizang product video skill: make software update promos from your real product components

Who It's For

• Developers or teams who want to create an update promo video for their software product (default 45–60 seconds, landscape, Chinese) and have access to the real product codebase.
• Its core approach is to reuse real product components and design language, using code to handle storyboard copy, animation, original music, sound effects, and video rendering, ultimately exporting a video plus an editable project. It suits those willing to let an agent read their local codebase and run the rendering pipeline.
• Not suitable if: you don't have a product codebase you can hand to AI; your local environment can't install dependencies like Node.js, Python, or FFmpeg; or you don't want the agent to run installation and rendering commands. The official docs don't specify the exact boundaries of unsupported scenarios — for anything not stated, rely on actual use.
• Supports two agent tools: Claude Code and Codex. Official docs don't mention support for other clients, nor which languages the copy supports (Chinese by default).

First Try in 3 Minutes

• Prerequisites: a supported agent tool installed (Claude Code or Codex). On first use, the skill checks for Node.js, Python, FFmpeg, and rendering dependencies; install whatever's missing via the official installation methods, following your tool's system permission settings when installing system software. The starter project uses React, esbuild, and Playwright to produce frames, then FFmpeg for encoding.
• Installation: run the officially recommended skills CLI command in your terminal:

npx skills add https://github.com/op7418/guizang-product-video-skill --skill guizang-product-video-skill

• An official manual alternative: with Claude Code, clone the repo to ~/.claude/skills/guizang-product-video-skill (Codex uses ~/.agents/skills), then restart your session so the tool picks up the new skill. This is only a fallback note, not a co-equal path with the CLI above.
• Triggering: inside your product repo, hand the codebase to the AI and tell it what this update changed — you can start making a software update promo. The agent will organize selling points, write copy, hook in original components, build animations, compose music, add sound effects, and finally export a video plus an editable project.
• Success signals: once the pipeline finishes, you'll have the exported video file and an editable project. Note this is not a one-off analysis task — it writes copy, generates audio/video, and executes rendering commands; for the first run, it's wise to have it produce a storyboard or copy plan based on your update first, confirm it, and only then move to rendering, to avoid unnecessary generation costs.
• Type of text involved: this is promotional copy for software updates, not fiction writing; the official pipeline centers on organizing selling points and storyboard copy — don't expect literary-narrative-style review processes.

Use Cases

Case 1: Version Update Promo

• Scenario: you've just released a new version and want to quickly produce a landscape promo video.
• How: enter the product repo and tell the skill-supporting agent, "This update added XX feature — please use this skill to make a software update promo video."
• Workflow: the skill follows the official pipeline to organize selling points, write storyboard copy, hook in your real components for animation, write music and sound effects, then render and export.
• Result: you get a finished 45–60 second, Chinese-language video (defaults) plus an editable project you can fine-tune further.

Case 2: Confirm the Plan Before Rendering

• Scenario: you're unsure how to present this update's selling points and want to validate the direction at low cost.
• How: hand the codebase to the AI, describe the update, and ask it to output a storyboard and copy plan without rendering yet.
• What can be confirmed: per the official description, the pipeline's front stage is "organize selling points, write copy," and the back stage is "hook in original components, animate, compose music, add sound effects, export video." Evidence confirms the front stage produces plan text, but the official docs don't describe the exact instruction wording for "copy only, no rendering" — where it stops depends on your actual session.

Usage Boundaries

• Files and commands: the skill reads your product codebase to reuse components and design language, and runs rendering-related commands (esbuild, Playwright, FFmpeg); on first use it also checks and may install missing dependencies. All of this follows your tool's system permission settings — watch for confirmation prompts before execution.
• Network: official docs don't clarify whether the rendering process involves network access — unknown.
• Environment: requires Node.js ≥ 22, Python ≥ 3.9, FFmpeg/ffprobe, and Playwright Chromium; the skill installs whatever's missing via official installation methods.
• Clients: officially supported are Claude Code and Codex; manual install paths differ between the two (~/.claude/skills vs ~/.agents/skills); availability for other clients is unconfirmed.
• Outputs: the exported video and editable project are definite deliverables; whether the skill directly rewrites your source code files isn't stated in official evidence — proceed cautiously.
• Language and duration: defaults are 45–60 seconds, landscape, Chinese; other languages or durations are unconfirmed.