How to Install the Watch Skill for Various AI Assistants

The watch skill is a self-contained package located in skills/watch/ that can be installed via the built-in plugin manager for Claude Code, the universal npx skills CLI for editors like Cursor and Copilot, or as a bundled archive for claude.ai web, requiring no host-specific environment variables.

The bradautomates/claude-video repository provides a harness-agnostic video analysis tool that processes media through yt-dlp, ffmpeg, and optional Whisper transcription. Because the skill computes its base directory from the location of SKILL.md rather than relying on host-specific variables like CLAUDE_SKILL_DIR, the same installation folder works across every supported platform.

Installation via Claude Code Plugin Manager

For Claude Code, use the built-in marketplace to install the watch skill directly from the repository.

1. Add the marketplace entry:

   /plugin marketplace add bradautomates/claude-video

2. Install the specific skill:

   /plugin install watch@claude-video

This command copies the skills/watch/ folder into Claude Code's plugin directory and registers the /watch slash command immediately.

Installation for Codex, Cursor, Copilot, and 50+ Other Hosts

For AI assistants that support the universal skills protocol—including Codex, Cursor, Copilot, and over 50 other hosts—use the npx CLI to install the skill globally.

Run the following command in your terminal:

npx skills add bradautomates/claude-video -g

The -g flag copies the entire skills/watch/ directory into the host's skill folder, making the /watch command available without additional configuration.

Installation for claude.ai Web Interface

When working with the claude.ai web interface, you must upload the skill as a bundled archive.

First, build the distributable package:

bash skills/watch/scripts/build-skill.sh

This produces dist/watch.skill, a zip archive containing the complete skill structure. Then upload this file via the claude.ai "Upload Skill" UI to activate the functionality.

Internal Skill Structure and Key Files

Regardless of the installation method, the watch skill deploys the identical folder layout:

skills/
 └─watch/
    ├─SKILL.md          ← Canonical contract defining the /watch command
    └─scripts/
        ├─watch.py      ← Main orchestrator for download → frames → transcript
        ├─download.py   ← yt-dlp wrapper for fetching video
        ├─frames.py     ← ffmpeg frame extraction with auto-FPS detection
        ├─transcribe.py ← Whisper API integration
        ├─setup.py      ← Pre-flight checks for dependencies
        └─config.py     ← Configuration management
    └─scripts/build-skill.sh  ← Archive builder for web uploads

As implemented in bradautomates/claude-video, the entry point at skills/watch/scripts/watch.py coordinates the video processing pipeline, while SKILL.md serves as the canonical contract that declares the slash command name and its invocability.

Using the Watch Skill After Installation

Once installed, invoke the skill using the slash command:

/watch <YouTube-URL-or-local-path> [optional-question]

The skill executes the following sequence defined in watch.py:

• Download: Fetches the video via download.py using yt-dlp
• Frame Extraction: Processes frames via frames.py using ffmpeg with automatic FPS detection
• Transcription: Optionally generates captions via transcribe.py using the Whisper API

Summary

• The watch skill lives in skills/watch/ and is harness-agnostic, working across Claude Code, Cursor, Copilot, and claude.ai without modification.
• Claude Code users install via /plugin marketplace add followed by /plugin install watch@claude-video.
• Codex/Cursor/Copilot users run npx skills add bradautomates/claude-video -g to copy the skill folder globally.
• claude.ai web users must run bash skills/watch/scripts/build-skill.sh to create dist/watch.skill, then upload through the web UI.
• The skill computes its own path from SKILL.md location, eliminating dependency on host-specific environment variables like CLAUDE_SKILL_DIR.

Frequently Asked Questions

Do I need to set environment variables like CLAUDE_SKILL_DIR to install the watch skill?

No. The watch skill is designed to be harness-agnostic and computes its base directory from the location of SKILL.md rather than relying on host-specific environment variables. This ensures the same installation works across Claude Code, Cursor, Copilot, and other platforms without configuration changes.

What dependencies does the watch skill require?

The skill requires yt-dlp for video downloading, ffmpeg for frame extraction, and optionally the Whisper API for transcription. The setup.py script in skills/watch/scripts/ performs pre-flight checks to verify these dependencies are installed before execution.

Can I install the watch skill on AI assistants not listed in the installation methods?

Yes. Any AI assistant that supports the universal skills protocol can install the watch skill using the npx skills add bradautomates/claude-video -g command. The skill's self-contained structure in skills/watch/ ensures compatibility with any host that can copy the folder into its plugin directory.

How do I update the watch skill to the latest version?

For Claude Code, reinstall via /plugin install watch@claude-video after updating the marketplace entry. For npx-based installations, run the npx skills add command again with the -g flag to overwrite the existing skill folder. For claude.ai web, rebuild the archive using skills/watch/scripts/build-skill.sh and re-upload the generated dist/watch.skill file.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}