danielrosehill/claude-video-editor-plugin/deps-setup

Deps Setup

Optional installer for third-party video-editing tools. The user picks a subset; each is installed using its native ecosystem's tooling and recorded in preferences.json so other skills (and future skills) can find it.

Python-based tools share a single uv-managed virtualenv at:

${CLAUDE_USER_DATA:-${XDG_DATA_HOME:-$HOME/.local/share}/claude-plugins}/video-editing/venv

The venv path is stored as preferences.python_venv so it can be sourced later (e.g. "$PYTHON_VENV/bin/python" -m moviepy ...).

Tools

Procedure

1. Resolve config

DATA_DIR="${CLAUDE_USER_DATA:-${XDG_DATA_HOME:-$HOME/.local/share}/claude-plugins}/video-editing"
PREFS_FILE="$DATA_DIR/preferences.json"
TOOLS_DIR="$DATA_DIR/tools"
VENV_DIR="$DATA_DIR/venv"
mkdir -p "$DATA_DIR" "$TOOLS_DIR"
test -f "$PREFS_FILE" || echo '{}' > "$PREFS_FILE"

2. Bootstrap uv (if any Python tool selected)

if ! command -v uv >/dev/null; then
  echo "uv not installed. Run: curl -LsSf https://astral.sh/uv/install.sh | sh"
  # Ask user before running. Don't silently curl|sh.
fi

Confirm with the user before running the install script.

3. Create the shared venv (lazy)

if [ ! -d "$VENV_DIR" ]; then
  uv venv "$VENV_DIR"
fi

4. Ask the user which to install

Multi-select. Show each tool's purpose in one line:

• moviepy — Python video editing library (programmatic clip ops)
• auto-editor — auto-cut silences and dead air from video
• video-use — agentic video understanding from browser-use
• VideoAgent — research multi-agent video understanding (HKU)
• vit — video transformer toolkit (LucasHJin)
• faster-whisper — fast CPU/GPU whisper inference (used by burn-subtitles, karaoke-video)
• demucs — Facebook Research stem separation (used by karaoke-video)
• LosslessCut — GUI for fast lossless trim/cut
• editly — declarative JSON → video composition (Node)
• whisper.cpp — alternative pure-C++ whisper backend (manual install, see notes)
• exiftool — uniform metadata reader for both images and video (sudo apt install -y libimage-exiftool-perl). Single tool covers ImageWidth, ImageHeight, Rotation, EXIF on JPEG/HEIC/DNG/WEBP/MP4/MOV — useful as a uniform parser when ffprobe's CSV output is fragile.
• mediainfo — purpose-built media-inspection CLI (sudo apt install -y mediainfo). Reports DisplayAspectRatio and Rotation already corrected for what the player would show; cleanest single source of truth for video orientation.

5. Install per selection

For each chosen tool, show the exact command first, ask for confirmation, run, capture exit status.

Python (PyPI)

uv pip install --python "$VENV_DIR/bin/python" moviepy

Python (git clone + requirements)

git clone https://github.com/HKUDS/VideoAgent "$TOOLS_DIR/VideoAgent"
uv pip install --python "$VENV_DIR/bin/python" -r "$TOOLS_DIR/VideoAgent/requirements.txt"

LosslessCut (flatpak)

if command -v flatpak >/dev/null; then
  flatpak install -y --user flathub no.mifi.losslesscut
else
  echo "flatpak missing — fall back to AppImage download from GitHub releases"
fi

editly (npm)

if ! command -v npm >/dev/null; then
  echo "Node.js not installed. Install Node first (see nodejs.org)."
  # Don't auto-install Node — too invasive for a video plugin.
fi
npm install -g editly

exiftool / mediainfo (apt)

sudo apt install -y libimage-exiftool-perl mediainfo

Verify with exiftool -ver and mediainfo --Version. Record absolute paths in preferences.json under tools.exiftool.path and tools.mediainfo.path (command -v exiftool, command -v mediainfo) so other skills (and other plugins, e.g. phone-ops:import-media) can locate them without re-probing.

6. Record results

Update preferences.json:

{
  "python_venv": "/abs/.../venv",
  "tools": {
    "moviepy":     { "installed": true, "method": "uv-pip", "venv": "...", "version": "1.0.3" },
    "auto-editor": { "installed": true, "method": "uv-pip", "venv": "...", "version": "..." },
    "VideoAgent":  { "installed": true, "method": "git+uv-pip", "path": "...tools/VideoAgent" },
    "lossless-cut":{ "installed": true, "method": "flatpak", "id": "no.mifi.losslesscut" },
    "editly":      { "installed": true, "method": "npm-global", "version": "..." },
    "faster-whisper": { "installed": true, "method": "uv-pip", "venv": "..." },
    "demucs":      { "installed": true, "method": "uv-pip", "venv": "..." },
    "exiftool":    { "installed": true, "method": "apt", "path": "/usr/bin/exiftool", "version": "..." },
    "mediainfo":   { "installed": true, "method": "apt", "path": "/usr/bin/mediainfo", "version": "..." }
  }
}

For whisper.cpp, deps-setup does not auto-install (build environment varies — clang vs gcc, BLAS deps, AVX support). Print a one-block install hint and ask the user to run it themselves, then re-run onboard to register the binary + model paths:

git clone https://github.com/ggerganov/whisper.cpp ~/whisper.cpp
cd ~/whisper.cpp && make
bash ./models/download-ggml-model.sh small.en
# then: re-run onboard and point subtitle_backend=whisper.cpp at:
#   subtitle_model_path=~/whisper.cpp/models/ggml-small.en.bin
#   binary on PATH: ~/whisper.cpp/main  (newer builds: whisper-cli)

Write atomically (.tmp + mv).

7. Report

Per tool: success / failed / skipped. For failures, surface stderr and a one-line hint (e.g. "Cairo dev headers missing — apt install libcairo2-dev"). Don't pretend a tool is installed if its command failed.