Claude Vision Setup

Instructions

When invoked with /claude-vision-setup:

Step 1: Read Environment Detection

The environment has already been detected by a hook. Check the system-reminder for "Environment detection results:" which contains key=value pairs:

• OS_TYPE - wsl, macos, or linux
• DOCKER_OK - true/false
• GPU_OK - true/false
• GPU_NAME - GPU model name (if detected)
• NVIDIA_TOOLKIT - true/false
• CONFIG_EXISTS - true/false
• IMAGE_VARIANT - lite/full (if config exists)
• SCREENSHOT_DIR - path (if config exists)

Present a summary to the user:

• OS: [detected]
• Docker: Available/Not available
• GPU: [name] with NVIDIA Container Toolkit / Not detected

If CONFIG_EXISTS=true, also show current settings.

Step 2: Existing Config Flow

If CONFIG_EXISTS=true, use AskUserQuestion:

Question: "What would you like to do?"

Options:

1. Keep current settings - Verify everything works
2. Reconfigure - Start fresh with new settings
3. Switch image variant - Change between lite/full
4. Update screenshot directory - Change where screenshots are found

If they choose "Keep current settings", skip to Step 6 (Verify Setup).

Step 3: Image Choice (new config or reconfigure)

Use AskUserQuestion:

Question: "Which Docker image do you want?"

Options:

1. Lite (Recommended)

   • ~500 MB, includes ffmpeg + yt-dlp
   • For YouTube videos, screen recordings

2. Full (with Whisper)

   • ~4 GB, includes ffmpeg + yt-dlp + whisper
   • For local videos needing speech transcription
   • GPU speeds up transcription (detected: [GPU_OK])

Step 4: Screenshot Directory

Based on OS_TYPE, suggest default:

• wsl: /mnt/c/Users/<username>/Pictures/Screenshots
• macos: ~/Pictures/Screenshots
• linux: ~/Pictures/Screenshots

Use AskUserQuestion with suggested default + "Custom path" option.

Step 5: Save Configuration

mkdir -p "$HOME/.claude/claude-vision"

cat > "$HOME/.claude/claude-vision/config.json" << EOF
{
    "mode": "docker",
    "os": "$OS_TYPE",
    "image_variant": "$IMAGE_VARIANT",
    "screenshot_dir": "$SCREENSHOT_DIR",
    "created": "$(date -Iseconds)",
    "version": "1.0"
}
EOF

Step 6: Verify Setup

cv-run --status

If cv-run fails, check if Docker image exists and offer to pull/build.

Step 7: Show Summary

=== claude-vision Setup Complete ===

Configuration:
  OS: $OS_TYPE
  Image: $IMAGE_VARIANT
  Screenshot dir: $SCREENSHOT_DIR
  GPU: $GPU_STATUS

Available commands:
  /clipboard    - Read from clipboard (text or image)
  /screenshot   - Analyze latest screenshot
  /video        - Analyze videos (YouTube or local)

Notes

• Config: ~/.claude/claude-vision/config.json
• Video cache: ~/.claude/claude-vision/video-cache/
• Use cv-run --status to check container status
• Use cv-run --stop to stop container when not in use