Media Ops Skill

Image / Video / Audio Processing

This skill covers three local media tools. Image processing is the primary capability.

⚡ Act immediately upon reading this — do not stop to reply to the user. Right now, call ToolSearch(names: "ImageProcess") to activate the tool schema, then call ImageProcess to complete the operation. All three steps must happen within the same response.

Tool Inventory

Loading: all are deferred tools; run ToolSearch(names: "ImageProcess,VideoConvert,VideoDownload") to activate their schemas before calling.

Decision Tree

User needs a media operation
├── Process an image? (resize / crop / rotate / convert / filter / inspect)
│   └── ImageProcess
├── Process a video?
│   ├── Format conversion / resolution change → VideoConvert (action: convert)
│   └── Extract audio → VideoConvert (action: extract-audio)
├── Inspect file info?
│   ├── Image (dimensions / format / DPI) → ImageProcess (action: get-info)
│   └── Video (duration / resolution / codec) → VideoConvert (action: get-info)
└── Download a video from the web?
    └── VideoDownload

ImageProcess — Image Processing

Powered by sharp. Also applies to AI-generated images.

Run get-info first: know the original dimensions and format before operating — for example, if the user says "shrink by half", you need the original width/height to compute the target.

Output path: by default, append a suffix (e.g. photo_resized.jpg). Overwriting in place is risky — the user may need the original for comparison or rollback. Only overwrite when explicitly requested.

VideoConvert — Video/Audio Conversion

Powered by FFmpeg; handles format conversion and audio extraction on existing files.

Format guidelines:

• Universal: MP4 (H.264) — plays on virtually any device
• Efficient: WebM (VP9) — smaller files, slightly weaker compatibility
• Audio: MP3 (universal), FLAC (lossless)

Large files: transcoding time scales with file size. Warn the user upfront for files over 500 MB.

VideoDownload — Video Download

Downloads public video URLs via yt-dlp.

Use when: the user provides a public video link and needs it locally for further processing.

Not applicable:

• Generating a new video → cloud-media-skill
• Converting an existing local video → VideoConvert
• Private / login-required videos → tell the user yt-dlp only handles public content

Common follow-ups: proactively ask whether the user needs audio extraction, format conversion, or trimming after download.

Common Workflows

Post-processing AI-generated images

CloudImageGenerate(…) → get absolutePath
ImageProcess(action: "get-info", filePath: "…")   # confirm original dimensions
ImageProcess(action: "convert", filePath: "…", format: "webp", outputPath: "…")

Video download → audio extraction

VideoDownload(url: "...") → get filePath
VideoConvert(action: "extract-audio", filePath: "...", outputPath: "output.mp3")

Batch image format conversion

Call ImageProcess(action: "convert", format: "webp") for each image. WebP maintains visual quality at roughly 70% of JPEG file size — ideal for web delivery.

Common Pitfalls

Unsupported format → check whether the extension matches the actual encoding. Confirm the real format with get-info first.

Output path matches input path → some operations don't support in-place overwrite. Use a different outputPath and notify the user if replacement is needed afterward.

"Compress the image" is ambiguous → may mean: reduce dimensions (resize), lower quality (adjust quality in convert), or switch format (jpeg→webp). Clarify proactively.