frames-cli - SKILL.md Agent Skill

name: frames-cli description: Frame screenshots and screen recordings with the `frames` CLI. Use this skill when the user asks to add Apple device bezels, frame screenshots or videos, create device mockups, inspect screenshot/device matches, or batch-process screenshot/video folders with the command line.

Apple Frames CLI

frames 1.3.0 is a single-file Python CLI that applies Apple device bezels to screenshots and videos, auto-detects devices from input dimensions, applies masks when needed, and can merge multiple framed results. Video support uses external ffmpeg 5.1+ and ffprobe 5.1+ with no extra Python media stack.

What Agents Should Know

frames is the default command. frames screenshot.png and frames frame screenshot.png are equivalent.
Use frames --json ... for automation. Put global flags such as --json, --assets, and --verbose before subcommands for the broadest compatibility.
Device support comes from the installed asset bundle, not hardcoded names in this skill. Use frames list and frames list-colors as the source of truth when exact names matter.
On macOS, the default asset location is the Apple Frames shortcut folder in iCloud Drive. That avoids downloading a second copy when the user already has the shortcut assets installed.
Use frames video ... for .mp4, .mov, and .m4v; the default frames ... path is for images.
Use frames video-info ... to probe videos and resolve the matching frame metadata without rendering.
Video framing requires external ffmpeg 5.1+ and ffprobe 5.1+. frames setup checks this and can offer a Homebrew install on macOS; frames doctor reports video tool readiness.
frames video preserves single-video audio by default. Pass --strip-audio when the user asks for silent output, privacy-safe delivery, or validation clips where audio does not matter.
frames video --preset compact|balanced|best controls MP4 export size/quality; best is the default. Video exports report output file size and source-vs-output savings in human and JSON output.
Use frames video --alpha ... or frames video --background transparent ... for transparent ProRes 4444 .mov output. This works for single videos and merged videos; MP4/H.264 and MP4/HEVC do not support alpha.
Explicit transparent output file paths must use a .mov extension. Use an output directory when you want the CLI to pick the .mov filename.

Quick Reference

# Frame one screenshot
frames screenshot.png

# Frame every top-level image in a folder
frames ~/Screenshots/

# Pick a specific color or randomize colors
frames -c "Cosmic Orange" screenshot.png
frames -c random *.png
frames --colors "Silver,Space Black,random" one.png two.png three.png

# Frame videos
frames video-info recording.mp4
frames --json video-info recording.mp4
frames video recording.mp4
frames video --preset compact recording.mp4
frames video --preset balanced recording.mp4
frames video --preset best recording.mp4
frames video --alpha recording.mp4
frames video -m --alpha 1.mp4 2.mp4
frames video -m --background transparent 1.mp4 2.mp4
frames --json video --strip-audio recording.mp4
frames video --strip-audio recording.mp4
frames video --background "#f5f5f5" --codec hevc recording.mp4
frames video -m --playback-offset 1.mp4 2.mp4
frames video -m --colors "Silver,random" 1.mp4 2.mp4

# Force an exact frame instead of auto-resolving the newest variant
frames -d "iPhone 15 Pro Portrait" screenshot.png

# Merge framed outputs side by side
frames -m shot1.png shot2.png

# Merge without proportional physical scaling
frames -m --no-scale shot1.png shot2.png

# Batch merge sequential groups
frames -b 3 *.png

# Save to a separate output directory
frames -o ~/Framed *.png

# Save to a subfolder next to originals
frames -f *.png
frames --subfolder mockups *.png

# Inspect matches without framing
frames info screenshot.png
frames --json info ~/Screenshots/

# Discover supported devices and colors
frames list
frames list-colors "17 Pro"

# Manage default colors
frames colors

# Diagnose asset/config problems
frames doctor

# Download or re-point assets
frames setup
frames setup /path/to/Frames

Current Feature Surface

Auto-detect devices from screenshot width, with height-based overlap disambiguation where needed.
Resolve width-sharing devices to the newest default frame automatically. Pass --device to skip variant resolution.
Apply frame colors by exact name, 1-based numeric index, default, or random.
Use --colors for per-input colors on images and videos; it maps comma-separated values to expanded inputs by order.
Apply masks when the asset entry requires clipping.
Merge multiple framed outputs with physical-size normalization by default.
Frame .mp4, .mov, and .m4v files through frames video; single-video audio is preserved unless --strip-audio is passed.
Use --preset compact, --preset balanced, or --preset best to tune MP4 export size/quality. best is the default.
Report output file size and savings after video export in both human output and --json agent output.
Use --alpha or --background transparent to create transparent ProRes 4444 MOV output for single or merged videos.
Show a live terminal progress bar during video renders in interactive mode; JSON and non-TTY runs stay pipeline-friendly.
Inspect video/device matches without rendering through frames video-info.
Merge videos simultaneously with frames video -m, or sequentially left-to-right with frames video -m --playback-offset.
Use --no-scale to disable proportional scaling and keep native framed sizes when merging.
Use --batch N to merge sequential groups. --batch implies merge and N must be at least 2.
Save next to originals, into --output, or into a validated single-name subfolder via -f or --subfolder.
Copy a single output to the macOS clipboard with --copy.
Inspect screenshots and folders with info.
Diagnose assets and config with doctor.
Save per-device default colors with colors.

Command Notes

Global flags:
- --json for machine-readable output
- --no-color to disable ANSI color
- --assets PATH to override asset discovery
- -v / --verbose for variant, resize, and mask details
Use frames video --help and frames video-info --help for the human-facing video examples and option summaries.
frames colors is interactive in a real terminal. If stdin is not a TTY, or curses is unavailable, it falls back to a plain printed color list.
frames list-colors does partial device-name matching. frames list-colors "17 Pro" is valid.
Directory inputs are expanded one level deep only. The CLI scans top-level image files inside the directory; it is not recursive.
Supported input image extensions are .png, .jpg, .jpeg, .heic, .tiff, and .webp.
Supported input video extensions are .mp4, .mov, and .m4v.
Images larger than 20,000px on either side trigger a warning. Images larger than 50,000px are rejected.
--subfolder only accepts a single directory name, not a path like ../out or foo/bar.

Video Notes

frames video-info FILE reports dimensions, duration, fps, codec, audio state, matched device, selected color, frame size, mask state, and resize metadata without creating an output video.
frames video FILE writes FILE_framed.mp4 by default, or .mov when --alpha, --codec prores, or --background transparent is used.
--alpha defaults the background to transparent unless the user explicitly passes another --background.
Explicit alpha/transparent output files must end in .mov; otherwise frames video fails before rendering.
For one video, --output may be an explicit output file path or a directory. For multiple individual videos, use an output directory.
--merge creates one horizontal video. Without --playback-offset, videos play simultaneously and duration is the longest input.
--playback-offset requires --merge; videos play left to right, inactive future videos hold their first frame, and completed videos hold their final frame.
Merged videos are proportionally scaled by device physicalHeight and bottom-aligned by default. Use --no-scale only when native framed pixel sizes matter more than physical proportion.
Single-video output preserves audio unless --strip-audio is passed. Sequential --playback-offset merges concatenate audio and generate silence for inputs without audio. Simultaneous merges omit mixed audio in this version.
--background accepts only white, black, transparent, or #RRGGBB. transparent implies alpha/ProRes MOV output.
--codec h264 is the default. --codec hevc is smaller but may be slower or less compatible. Use --codec prores or --alpha for transparent .mov.
Transparent merged videos preserve alpha with ProRes 4444 (yuva444p10le), suitable for compositing in video editors and presentation tools.
--preset compact|balanced|best controls H.264/HEVC hardware bitrate and software CRF. best is the default; balanced and compact trade quality for smaller files. ProRes/alpha output keeps ProRes settings.
--quality N is an expert software CRF override; lower is higher quality.
--color random randomizes independently per input. --colors "A,B,random" maps values to expanded inputs by order and the count must match after directory expansion.
video-info accepts --device, --color, and --colors; it uses the same resolution rules as frames video.
Taildrop or other delivery is separate from frames; render first, verify the output, then use the delivery workflow the user requested.

JSON Output Behavior

frames --json screenshot.png returns one framed-image object with fields such as source, device, color, dimensions, frame_size, resized, masked, physicalHeight, and output.
frames --json *.png returns { "frames": [...] } for multiple individual outputs.
frames --json -m ... returns a merged result with merged, count, and frames.
frames --json -b N ... returns batches, batch_size, total, and frames.
When proportional merge scaling is applied, per-frame scale_factor values are included in the JSON output.
frames --json info ... returns either one object or a list of objects, including device, primary_match, colors, color_count, has_mask, resize_width, variants, and is_variant.
frames --json video-info ... returns one video metadata object, or a list for multiple videos, including dimensions, duration, fps, codec, audio, device, primary_match, color, frame_size, resize_width, and mask_missing.
frames --json doctor returns asset path/source/version, PNG count, config presence, issues, notes, and suggested next steps.
frames --json video ... returns one video object, or { "videos": [...] } for multiple individual outputs, including preset, output_size_bytes, output_size, source_size_bytes, source_size, savings_bytes, savings, and savings_percent.
frames --json video -m ... returns merged, duration, dimensions, playback_offset, audio state, top-level output size/savings fields, and per-input frames. For transparent exports, alpha is true on each frame and background reports transparent unless another background was explicitly requested.

Assets and Config

Config file:

~/.config/frames/config.json

Config keys:

assets_path: explicit asset folder path
default_colors: saved default frame colors by base device name
use_subfolder: true for framed, or a custom folder name string

Asset resolution order:

--assets
FRAMES_ASSETS
assets_path from config
macOS default: ~/Library/Mobile Documents/iCloud~is~workflow~my~workflows/Documents/Frames
macOS fallback if valid: ~/Library/Application Support/frames/assets
On non-macOS platforms, the standard per-platform app-data directory is the default

Setup behavior:

frames setup downloads the current asset archive from https://cdn.macstories.net/AppleFrames401.zip.
frames setup /path/to/Frames points the CLI at an existing asset folder instead of downloading.
The asset folder must contain NewFrames.json, version.txt, and the frame/mask PNGs.
Setup checks ffmpeg/ffprobe for video framing and can install ffmpeg through Homebrew on macOS when run interactively.
frames setup --subfolder and frames setup --no-subfolder update the default save behavior in config.
frames doctor is the first command to run when assets were moved, edited, corrupted, or when video dependencies need verification.

Current Supported Device Families

The current v4 asset bundle used by frames 1.3.0 includes these primary families:

iPhone: iPhone 17, iPhone 17 Pro, iPhone 17 Pro Max, iPhone Air, iPhone 16, iPhone 16 Plus, iPhone 12-13 Pro, iPhone 12-13 Pro Max, iPhone 12-13 mini, iPhone 8 / 2020 SE
iPad: iPad mini 2021, iPad 2021, iPad Air 2020, iPad Pro 2018-2021 11-inch, iPad Pro 2018-2021 12.9-inch, iPad Pro 2024 11-inch, iPad Pro 2024 13-inch
Mac: MacBook Neo, MacBook Pro 13, MacBook Air 2020, MacBook Air M5 13, MacBook Air M5 15, MacBook Pro M5 14, MacBook Pro M5 16, iMac M4, Studio Display, Studio Display XDR
Watch: Watch Series 7 41, Watch Series 7 45, Watch Series 11 42, Watch Series 11 46, Watch Ultra 2024, Watch Ultra 3

Current default variant resolution favors the newest matching frame for shared screenshot sizes:

iPhone 17 Portrait resolves to iPhone 17 Pro Portrait
iPhone 17 Landscape resolves to iPhone 17 Pro Landscape
iPhone 17 Pro Max shares sizes with iPhone 16 Pro Max
iPhone 16 shares sizes with iPhone 15 Pro
iPhone 16 Plus shares sizes with iPhone 15 Pro Max
MacBook Pro M5 shares sizes with MacBook Pro 2021
MacBook Air M5 13 shares sizes with MacBook Air 2022
Studio Display shares sizes with Studio Display XDR
Watch Series 11 shares sizes with Watch Series 10

If an exact older variant is required, pass --device with the exact frame name instead of relying on auto-detection.

Recommended Agent Workflow

Run frames --json info ... first when you do not control the screenshot source. For videos, run frames --json video-info ... when you need dimensions, duration, audio state, or the matched frame before rendering.
If the user asks for a specific device or color, confirm the exact names with frames list or frames list-colors.
Use frames doctor before changing config or re-downloading assets.
Use --device only when the user wants an exact frame, older variant, or a non-default shared-size match.
Use --no-scale only when the user explicitly wants native framed sizes instead of physically proportionate merges.
For multi-video previews, choose frames video -m for simultaneous playback and frames video -m --playback-offset for left-to-right sequential playback.
Prefer --strip-audio for generated test artifacts unless the user specifically wants audio preserved.
After video rendering, verify with ffprobe when final output audio state, duration, codec, or stream count matters.