New: 98,000+ skills indexed from GitHub repos with 100+ stars · every skill safety-checkedBrowse top skillsTop skills
Esc to closeBrowse all skills
Submit a skillBrowse skills
Submit a skillBrowse skills

Agent skill

Watch Together

by vellum-ai in vellum-ai/vellum-assistant

Watch local TV shows and movies with the user in real time. An agent skill from vellum-ai/vellum-assistant.

MITAuto-check: notesMedia & Creative

Install Watch Together

skills CLI
$ npx skills add vellum-ai/vellum-assistant --skill watch-together -a claude-code

Project install by default; add -g for ~/.claude/skills/.

GitHub CLI
$ gh skill install vellum-ai/vellum-assistant watch-together --agent claude-code

Project scope by default; add --scope user for a personal install. Needs GitHub CLI 2.90.0 or later (public preview).

Manual copy
$ git clone --depth 1 https://github.com/vellum-ai/vellum-assistant.git skills-src && mkdir -p .claude/skills && cp -r skills-src/skills/watch-together .claude/skills/watch-together && rm -rf skills-src

Use ~/.claude/skills/ instead of .claude/skills for a personal install. The folder must contain SKILL.md.

Claude Code skills documentation · loads skills from .claude/skills/

Facts

Skill name
watch-together
GitHub stars
1.4k
Token cost
~2.1k tokens
SKILL.md length
829 words
Files
7 (incl. scripts)
Skills in repo
108
Repo updated
First seen
Licence
MIT

At a glance

Watch local TV shows and movies with the user in real time. An agent skill from vellum-ai/vellum-assistant.

  • Works in 4 steps: Playback is divided into 60-second… → editor.py reviews each window, selects… → A wake extracts the selected frames at… → …
  • Tasks that involve Transcription
  • SKILL.md covers Choose a mode, How it works, When a [WATCH] message arrives and Start a session, plus 4 more sections
  • Runs Python and Shell scripts from its folder; calls python3, ffmpeg and brew; needs GEMINI_API_KEY

What it does

Watch Together is an agent skill from vellum-ai/vellum-assistant. Watch local TV shows and movies with the user in real time. An editor model watches each playback window, transcribes dialogue, and picks story-critical frames. Source mode plays local media in mpv; screen mode captures supported desktops as a fallback.

Its SKILL.md is about 2.1k tokens, which your agent loads only when the skill is triggered. The skill folder holds 7 other files, including scripts (for example `scripts/capture-live.sh`, `scripts/capture_live.py` and `scripts/editor.py`).

It sits in Media & Creative, covering Transcription. The repository describes itself as: An AI Assistant that’s easy to setup, does your work 24/7, knows your preferences and gets better over time. The licence is MIT.

When your agent uses it

  • Tasks that involve Transcription

Example prompts

  • “/watch-together”

Requirements

  • Python 3
  • A Bash shell
  • A credential in GEMINI_API_KEY

Workflow steps

4 steps, taken from the first numbered list in SKILL.md.

  1. Playback is divided into 60-second windows.
  2. editor.py reviews each window, selects story-critical frames, collects
  3. A wake extracts the selected frames at 720p and sends a [WATCH] message to
  4. The assistant looks at the frames and dialogue and reacts only when useful.

What it can do on your machine

Read from SKILL.md and the folder at commit 33cc983. It shows what the files ask for, not the result of running them.

  • Tool permissions

    Pre-approves nothing: there is no allowed-tools line, so your agent's usual permission prompts apply.

    From allowed-tools in the SKILL.md frontmatter.

  • Runs code

    Ships 6 files in scripts/ (Python and Shell), which the agent can run.

    Shell commands in SKILL.md call:

    • python3
    • ffmpeg
    • brew
    • ffprobe
    • apt
    • dnf
    • python

    From the folder's file list and the shell code blocks in SKILL.md.

  • Network

    No URLs in SKILL.md.

    From URLs in SKILL.md, links to its own repository left out.

  • Credentials

    Names these keys or tokens, usually read from environment variables:

    • GEMINI_API_KEY

    From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.

Context cost

Watch Together loads about 2.1k tokens when it runs. Until then it costs about 67 tokens; SKILL.md has 829 words of instructions outside code blocks.

Always · name and description, kept in context so the agent knows when to use it
~67
When it runs · the whole SKILL.md, loaded when a task matches
~2.1k

Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.

Safety

Auto-check: notes

The automated check noted patterns worth knowing about, such as sudo or a known installer.

  • NoteRuns commands with sudoSKILL.md:237
    sudo apt install python3 ffmpeg mpv
  • NoteRuns commands with sudoSKILL.md:243
    sudo dnf install python3 ffmpeg mpv

Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); the scripts in this folder are not scanned.

SKILL.md

The full file from vellum-ai/vellum-assistant at commit 33cc983, republished under its MIT licence (© vellum-ai). 829 words, ~2,113 tokens.

Download SKILL.mdSave it as .claude/skills/watch-together/SKILL.md (or your agent's skills folder). This skill also uses 6 other files; get the full folder from GitHub.
name
watch-together
description
Watch local TV shows and movies with the user in real time. An editor model watches each playback window, transcribes dialogue, and picks story-critical frames. Source mode plays local media in mpv; screen mode captures supported desktops as a fallback.
metadata.emoji
📺

Watch Together

Real-time co-watching. An editor model watches playback continuously and wakes the assistant only at moments worth reacting to. Each wake includes dialogue and story-critical frames. Wakes arrive through signal files, so do not poll.

Choose a mode

Source mode (preferred)

watch-file.py plays a local media file in mpv and reads that file directly. It does not capture the screen or require a loopback audio driver. Frames come from the source at full quality. Sidecar .srt subtitles or the first embedded text subtitle track provide exact dialogue. Without subtitles, the editor transcribes audio. Pause stops the flow and seeks resynchronize it.

Use source mode whenever the user has a local media file. The bundled script does not accept stream URLs.

Screen mode (fallback)

capture_live.py records supported desktops through ffmpeg and feeds completed segments to the same editor pipeline.

  • macOS: Uses AVFoundation. It auto-detects BlackHole for audio and records video only when BlackHole is unavailable.
  • Windows: Uses ffmpeg gdigrab. It records video only by default. System audio requires an existing DirectShow playback source such as Stereo Mix or a configured virtual audio cable, passed explicitly as the audio device.
  • Linux: Uses X11 capture and $DISPLAY. It records video only by default. System audio requires an existing PulseAudio or PipeWire-Pulse monitor source passed explicitly. Native Wayland screen capture is not supported by this script.

Screen capture cannot bypass DRM or protected-video blanking. If browser video is black in the recording, explain that screen mode is unsupported for that content. Do not claim the fallback works around content protection.

How it works

  1. Playback is divided into 60-second windows.
  2. editor.py reviews each window, selects story-critical frames, collects dialogue, and decides whether to wake now or hold a developing moment.
  3. A wake extracts the selected frames at 720p and sends a [WATCH] message to the active conversation.
  4. The assistant looks at the frames and dialogue and reacts only when useful.

The editor chooses when the assistant looks and what evidence it receives. The assistant decides whether to speak and how to react. A hard cap controlled by WATCH_MAX_HOLD prevents indefinite holds.

When a [WATCH] message arrives

  1. Inspect the attached frames with vision.
  2. Read the dialogue as the primary source for the window.
  3. React naturally. A short line, a stage direction, or silence can all be appropriate. Reserve longer reactions for moments that earn them.
  4. Track theories, callbacks, and running jokes in replies so memory can carry them between sessions.

If a moment needs closer inspection, extract dense 720p frames from the source named in the wake.

macOS or Linux:

bash
python3 "$VELLUM_WORKSPACE_DIR/watch-together/scripts/rewind.py" \
  <source-file> <output-dir> <start-seconds> <end-seconds>

The existing rewind.sh command remains a POSIX compatibility wrapper.

Windows PowerShell:

powershell
py -3 "$env:VELLUM_WORKSPACE_DIR\watch-together\scripts\rewind.py" `
  <source-file> <output-dir> <start-seconds> <end-seconds>

If the Python launcher is unavailable but python --version reports Python 3, replace py -3 with python.

Start a session

Choose a stable lowercase session ID such as example-show-s1e1. Do not include personal data in it.

Show full SKILL.md (349 more words)Show less
macOS or Linux

Create the session directory:

bash
SESSION_ID="example-show-s1e1"
mkdir -p "$VELLUM_WORKSPACE_DIR/watch-together/sessions/$SESSION_ID"

Source mode:

bash
python3 "$VELLUM_WORKSPACE_DIR/watch-together/scripts/watch-file.py" \
  <local-media-file> \
  "$VELLUM_WORKSPACE_DIR/watch-together/sessions/$SESSION_ID" \
  <conversation-id>

Screen mode:

bash
python3 "$VELLUM_WORKSPACE_DIR/watch-together/scripts/capture_live.py" \
  "$VELLUM_WORKSPACE_DIR/watch-together/sessions/$SESSION_ID" \
  <conversation-id>

On macOS, the optional positional arguments are chunk seconds, AVFoundation screen index, and AVFoundation audio index. The default screen index remains 2, and BlackHole is auto-detected when no audio index is supplied.

On Linux, $DISPLAY is used by default. To add system audio, pass chunk seconds, the X11 display, and a Pulse monitor source:

bash
python3 "$VELLUM_WORKSPACE_DIR/watch-together/scripts/capture_live.py" \
  "$VELLUM_WORKSPACE_DIR/watch-together/sessions/$SESSION_ID" \
  <conversation-id> 60 "$DISPLAY" <pulse-monitor-source>
Windows PowerShell

Create the session directory:

powershell
$SessionId = "example-show-s1e1"
$SessionDir = Join-Path $env:VELLUM_WORKSPACE_DIR "watch-together\sessions\$SessionId"
New-Item -ItemType Directory -Force $SessionDir | Out-Null

Source mode:

powershell
py -3 "$env:VELLUM_WORKSPACE_DIR\watch-together\scripts\watch-file.py" `
  <local-media-file> $SessionDir <conversation-id>

Screen mode, video only:

powershell
py -3 "$env:VELLUM_WORKSPACE_DIR\watch-together\scripts\capture_live.py" `
  $SessionDir <conversation-id>

Screen mode with a configured DirectShow playback source:

powershell
ffmpeg -list_devices true -f dshow -i dummy
py -3 "$env:VELLUM_WORKSPACE_DIR\watch-together\scripts\capture_live.py" `
  $SessionDir <conversation-id> 60 desktop "<DirectShow-audio-device>"

Do not describe the optional Windows audio path as automatic loopback. The user must already have a playback capture source exposed through DirectShow.

After giving the matching command, tell the user to start the show. When mpv exits or screen capture stops, the final window is flushed.

The conversation ID is the bare UUID from the conversation database record, not the timestamped conversation folder name. Using the folder name routes the wake incorrectly.

Cost setup

Watching produces many small turns. Recommend normal inference mode and an inference profile with a reduced context ceiling so the conversation compacts regularly. Example:

jsonc
"llm": {
  "profiles": {
    "watch-mode": {
      "contextWindow": { "maxInputTokens": 200000 }
    }
  }
}

Environment variables

  • GEMINI_API_KEY: Enables editor verdicts and transcription. Without it, fixed-cadence wakes use evenly spaced frames and no editor transcription.
  • GEMINI_MODEL: Editor model. Defaults to gemini-3.7-flash.
  • WATCH_MAX_HOLD: Maximum seconds between wakes. Defaults to 240.
  • WATCH_MAX_FRAMES: Maximum frames attached per wake. Defaults to 8.
  • WATCH_MPV_ARGS: Extra mpv arguments for source mode. Quoted arguments use the current platform's command-line rules.

Prerequisites

All modes require Python 3 and ffmpeg. Source mode also requires mpv. Confirm each command is on PATH before starting.

macOS:

bash
brew install ffmpeg mpv
brew install blackhole-2ch  # optional screen-mode system audio

Configure a Multi-Output Device in Audio MIDI Setup when using BlackHole.

Windows: Install current Python 3, ffmpeg, and mpv builds, add them to PATH, then verify:

powershell
py -3 --version
ffmpeg -version
ffprobe -version
mpv --version

Linux: Install the distro packages. For Debian or Ubuntu:

bash
sudo apt install python3 ffmpeg mpv

For Fedora:

bash
sudo dnf install python3 ffmpeg mpv

File locations

  • Scripts: $VELLUM_WORKSPACE_DIR/watch-together/scripts/
  • Sessions: $VELLUM_WORKSPACE_DIR/watch-together/sessions/<session-id>/
    • chunks/: Recorded segments from screen mode
    • editor/verdicts/: Per-window editor output
    • wakes/wake-NNN/: Frames attached to each wake
    • editor-state.json: Held-window state
    • subs.srt: Extracted source-mode subtitles
    • mpv.sock: POSIX source-mode IPC socket; Windows uses a named pipe
  • Signals: JSON files at $VELLUM_WORKSPACE_DIR/signals/user-message.<requestId> with optional image attachments

© vellum-ai, MIT. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file

Files

SKILL.md and 6 other files (scripts) in skills/watch-together of vellum-ai/vellum-assistant.

  • SKILL.md
  • scripts/capture-live.sh
  • scripts/capture_live.py
  • scripts/editor.py
  • scripts/rewind.py
  • scripts/rewind.sh
  • scripts/watch-file.py

Open the folder on GitHubat commit 33cc983

Compare with similar skills

Watch Together next to the 5 skills that share the most tags, products or categories with it. Stars are the repository's; “used in” counts other GitHub owners with a copy.

Watch Together compared with similar skills
SkillStarsUsed inTokensAuto-checkLicenceRepo updated
Watch Together this skillvellum-ai/vellum-assistant1.4k—~2.1kAutomated safety check: NotesMIT
HyperFrames Media Useheygen-com/hyperframes60k—~2.4kAutomated safety check: PassApache-2.0
Native Subtitle Quote Imagechengyi-ai/native-subtitle-quote-image2.6k—~2.4kAutomated safety check: PassMIT
Edu Chem Videowy51ai/edulab1.4k—~2.1kAutomated safety check: NotesApache-2.0
Transcription Memory ReconstructionNxcoreAI/EverRoom3k—~714Automated safety check: PassCustom licence
Edu Math Videowy51ai/edulab1.4k—~2.5kAutomated safety check: NotesApache-2.0

Similar skills

  • HyperFrames Media Use

    heygen-com/hyperframes

    Finds, generates and edits media for HyperFrames video projects: music, sound effects, images, icons, logos, voiceovers, captions and color grades.

    60k GitHub stars~2.4k tokensUpdated today
    Media & CreativeAuto-check passed

  • Native Subtitle Quote Image

    chengyi-ai/native-subtitle-quote-image

    将本地视频或用户有权处理的在线视频，经过来源获取、文字稿定位、选题选句、精确取帧、紧凑裁切、拼图和逐张质检，制作成 3:4 或保留画面原比例的视频字幕长图。支持两种明确分开的输出：保留画面内已烧录字幕的原生字幕模式，以及把已审核的时间点与台词绘制到真实视频帧上的脚本字幕模式。用户要求原生字幕截图、字幕帧拼图、YouTube…

    2.6k GitHub stars~2.4k tokensUpdated today
    Media & CreativeAuto-check passed

  • Edu Chem Video

    wy51ai/edulab

    A skill your agent uses when asked to make an explainer / walkthrough video (讲解视频、解题视频、例题精讲、微课) for a chemistry problem (化学题: 氧化还原配平 双线桥 电子守恒, 物质的量计算, 化学平衡 三段式 平衡常数 转化率 反应速率, 离子反应, 电化学, 溶液 滴定…

    1.4k GitHub stars~2.1k tokensUpdated yesterday
    Media & CreativeAuto-check: notes

  • Transcription Memory Reconstruction

    NxcoreAI/EverRoom

    Reconstruct a complete, searchable memory from an untrusted meeting or conversation transcript.

    3k GitHub stars~714 tokensUpdated 2 days ago
    Media & CreativeAuto-check passed

  • Edu Math Video

    wy51ai/edulab

    A skill your agent uses when asked to make an explainer / walkthrough video (讲解视频、解题视频、例题精讲、微课) for a math problem (数学题, geometry, algebra, functions, motion/行程 problems), from a problem screenshot…

    1.4k GitHub stars~2.5k tokensUpdated yesterday
    Media & CreativeAuto-check: notes

  • Transcribe

    JetBrains/skills

    Official

    Transcribe audio files to text with optional diarization and known-speaker hints.

    366 GitHub starsUsed in 4 repos~776 tokens
    Media & CreativeAuto-check passed

More from vellum-ai/vellum-assistant

All 108 skills in this repo

  • Vellum GitHub App Setup

    vellum-ai/vellum-assistant

    Create and configure a GitHub App so the assistant can push commits, open PRs, and comment under its own bot identity.

    1.4k GitHub stars~3.1k tokensUpdated yesterday
    Auto-check passed

  • Discord App Setup

    vellum-ai/vellum-assistant

    Connect a Discord bot to the assistant via the Discord Gateway with guided application creation and intent configuration

    1.4k GitHub stars~4.2k tokensUpdated yesterday
    Auto-check passed

  • Sentry App Setup

    vellum-ai/vellum-assistant

    Create and configure a Sentry internal integration so the assistant can manage issues, alerts, and releases under its own identity

    1.4k GitHub stars~1.3k tokensUpdated yesterday
    Auto-check passed

  • Memory Corpus Ingest

    vellum-ai/vellum-assistant

    Ingest a large dataset into memory as a skimmed map. An agent skill from vellum-ai/vellum-assistant.

    1.4k GitHub stars~3k tokensUpdated yesterday
    Auto-check: notes

  • Plugin Builder

    vellum-ai/vellum-assistant

    A skill your agent uses when the user wants to build, scaffold, ship, or edit a Vellum plugin that bundles multiple surfaces (hooks, tools, skills, and more) into one installable package.

    1.4k GitHub stars~3.1k tokensUpdated yesterday
    Auto-check passed

  • Slack App Setup

    vellum-ai/vellum-assistant

    Connect a Slack app to the Vellum Assistant via Socket Mode.

    1.4k GitHub stars~2.5k tokensUpdated yesterday
    Auto-check: warnings

Categories

Questions about Watch Together

What does Watch Together do?

Watch local TV shows and movies with the user in real time. An agent skill from vellum-ai/vellum-assistant. Watch Together is an agent skill from vellum-ai/vellum-assistant. Watch local TV shows and movies with the user in real time.

When should I use Watch Together?

Watch Together fits situations like: tasks that involve Transcription.

How do I install Watch Together in Claude Code?

Run `npx skills add vellum-ai/vellum-assistant --skill watch-together -a claude-code`. Or copy the skill folder (skills/watch-together in vellum-ai/vellum-assistant) into .claude/skills/watch-together in your project. Claude Code loads it when a task matches its description.

How do I install Watch Together in Codex?

Run `npx skills add vellum-ai/vellum-assistant --skill watch-together -a codex`. Or copy the skill folder (skills/watch-together in vellum-ai/vellum-assistant) into .agents/skills/watch-together in your project. Codex loads it when a task matches its description.

Can I use Watch Together in Cursor, Gemini CLI or GitHub Copilot?

Cursor, Gemini CLI, GitHub Copilot and OpenCode also load SKILL.md folders. With the skills CLI, run `npx skills add vellum-ai/vellum-assistant --skill watch-together -a cursor` (or -a gemini-cli, github-copilot or opencode for the others). To copy it by hand, put the folder in .cursor/skills/watch-together, .gemini/skills/watch-together, .github/skills/watch-together and .opencode/skills/watch-together in your project.

What does Watch Together need to run?

Going by SKILL.md and its folder, Watch Together needs Python and a shell for the scripts in its folder, the command-line tools its instructions call (python3, ffmpeg, brew, ffprobe, apt and dnf) and credentials named GEMINI_API_KEY. Our summary lists: Python 3; A Bash shell; A credential in GEMINI_API_KEY.

Does Watch Together access the network?

SKILL.md contains no URLs. Any network use would come from the scripts or tools the agent runs. This is read from the text; nothing was executed.

Is Watch Together safe to install?

Our automated static check of SKILL.md found notes only (runs commands with sudo), nothing it rates as a warning. It is not a guarantee. The check reads SKILL.md only: the scripts in the folder are not scanned, so read them before running anything.

What licence does Watch Together use?

Watch Together is published under the MIT licence (the repository's licence). It allows redistribution, so the full SKILL.md is shown on this page.

How many tokens does Watch Together use?

About 2.1k tokens (SKILL.md is roughly 8.5k characters). Agents keep only the skill's name and description in context until a task matches; then they load SKILL.md in full.

What are the alternatives to Watch Together?

Skills that share tags, products or a category with Watch Together: HyperFrames Media Use (heygen-com/hyperframes, 60k stars), Native Subtitle Quote Image (chengyi-ai/native-subtitle-quote-image, 2.6k stars), Edu Chem Video (wy51ai/edulab, 1.4k stars) and Transcription Memory Reconstruction (NxcoreAI/EverRoom, 3k stars). The comparison table on this page puts their stars, adoption, token cost, safety result and licence side by side.

Who maintains Watch Together?

vellum-ai (a GitHub organization) maintains it in vellum-ai/vellum-assistant, which has 1,408 GitHub stars. The repository holds 108 skills in this directory. The repository was last updated on October 9, 2026.

Source: vellum-ai/vellum-assistant on GitHub. Facts on this page come from the repository at the commit we read; the author's words are quoted as theirs.