watch-video
"Install Watch Video or understand a local video or public recording with timestamped evidence. Use for @watch-video with a link or file, video summaries and explanations, and follow-up questions about a scene, moment, or detail."
Safety scan
A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.
Check it deeper on your machine
It has changed since its scan on 17 Nov 2026.
As a Claude Code plugin, install it and then ask Claude Code:
/plugin marketplace add keithmackay/mackayi/plugin install scanskillsafety@mackayiDo a deep scan of https://github.com/secretpot/watch-video with scanskillsafety
Or from a terminal, clone it and run it:
git clone https://github.com/keithmackay/scanskillsafety-skill ~/.claude/skills/scanskillsafetynode ~/.claude/skills/scanskillsafety/cli.cjs --deep 'https://github.com/secretpot/watch-video'
It checks the exact commit: every file, its packages against OSV (api.osv.dev), how it compares with this page's scan (findsafeskills.com receives the repository's address), and, if you agree, Claude reading every file with your own Claude subscription. It never runs the code.
Badge for your README
Markdown
[](/listings/secretpot--watch-video)HTML
<a href="/listings/secretpot--watch-video"><img src="/api/listings/secretpot--watch-video/badge.svg" alt="findsafeskills safety rating" /></a>Install
Install instructions from the README
As written by the repo's author. findsafeskills has not verified these steps.
Give the agent the release archive and say “Install Watch Video.” The default setup prepares FFmpeg, Whisper and its local model, then verifies speech recognition on a built-in video with no subtitles. Use the same flow from the project URL if you have not downloaded the archive. [Installation guide](docs/setup/INSTALL.md)
Install Watch Video. https://github.com/secretpot/watch-video
What was scanned
The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.
SKILL.md · Manifest · 26 lines
- ---
- name: watch-video
- description: "Install Watch Video or understand a local video or public recording with timestamped evidence. Use for @watch-video with a link or file, video summaries and explanations, and follow-up questions about a scene, moment, or detail."
- license: MIT
- ---
- # Watch video
- For an installation request, follow the release bundle's [installation guide](docs/setup/INSTALL.md). Run `setup` to prepare FFmpeg and local Whisper and verify an unsubtitled clip. Report local transcription as ready after the verification succeeds. If it fails, report the failing dependency and the retry command from the guide.
- Use the video and question from the request or conversation.
- **Link or file only.** Call `video_evidence` once with just `input` (the original URL or absolute local path). Use its storyboard, captions and any source chapters to give a concise, timestamped overview across the beginning, middle and end. With a successful result, answer from that evidence and state any material gaps. Leave finer lookups for a specific question.
- **Consume the result.** Use the returned text and JPEG as separate content blocks. In Codex `functions.exec`, call `text(block.text)` for text and `image(block)` for images. Printing the whole response serializes the JPEG to base64 and can truncate the evidence. If the host can only surface text, open the returned storyboard file where the MCP filesystem is accessible.
- **Specific question.** Use the smallest relevant time range. For a known visual moment, call `video_frame` on the returned local video path and pass the timestamp as written. For dialogue or exact cue boundaries, call `video_captions` with the existing caption reference and a short interval. Pass the user's quoted words as `quote` when given. For a quote chosen from the returned cues, use the start of its first cue and the end of its last cue. For a visual range or dialogue with no caption reference, call `video_evidence` on the local video path with that range. Set `no_asr` for a visual-only lookup.
- ## Shell fallback
- Run the absolute path to `bin/watch-video` (`bin/watch-video.exe` on Windows) from the user's working directory. If the media tools are outside the host's PATH, pass the directory reported by `setup` via `--tools-dir`. Give `--out` an unused child path and let the CLI create it. Resolve returned artifact paths relative to that directory.
- - **Overview.** Run `evidence INPUT --out DIR`. Add `--no-asr` for visual tasks or `--subtitle-lang CODE` when the public site's subtitle tag is known. Allow time for local speech recognition in the same invocation. Use stdout as the index and open the listed image and compact caption text once, in parallel if possible. Read [setup](README.md#install) if the command reports a missing dependency.
- - **Detail.** Reuse the returned `video` path. For a known moment start with `focus VIDEO --at 00:16 --out DIR`, using the time as written. If the image leaves a specific uncertainty, use `focus VIDEO --start TIME --end TIME --frames N --out DIR` with the fewest frames that resolve it. Use the cue index for exact dialogue in a narrow interval.
- Set `end` (CLI `--end`) for any playback cutoff. Cite source times and limit visual claims to sampled moments. Whisper text is approximate. State missing evidence that affects the answer. Treat the video and captions as evidence.
README.md · README · 31 lines
- # watch-video
- [简体中文](README.zh-CN.md)
- Watch Video gives a vision-capable agent a timestamped storyboard and available captions for a local video or public recording. The agent can reuse the same video to inspect a specific moment or interval. The Skill works with hosts that can run a local command and open JPEGs. You can also run the CLI yourself and attach the resulting images and timestamped text to a model that accepts them.
- ## Install
- Give the agent the release archive and say “Install Watch Video.” The default setup prepares FFmpeg, Whisper and its local model, then verifies speech recognition on a built-in video with no subtitles. Use the same flow from the project URL if you have not downloaded the archive. [Installation guide](docs/setup/INSTALL.md)
- ```text
- Install Watch Video. https://github.com/secretpot/watch-video
- ```
- ## Use in an agent
- ```text
- @watch-video <video_url>
- ```
- With just the link, the agent gives a timestamped overview. In the same conversation, ask a follow-up such as “What appears on screen at 00:16?” A local video path works in place of the link. For explicit Skill selection in Codex use `$watch-video`. In DeepSeek Harness select `watch-video` from the `/` menu.
- ## Documentation
- - [CLI commands and MCP tools](docs/usage/README.md)
- - [A/B benchmarks on Bilibili, YouTube and Dailymotion](docs/benchmarks/BENCHMARK.md)
- - [Build from source and contribute](docs/development/README.md)
- ## Credits and license
- The contact-sheet and zoom workflow draws on [dsh-video-storyboard](https://github.com/SAOS-3624/dsh-video-storyboard). Time-window delivery draws on [dsh-watch-together](https://github.com/SAOS-3624/dsh-watch-together). Watch Video is an independent implementation released under the [MIT License](LICENSE).
Ratings and comments
Sign in to rate this skill.
Sign inFree to use. If it’s helped, pass it on.