Claude Video Watch Skill — Setup Guide¶
Source: bradautomates/claude-video
GitHub: bradautomates/claude-video (15.4K stars)
Skills: 1 skill (watch) · 10.1K installs
Category: Video Understanding
First Seen: Apr 29, 2026 (catalogued August 15, 2026 midday sweep)
Quality Tier: 🟡 Trusted (Gen Agent Trust Hub and Socket pass; Snyk Fail on the skill page — named in Limitations)
The watch skill gives an agent a video input. A bundled Python script fetches captions first, optionally downloads the video, extracts frames as JPEGs (scene-aware, or fast keyframes at efficient detail), builds a timestamped transcript (native captions first, Whisper API as fallback), and prints frame paths. The agent then reads frames and combines them with the transcript to answer questions about the video.
Installation¶
npx skills add bradautomates/claude-video --skill watch
Prerequisites¶
| Requirement | Details |
|---|---|
| Python 3 | Runs the bundled watch.py script |
| Node.js + npx | For the skill installer |
| ffmpeg | For frame extraction (scene-aware mode) |
| Whisper API key (optional) | Fallback when native captions are absent |
What It Provides¶
| Capability | Notes |
|---|---|
| Caption-first pipeline | Native captions used before any transcription |
| Frame extraction | Scene-aware or fast keyframe JPEG extraction at efficient detail |
| Timestamped transcript | Captions or Whisper API fallback |
| Harness-agnostic pathing | SKILL_DIR resolution works across Claude Code, Codex, Cursor, Gemini CLI |
Quick Start¶
- Install:
npx skills add bradautomates/claude-video --skill watch - Resolve SKILL_DIR per the skill's layout table, then ask: "watch this video and summarize what happens at each timestamp"
CorpusIQ Use Cases¶
| Use Case | How |
|---|---|
| Video intelligence | Complements our video-transcript-analysis pipeline with frame-level grounding |
| Creator research | Extract timestamped insights from YouTube and TikTok content |
| Multimodal answers | Answer visual questions about recordings without a native video input |
| Harness pattern | The SKILL_DIR resolution pattern is reusable for multi-harness skills |
Limitations / Verification¶
- Snyk audit Fail on the skill page — review the flagged dependency surface before production use
- Frame extraction requires ffmpeg; Whisper fallback requires an API key
- Publisher page verified: 10.1K installs, 15.4K GitHub stars
npx skills add bradautomates/claude-video # verify install works
Related¶
Powered by CorpusIQ