video-analyzer
Skill1 starsChanged 3 months agoMITView on GitHub
Analyze video files by extracting frames with FFmpeg and visually Reading them to learn what's actually inside — on-screen text, products, people, quotes, branding — before any content decision. Triggers on "look at these videos", "what's in these clips", "review the footage", "analyze the reels", or whenever client-provided videos need captions, calendars, or audits. If video files exist and decisions depend on them, run this first — never guess from filenames.
- Installs packages
6 other projects use a version of this.
Adopted by
Most recently taken up by
- TsinHzl/claude-code-configtheir own version, first seen Nov 1, 2026
- zrong/skillstheir own version, first seen Oct 29, 2026
- Valeriy1980/claude-skillstheir own version, first seen Oct 13, 2026
- pgomezurrea-eng/claude-skillstheir own version, first seen Sep 12, 2026
- VjAlbert/video-analyzer-repotheir own version, first seen Jun 27, 2026
- Agentchengfeng/video-analyzer-skilltheir own version, first seen Feb 14, 2026
Other versions6 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 6
- Agentchengfeng/video-analyzer-skill★ 16same name
- zrong/skills★ 13same name
- TsinHzl/claude-code-config★ 2same name
- pgomezurrea-eng/claude-skills★ 0same name
- Valeriy1980/claude-skills★ 0same name
- VjAlbert/video-analyzer-repo★ 0same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository47 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- access-transfer-trackerSkill
- ai-3d-generatorSkill
- ai-asset-generatorSkill
- ai-demo-builderSkill
- apple-grade-designSkill
- brand-guide-builderSkill
- client-onboardingSkill
- cold-email-campaignSkill
- cold-emailSkill
- company-teardownSkill
- competitive-ads-extractorSkill
- content-calendarSkill
- content-repurposerSkill
- content-research-writerSkill
- content-strategySkill
- daily-briefSkill
- discovery-call-prepSkill
- email-composerSkill
- follow-up-nurtureSkill
- gbp-local-seoSkill
- hormozi-100m-leadsSkill
- interactive-website-builderSkill
- invoice-trackerSkill
- lead-flow-verifierSkill
- lead-scraperSkill
- mcp-integrationSkill
- meeting-listenerSkill
- post-purchase-email-seriesSkill
- productization-journalSkill
- proposal-generatorSkill
- prospect-workupSkill
- regulated-copy-complianceSkill
- retainer-reportingSkill
- review-managerSkill
- scroll-demo-builderSkill
- seo-auditSkill
- site-launch-cutoverSkill
- site-qa-checklistSkill
- social-media-auditorSkill
- social-media-managerSkill
- social-video-producerSkill
- stylized-3d-worldbuildingSkill
- web-researchSkill
- website-builderSkill
- weekly-client-checkinSkill
- welcome-email-seriesSkill
The file
We split this skill into the rules it gives an agent
4 rules. 0 are common, 2 are distinctive, and 2 are very specific to this project.
0
Common
Most of the 3 edited versions we read keep these.
2
Distinctive
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 14
Before any of these tasks, whenever video files are involved:
- Very specific: line 58
Read each PNG. Start with the _mid frames — they usually capture the main content.
- Very specific: line 81
Check brand consistency against clients/<slug>/brand/.
- Distinctive: line 89
Countdown/launch videos: verify the sequence is complete.
“Common” is measured: kept by at least 6 in 10 of the 3 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Video Analyzer
Claude cannot play video, but it CAN view images. This skill bridges the gap: extract key frames with FFmpeg, then Read them. The result is planning accuracy equal to having watched the footage.
Why it matters: planning content around unseen videos produces wrong captions, wrong scheduling, and missed details. A filename like Review (FB).mp4 doesn't tell you which product is reviewed, what the reviewer said, or what the visual style is. Frames do.
When to Run This First
Before any of these tasks, whenever video files are involved:
- Content calendar creation — know what each video contains before scheduling or captioning it
- Social media planning — match videos to pillars, platforms, cadence
- Client video audit — inventory existing content before building strategy
- Caption writing — see the screen so captions complement rather than repeat
- Quality review — brand consistency, text readability, style across a set
Workflow
Step 1 — Probe metadata, then extract frames
# Metadata for every video in a folder (duration, resolution, audio streams)
for f in videos/*.mp4; do
ffprobe -v quiet -print_format json -show_format -show_streams "$f" \
> "analysis/$(basename "$f" .mp4).json"
done
# Extract start / middle / end frames (repeat per file; MID = duration/2, END = duration - 0.5)
ffmpeg -ss 0.5 -i video.mp4 -frames:v 1 thumbs/video_start.png
ffmpeg -ss "$MID" -i video.mp4 -frames:v 1 thumbs/video_mid.png
ffmpeg -ss "$END" -i video.mp4 -frames:v 1 thumbs/video_end.pngSave output alongside the videos, e.g. clients/<slug>/deliverables/content-calendar/video-analysis/. Write a summary.md inventory table (filename, duration, resolution, aspect, audio y/n) as you go.
Frame count by duration:
| Duration | Frames | Why |
| Under 5s | 1 (mid) | One frame captures it |
| 5-15s | 3 (start/mid/end) | Default — catches transitions and the final CTA |
| 15-60s | 5 | Multiple scenes likely |
| Over 60s | 5-8 | Full scene coverage |
Step 2 — Read the metadata for signals
- No audio stream = animated graphic or text overlay; captions carry 100% of the story.
- Under 10 seconds = likely an animated social graphic, not filmed content.
- Portrait (1080x1350, 1080x1920) = built for Instagram/TikTok; landscape = Facebook/YouTube.
- Same content at two resolutions = client pre-made platform pairs (filenames often end "(FB)"/"(IG)").
Step 3 — View the frames
Read each PNG. Start with the _mid frames — they usually capture the main content; check _start/_end only when mid is ambiguous or the video runs longer than 15s.
Document per video:
- On-screen text — headlines, quotes, CTAs, review text. Transcribe exactly.
- Visuals — people, products, packaging, scenes, illustrations.
- Visual style — colors, fonts, template, brand consistency.
- Which product/service is featured — read the actual labels and covers in the frame, not the filename. Filenames lie.
- Content category — countdown, announcement, review, promotional, educational, behind-the-scenes.
- Platform fit — from aspect ratio plus content type.
Step 4 — Write the content brief
Save video-content-briefs.md alongside the analysis. Include:
- Discoveries — things filenames could not have told you ("all the review videos feature the flagship product, not the new releases")
- Per-category breakdowns with the actual visible text/quotes
- Scheduling recommendations — lead order, sequences, what to hold back
- Caption guidance — what the caption should ADD versus what's already on screen; never just repeat the visible quote
- Summary table — counts by category, duration, audio, format
This brief is the handoff artifact: content-calendar, social-media-manager, and content-strategy consume it by path instead of re-analyzing the videos. Gaps it reveals feed social-video-producer.
Step 5 — Verify critical details
- Cross-reference product shots against client records; when frame and filename disagree, trust the frame.
- Partially visible or cut-off text: note it and flag for the operator to verify.
- Check brand consistency against
clients/<slug>/brand/.
Patterns to Watch For
Client-provided social graphics: often silent ~5s animations in FB/IG resolution pairs sharing a template; the product shown may differ from the filename. Review/testimonial videos: identify the exact product reviewed from the frame; transcribe the full quote (gold for captions and site testimonials); note reviewer name and credentials for authority positioning; schedule multi-part reviews together. Countdown/launch videos: verify the sequence is complete; map to calendar dates counting backward from launch day. Quote/verse graphics: transcribe the exact text and attribution; note deliberate pairings of subject and quote; match the caption's tone to the content's register, not a sales register.
Troubleshooting
- FFmpeg missing:
brew install ffmpeg(macOS) or your package manager. - Black frames: slow fade-in — extract more frames at more timestamps.
- 50+ files: view only
_midframes first, bucket by filename pattern, then deep-dive one representative per bucket.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

