agentleFS
sharon-laiccsharon-laicc/viral-video-decomposer/skill/SKILL.md

video-structure-analyzer

Skill4 starsChanged 5 months agoMITView on GitHub

拆解爆款短视频、Vlog、小红书/抖音/B站视频、广告素材或竞品视频时使用。用于把视频链接、截图、录屏、字幕、逐字稿或参考视频拆成镜头级拉片表、爆款机制、可复用结构、批量选题变量、脚本/分镜/prompt,以及给后续视频生成 agent 使用的 JSON brief。也适用于用户说“分析爆款视频”“拆视频”“拉片”“拆解短视频”“复刻这种视频结构但不要抄”“根据爆款批量产出视频方案”等场景。

1 other project uses a version of this, 1 of them edited.

Adopted by

Most recently taken up by

  • Other versions1 project uses a version of this.

    Word for word · 0

    None found.

    Wrote their own version · 0

    None found.

    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 repositoryOne other file it gives its agents.

The file

Show its rules and text

We split this skill into the rules it gives an agent

23 rules. About 0 are common, 12 are distinctive, and 11 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈12

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 11

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Use this skill as the pre-production researcher for video-generation agents.

    line 10
  2. Very specific:

    When the user asks for an HTML report, the report is not a free-design task.

    line 18
  3. Very specific:

    Keep the interface name Viral Video Lab.

    line 22
  4. Very specific:

    Keep the same A/B switch behavior, labels, navigation, module order, spacing scale, card treatment, table treatment, shot-card grid, mobile …

    line 23
  5. Distinctive:

    Do not invent a new hero, color palette, card style, Chinese title bar, marketing landing page, decorative gradients, special blueprint …

    line 24
  6. Distinctive:

    If a platform blocks direct download, the skill should:

    line 55
  7. Distinctive:

    Use screenshots or metadata as alternatives

    line 58
  8. Very specific:

    Automatic video download: Use yt-dlp, ffmpeg, or browser automation to automatically download the video from the provided URL.

    line 67
  9. Very specific:

    If blocked, try ffmpeg with URL directly

    line 71
  10. Distinctive:

    If still blocked, use browser automation to capture video

    line 72
  11. Distinctive:

    Always record the extraction method used in the source facts.

    line 77
  12. Very specific:

    Create one S### row/card per visible cut or meaningful continuous action beat.

    line 93
  13. Distinctive:

    Every shot card in the HTML should have a matching real image file.

    line 97
  14. Distinctive:

    If exact frames are unavailable, do not invent precise timecodes.

    line 98
  15. Very specific:

    Do not include a visible JSON Brief(机器可读) module inside the HTML report unless the user explicitly asks to display raw JSON in the web page.

    line 133
  16. Distinctive:

    Use this table unless the user asks for another format:

    line 141
  17. Distinctive:

    Treat empty shots as meaningful: B-roll, pauses, object close-ups, and transition shots often carry rhythm and emotion.

    line 190
  18. Distinctive:

    If evidence is incomplete, mark confidence and avoid pretending exact frame timing is known.

    line 192
  19. Distinctive:

    Automatic video handling: Always attempt to download video from URL automatically.

    line 193
  20. Very specific:

    Before delivering an HTML report, verify that the HTML contains: .topbar-inner, .brand with Viral Video Lab, .style-switch, …

    line 194
  21. Very specific:

    Before delivering an HTML report, verify that every local <img src="..."> path resolves relative to the HTML file.

    line 195
  22. Very specific:

    Before delivering an HTML report, verify that it does not contain placeholder-only shot panels, a visible JSON Brief(机器可读) section, or …

    line 196
  23. Very specific:

    If possible, run references/validate_report.py <report.html> before final delivery.

    line 197

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Video Structure Analyzer

Use this skill as the pre-production researcher for video-generation agents. The goal is not only to explain why a video works, but to produce reusable, original, machine-readable production specs.

Core Rule

Analyze methods, structure, rhythm, and production logic. Do not copy protected expression. Avoid reproducing exact scripts, distinctive shot sequences, watermarks, creator identity, brand identity, or proprietary assets unless the user owns them or explicitly asks for analysis-only commentary.

Non-Negotiable Report Fidelity

When the user asks for an HTML report, the report is not a free-design task. You MUST reproduce the accepted Viral Video Lab report system exactly enough that a user can recognize it as the same product:

  • Load references/report-style.md before writing HTML.
  • Inline or copy the exact CSS from references/viral-video-lab.css; do not replace it with a new visual system.
  • Keep the interface name Viral Video Lab.
  • Keep the same A/B switch behavior, labels, navigation, module order, spacing scale, card treatment, table treatment, shot-card grid, mobile topbar behavior, and AI 生产蓝图 module styling.
  • Do not invent a new hero, color palette, card style, Chinese title bar, marketing landing page, decorative gradients, special blueprint panel, or unrelated template.
  • Content may change per video; layout and visual system should remain stable.

Input Modes

Primary mode: Video URL only - Users should only need to provide a video link. The skill automatically handles video download and processing.

Accepted video sources:

  • Xiaohongshu (小红书): Direct link or video ID
  • Douyin (抖音): Direct link or video ID
  • Bilibili: Direct link or video ID
  • YouTube: Direct link or video ID
  • TikTok: Direct link or video ID
  • Kuaishou (快手): Direct link or video ID
  • Instagram: Direct link or video ID
  • Ad libraries: Direct link

Alternative inputs (when video link is unavailable):

  • Local video file upload (for testing or when link is blocked)
  • Screenshots or screen recordings
  • Transcript, SRT/VTT subtitles, or OCR text
  • Existing analysis notes

Important: Users should NOT need to manually download or upload videos. The skill should automatically:

  1. Accept the video URL
  2. Download the video using appropriate tools (yt-dlp, ffmpeg, browser automation, etc.)
  3. Extract frames and metadata
  4. Generate the analysis report

If a platform blocks direct download, the skill should:

  • Attempt multiple download methods
  • Fall back to browser automation if needed
  • Use screenshots or metadata as alternatives
  • Clearly state any limitations in the output

Workflow

  1. Capture source facts: Title, creator/account, platform, URL, duration, visible engagement metrics, publish date if available, product/category, and any extraction limitations.
  2. Extract usable evidence: Automatic video download: Use yt-dlp, ffmpeg, or browser automation to automatically download the video from the provided URL. Do not ask the user to upload the video file.

    Extraction priority:

    1. Attempt automatic download using yt-dlp (supports most platforms)
    2. If blocked, try ffmpeg with URL directly
    3. If still blocked, use browser automation to capture video
    4. Extract frames using ffmpeg or ffprobe
    5. Extract subtitles/transcripts if available
    6. For blocked web video, sample screenshots at meaningful timestamps

    Always record the extraction method used in the source facts.

    For full HTML reports, evidence images are required. Create a sibling visual directory for every report, named with the same slug plus -visual, for example:

    • xhs-smelly-cat_图文拆解报告.html
    • xhs-smelly-cat-visual/contact_sheet.jpg
    • xhs-smelly-cat-visual/S001.png
    • xhs-smelly-cat-visual/S002.png

    The HTML must reference those local relative files. A final *_图文拆解报告.html must use real frame images, not decorative placeholders. If frames cannot be extracted, do not pretend the report is complete: create a clearly labeled draft analysis or ask the user for the video file/screenshots, and state that a final HTML report requires visual evidence. Do not silently omit the screenshot/contact-sheet modules.

  3. Build the shot-level table: The table is the main output. Include timecode, duration, visual, shot size, action, subtitle/dialogue summary, sound, camera angle, camera movement, transition, emotional function, narrative function, and generation prompt.

    For full reports, Shot-by-Shot means actual shots/cuts, not broad narrative chunks. Do not compress a long video into 6-8 generic sections unless the user explicitly asks for a compressed outline. When real frames are available:

    • Create one S### row/card per visible cut or meaningful continuous action beat.
    • For slow continuous footage, sample at least every 3-5 seconds.
    • For 20-60s videos, expect roughly 10-25 shots unless the video is truly static.
    • For 60-120s videos, expect roughly 18-45 shots unless the video is truly static.
    • Every shot card in the HTML should have a matching real image file.
    • If exact frames are unavailable, do not invent precise timecodes; label the section as a reconstructed outline instead of final shot-by-shot.
  4. Segment the narrative: Group shots into sections such as Hook, setup, proof, reveal, contrast, routine block, escalation, payoff, summary, CTA. Name the section by its job, not by vague labels.
  5. Explain the viral mechanism: Identify hook type, curiosity gap, emotional driver, information density, rhythm, visual density, trust signals, audience fantasy/pain, repeatable meme or comment trigger, and conversion move.
  6. Convert analysis into production assets: Produce reusable templates, variables, batch variants, script options, shot prompts, caption rules, BGM/SFX guidance, platform adaptations, and a video_generation_brief.json-style object.
  7. Add originality guardrails: List what can be reused at the pattern level and what must be changed to avoid copying.

Required Outputs

For quick requests, answer inline with these sections:

  • Source facts
  • One-paragraph thesis
  • Shot-level table or compressed shot table
  • Viral mechanism
  • Reusable production template
  • Batch variants or next-video ideas
  • JSON brief

For substantial requests, create files in the working project:

  • *_拆解报告.html or *_拆解报告.md
  • *_拉片拆解.md
  • *_video_generation_brief.json
  • Required for HTML reports: a *-visual/ directory containing contact_sheet.jpg plus sampled frame images named S001.png, S002.png, etc. Use visual/contact_sheet.jpg paths only when the HTML itself lives inside the same parent directory and the folder is actually named visual.

When producing HTML reports, read references/report-style.md and follow that style contract as a hard requirement. Prefer a dense professional analysis surface, not a marketing landing page. The accepted default is a Viral Video Lab report with source facts, stats, contact sheet, shot cards, production table, AI blueprint, variable slots, prompt template, and an A/B visual style toggle.

Do not include a visible JSON Brief(机器可读) module inside the HTML report unless the user explicitly asks to display raw JSON in the web page. The JSON brief belongs in the separate *_video_generation_brief.json file. The HTML should include Agent Prompt Template, not raw machine JSON.

The AI 生产蓝图 module should use the same module/card styling as adjacent sections in both A and B themes unless the user explicitly asks for a special treatment.

For key-shot overviews, generate a compact contact sheet with tight gutters. Do not use large screenshot row gaps that make the report excessively long.

Shot Table Schema

Use this table unless the user asks for another format:

排序镜号时间码时长秒画面景别内容/动作台词/字幕声音摄像机角度镜头运动转场情绪功能叙事功能生成提示词

For fast analysis, compress to:

镜号时间码画面/动作台词/字幕摘要情绪功能叙事功能可复用生成提示

JSON Brief

For the complete JSON contract and detailed heuristics, read references/output-contract.md when generating machine-readable assets.

Minimum JSON object:

{
  "source": {},
  "analysis_summary": "",
  "viral_pattern": {},
  "shots": [],
  "production_template": {},
  "batch_variants": [],
  "originality_guardrails": []
}

User Experience

Minimal user input required: Users should only provide:

  1. Video URL (or video ID)
  2. (Optional) Specific analysis focus or output format

Automatic handling:

  • Video download
  • Frame extraction
  • Subtitle/transcript retrieval
  • Metadata collection
  • Report generation

No file upload needed unless:

  • The video URL is blocked and user provides local file as fallback
  • User wants to analyze multiple videos at once

Quality Bar

  • Be concrete: prefer "0-3s close-up of messy desk, abrupt zoom, anxiety hook" over "strong opening".
  • Preserve timing: timecodes and durations matter because later agents need executable pacing.
  • Treat empty shots as meaningful: B-roll, pauses, object close-ups, and transition shots often carry rhythm and emotion.
  • Separate human-readable insight from machine-readable production specs.
  • If evidence is incomplete, mark confidence and avoid pretending exact frame timing is known.
  • Automatic video handling: Always attempt to download video from URL automatically. Only ask for file upload if automatic download fails after all retry methods.
  • Before delivering an HTML report, verify that the HTML contains: .topbar-inner, .brand with Viral Video Lab, .style-switch, body[data-style="swiss"], #overview, #shots, #blueprint, #prompt, a contact-sheet <img>, and at least one shot card image when frames are available.
  • Before delivering an HTML report, verify that every local <img src="..."> path resolves relative to the HTML file. Broken image references are a failed output.
  • Before delivering an HTML report, verify that it does not contain placeholder-only shot panels, a visible JSON Brief(机器可读) section, or compressed narrative beats mislabeled as final shot-by-shot analysis.
  • If possible, run references/validate_report.py <report.html> before final delivery. If it fails, fix the report instead of presenting it as complete.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.