Video Processor v1 — AI Video Processing Pipeline

Purpose

Pure video processing skill with no platform opinions. Scans 00_Inbox/ for new raw videos,
analyzes frames with multimodal AI, generates English copywriting, applies orientation-aware
smart cropping via FFmpeg, produces clip variants, and outputs to 02_PendingReview/.

Platform rules (FB supports all orientations, TikTok requires portrait) are handled by the
expert calling this skill — the skill itself just processes whatever it's given.

When to Use

• Processing new raw videos dropped into 00_Inbox/
• Generating short-form video clips for Facebook, TikTok, or Instagram
• Auto-creating captions and copywriting for video posts
• The user says: "process videos", "generate clips", "prepare video content", "run video pipeline"

Project Dependency

This skill depends on the shared project core modules and environment at:
- Project root: <your_workspace>
- Core modules: core/model_providers.py, core/logger.py, core/state_machine.py, core/sop_pipeline.py
- Environment: .env file with API keys (OpenRouter, DashScope, etc.)
- Dependencies: FFmpeg (system), Python packages (opencv-python, requests, python-dotenv)

All commands MUST be run from the project root directory.

How to Use

Basic — process all new videos in inbox

cd <your_workspace>
python scripts/run.py

Filter by orientation (e.g., exclude landscape for portrait-only platforms)

cd <your_workspace>
python scripts/run.py --exclude_orientation landscape

Process a specific video by name

cd <your_workspace>
python scripts/run.py --name "my_video_filename"

Force re-process already done videos

cd <your_workspace>
python scripts/run.py --force

Pipeline Steps

1. Scan: Detect new videos in 00_Inbox/ (excluding _done/)
2. Analyze: video.analyzer.v1 — extract keyframes, run multimodal LLM analysis
3. Copywriting: copywriting.llm.v1 — generate 3 caption variants (A/B/C)
4. Crop: FFmpeg smart cropping based on orientation
   - Portrait (h > w) → 9:16 center crop
   - Landscape (w > h) → 16:9 center crop + extra vp (9:16 padded portrait)
   - Square (0.9 ≤ ratio ≤ 1.1) → 1:1 center crop + extra vp (9:16 padded portrait)
5. Produce: 3 base clip variants (v1/v2/v3) + optional portrait variant (vp)
6. Output: Write clips + captions to 02_PendingReview/
7. Archive: Move source video to 00_Inbox/_done/

Output Structure

02_PendingReview/
└── {video_name}/
    ├── clip_v1.mp4          # Detail version
    ├── clip_v2.mp4          # Standard version
    ├── clip_v3.mp4          # Depth version
    ├── clip_vp.mp4          # Portrait 9:16 variant (landscape/square only, black-bar padded)
    ├── caption.json         # LLM-generated English copy
    └── analysis.json        # Frame analysis results

Orientation Rules