vidpipe

name	vidpipe
description	VidPipe CLI video processing — run the VidPipe pipeline on a recorded video, monitor progress, launch the review server, and handle the full flow from recording to review-ready. Use when user says "process video", "vidpipe", "run vidpipe", "video processing", "review server", "pipeline log", "process recording", or when the video-bridge extension sends a processing request via session.send().

VidPipe Processing Skill

How to process a video recording through VidPipe and prepare it for review. This is the canonical reference for any agent that receives a video processing request from the video-bridge extension.

⚠️ CRITICAL: Long-Running Process

VidPipe processing takes 5–30+ minutes depending on video length. It performs silence removal, transcription, captioning, clip generation, thumbnail creation, and social post drafting — all in one pipeline run.

NEVER spawn VidPipe from an extension or short-lived context. Always run it from a session/agent where long execution is expected.

Default Paths (from %APPDATA%\vidpipe\config.json)

Path	Purpose
C:\VidPipe\output	Default output directory — all processed videos land here
C:\VidPipe\brand.json	Brand config (logos, fonts, colors)
C:\VidPipe\schedule.json	Platform-specific posting schedules

DO NOT override --output-dir — let VidPipe use its default C:\VidPipe\output. This keeps all outputs in one predictable location and avoids fragmented output across run directories.

Step 1: Process the Video

Run VidPipe via npx (version pinned at 1.3.26):

npx vidpipe@1.3.26 process "<video-file-path>" --once --progress

Valid Flags

Flag	Purpose
--once	Process a single file and exit (don't watch for new files)
--progress	Show progress updates during processing
--verbose	Extra logging output
--no-silence-removal	Skip silence removal step
--no-shorts	Skip short clip generation
--no-medium-clips	Skip medium clip generation

Invalid flags: --no-git does NOT exist — never use it.

Monitoring Progress

VidPipe writes real-time progress to pipeline.log inside the output subfolder. To monitor:

1. Wait ~10 seconds after starting for the output subfolder to appear
2. Check C:\VidPipe\output\ for the newest subfolder (named after the video)
3. Tail pipeline.log inside that subfolder for live progress
4. Look for "Pipeline complete" to know processing finished

Output Structure

After processing, VidPipe creates a subfolder in C:\VidPipe\output\ containing:

C:\VidPipe\output\<video-name>\
├── pipeline.log          # Full processing log
├── transcript.json       # Word-level transcript
├── summary.json          # AI-generated summary
├── <video>-processed.mp4 # Primary processed video
├── shorts/               # Short clips (<60s)
├── medium-clips/         # Medium clips (1-5min)
├── social-posts/         # Generated social media copy
└── thumbnails/           # AI-generated thumbnails

Error Handling

If VidPipe fails:

1. Check pipeline.log for the specific error
2. Common issues: FFmpeg not found, Gemini API timeout, disk space
3. Notify {{PARENT_1}} via Telegram with the error details
4. Do NOT retry automatically — wait for {{PARENT_1}}'s instructions

Step 2: Launch the Review Server

After processing completes successfully, start the review server so {{PARENT_1}} can review on his phone:

npx vidpipe@1.3.26 review --port 3847

Set the OUTPUT_DIR environment variable to the VidPipe output directory:

OUTPUT_DIR=C:\VidPipe\output

The review server serves a web UI at http://localhost:3847 that shows:

• All processed videos with thumbnails
• Short/medium clip previews
• Approve/reject buttons for each item

Making the Review Server Accessible

The video-bridge extension registers a "review" service with the ngrok gateway. Once the review server is running on port 3847, the gateway routes traffic to it automatically. No separate tunnel is needed.

Step 3: Notify {{PARENT_1}}

After processing completes and the review server is ready, send {{PARENT_1}} a Telegram message:

🎬 *VidPipe review ready*
• Shorts: {count}
• Mediums: {count}
• Longs: {count}
• Review: {review_url}

Use the speak parameter: "Your video was processed. I found X shorts, Y medium clips, and Z long videos. Review link is ready."

Full Flow (triggered by video-bridge extension)

When the video-bridge extension receives a recording upload, it sends a session.send() message to the agent. Here's the expected flow:

1. Receive processing request — message includes file path, filename, size, duration, and run ID
2. Run VidPipe — npx vidpipe@1.3.26 process "<file>" --once --progress
3. Monitor progress — tail pipeline.log in the output subfolder
4. Count outputs — scan the output subfolder for shorts/, medium-clips/, and primary video
5. Start review server — npx vidpipe@1.3.26 review --port 3847 with OUTPUT_DIR=C:\VidPipe\output
6. Notify {{PARENT_1}} — Telegram with counts and review URL
7. Wait for approval — {{PARENT_1}} reviews on phone, approves/rejects via the review UI
8. On approval — the video-bridge extension handles the publish flow via a separate session.send()

Windows-Specific Notes

• VidPipe is a Node.js CLI tool — npx.cmd requires shell: true on Windows
• NVM-for-Windows may leave node missing from PATH — if spawning fails, check %LOCALAPPDATA%\nvm\ for installed versions and prepend to PATH
• FFmpeg must be in PATH (VidPipe depends on it for all video operations)