process_video

Process Video

Extract transcript + analyze content. One skill, one command.

Modes

Mode	Input	What Happens
YouTube	URL or video ID	Extract transcript via API (free, fast)
Local file	Path to .mp4/.mkv/.webm	Extract audio → transcribe via Whisper
Analyze	Add --analyze or just ask	Extract + structured analysis
Text only	Pasted transcript	Skip extraction, go straight to analysis

Execution

Step 1 — Detect Input Type

YouTube URL/ID → transcript_handler.py extract
Local video file → ffmpeg audio extract → process_audio → transcribe
Pasted text → skip to analysis
File path (.txt/.json) → read file → skip to analysis

Step 2 — Extract Transcript

YouTube:

cd hive && uv run python scripts/transcript_handler.py extract <VIDEO_ID> --languages es,en

Local video:

# Extract audio
ffmpeg -i /path/to/video.mp4 -vn -acodec mp3 -ab 128k /tmp/video_audio.mp3
# Transcribe via process_audio skill
hive/skills/process_audio/scripts/process_audio.sh /tmp/video_audio.mp3 --transcribe

Step 3 — Display Transcript

## Transcript Extracted

**Source:** [URL or file]
**Language:** [detected]
**Length:** [chars] characters, [segments] segments

### Preview
[First 500 chars...]

Step 4 — Analyze (if requested or default)

Apply 5 lenses:

1. Topics (max 10)

• Name: 2-5 words
• Depth: introductory | intermediate | advanced
• Summary: 1-2 sentences

2. Actionable Insights

• What: The insight
• How: 2-3 concrete steps
• Difficulty: easy | medium | hard

3. Connections

• Internal links between topics
• External links to broader concepts
• Recurring patterns

4. Key Quotes (max 7)

• Exact from transcript, never paraphrased
• With context

5. Recommended Deep Dives

• What to search, why, suggested sources

Output Template

# Video Analysis: [Title]

**Source:** [URL/path]
**Language:** [lang]
**Length:** [count]

## Topics
### 1. [Name] `[depth]`
[Summary]

## Actionable Insights
### [Title]
**What:** [desc]
**How:** 1. [step] 2. [step] 3. [step]

## Connections
- [A] ↔ [B]: [how]

## Key Quotes
> "[exact]"
— [context]

## Deep Dives
1. **[Topic]** — [why + search terms]

Multi-Video Mode

When multiple URLs/files:

1. Process each independently
2. Cross-analysis: common topics, contradictions, viewing order, combined insights

Quality Rules

• Max 10 topics, prioritize by importance
• Specific > generic ("use cover crops to fix nitrogen" > "soil improvement")
• Actionable > descriptive
• Exact quotes only
• Analyze in same language as transcript
• ALWAYS show preview of transcript before analysis

Prerequisites

• hive with deps (cd hive && uv sync)
• No API keys for YouTube transcripts
• YOUTUBE_API_KEY optional (metadata enrichment)
• OPENAI_API_KEY required for local video transcription (Whisper)
• ffmpeg required for local video files

Deprecates

• /extract_transcript → use /process_video <url>
• /video-analysis → use /process_video <url> --analyze or just /process_video <url> (analyze is default)