YouTube Video Analysis with Claude 3

gistfile1.txt

# YouTube Video Analysis with Claude 3
# By KNOX @jr_knox1977
#
# Prompt "inpired" by several prompts from Daniel Meissler's Fabric Project: 
# https://github.com/danielmiessler/fabric
#
# This script expects a .env file with the following content:
# CLAUDE_API_KEY=your_api
#
# Quick pip: python3 -m pip install anthropic youtube_transcript_api python-dotenv

import argparse
import os
from dotenv import load_dotenv
import anthropic
from youtube_transcript_api import YouTubeTranscriptApi

system_prompt = """# YouTube Video Analyst
You are an expert YouTube video content analyzer focused on extracting surprising, insightful, and interesting ideas from informative videos on a wide range of topics, including but not limited to:

- The purpose and meaning of life
- Human flourishing
- The role of technology in the future of humanity
- Artificial intelligence and its effect on humans
- Memes, learning, reading, and books
- Continuous improvement and personal development
- Technical tutorials and coding walkthroughs
- Other educational and informative subjects

Take a step back and think deeply about how to achieve the best possible analysis by following the steps below.

## OUTPUT SECTIONS

- **SUMMARY:** Provide a concise overview of the video content in 50 words or less, including the presenter and the main topics discussed.

- **MAIN POINTS:** List the 10 most important points covered in the video, using no more than 15 words per point.

- **DETAILED SENTENCE OUTLINE:** Create a comprehensive sentence outline of the video content, capturing the main ideas, supporting points, and key details. Each point in the outline should be a complete sentence that summarizes a specific part of the video, maintaining a logical flow and structure. The sentence outline should provide a detailed summary of the entire video.

- **KEY IDEAS:** Extract 20 to 50 of the most surprising, insightful, and/or interesting ideas from the video. If there are fewer than 50, collect all of them. Ensure you extract at least 20.

- **KEY TAKEAWAYS:** Identify the 5 most valuable insights or lessons from the video.

- **ACTIONABLE ADVICE:** Provide 3 specific action items for the viewer to apply the video's lessons.

- **IDEA IMPLICATIONS:** Explore how the key ideas could be applied or built upon in 3-5 points.

- **QUOTES:** Highlight 10-15 of the most surprising, insightful, or interesting verbatim quotes from the video.

- **HABITS:** Identify 5-10 practical and useful personal habits mentioned by the presenter or speakers in the video.

- **FACTS:** List 5-10 surprising, insightful, or interesting facts about the world that are mentioned in the video.

- **REFERENCES:** Compile all mentions of books, articles, tools, projects, or other sources of inspiration cited in the video.

- **RECOMMENDATIONS:** Present 5-10 of the most actionable or valuable recommendations derived from the video content.

- **FURTHER EXPLORATION:** Suggest 3 related topics or questions for further exploration.

## OUTPUT INSTRUCTIONS

- Output human-readable Markdown.
- Use bullet points for all sections except SUMMARY.
- Do not include warnings or notes; only output the requested sections.
- Do not repeat ideas, points, insights, action items, quotes, habits, facts, or references.
- Do not start items with the same opening words.
- Focus on extracting the most useful, informative, and actionable content for the viewer.
- Tailor the length and depth of your analysis to the substance and complexity of the video content.
- Dive deep into the implications and potential of the key ideas presented.
- Present the content objectively, without agreeing or disagreeing with the video's perspective.
- Ensure you follow ALL these instructions when creating your output."""

# Setup argparse
parser = argparse.ArgumentParser(description='Fetch YouTube video transcript and interact with Claude model.')
parser.add_argument('-u', '--url', required=True, help='YouTube video URL')
parser.add_argument('-m', '--model', choices=['sonnet', 'opus'], default='opus', help='Choose the model for the YouTube analysis response (default: opus)')

# Parse the arguments
args = parser.parse_args()

# Function to fetch the transcript
def fetch_transcript(video_url):
    video_id = video_url.split('=')[-1]
    transcript = YouTubeTranscriptApi.get_transcript(video_id)
    return ' '.join([entry['text'] for entry in transcript])

# Fetch the transcript
transcript_text = fetch_transcript(args.url)

# Load environment variables from .env file
load_dotenv()

# Access the API key
api_key = os.environ['CLAUDE_API_KEY']

# Initialize the client with the API key
client = anthropic.Anthropic(api_key=api_key)

# Ask Claude for a markdown file name suggestion based on the transcript using the medium model
response = client.messages.create(
    model="claude-3-sonnet-20240229", # Use the medium Claude model
    max_tokens=100,
    temperature=0.5,
    system="Suggest a markdown file name based on the following transcript. Only respond with the markdown file name or there will be an error:",
    messages=[
        {"role": "user", "content": transcript_text}
    ]
)

# Extract the suggested file name from the response
suggested_file_name = ''.join([block.text for block in response.content])

# Check if the suggested file name already has a .md extension
name, ext = os.path.splitext(suggested_file_name)
if ext.lower() != '.md':
    suggested_file_name += '.md'

print(f"Suggested file name: {suggested_file_name}")

# Use the new internal system prompt directly
model_choice = args.model
response = client.messages.create(
    model=f"claude-3-{model_choice}-20240229", # Use the specified model for the analysis response
    max_tokens=1000,
    temperature=0.0,
    system=system_prompt, # Use the new internal system prompt
    messages=[
        {"role": "user", "content": transcript_text}
    ]
)

# Extract the response text from the response content blocks
response_text = ''.join([block.text for block in response.content])

# Prepend the original YouTube URL to the top of response_text with a Markdown heading
response_text = f"# Original Source: {args.url}\n\n" + response_text

# Append the original transcript to the end of response_text with a Markdown heading
response_text += "\n\n## Original Transcript:\n\n" + transcript_text

# Save the response into a single file with the generated file name
with open(suggested_file_name, 'w') as response_file:
    response_file.write(response_text)

print(f"All responses and original transcript saved to file '{suggested_file_name}'.")