x-reader

The Video & Podcast Digest Skill is a pipeline that converts media links (video or podcast) into a full transcript and a structured summary. It automatically detects the media type, extracts subtitles or audio, transcribes the audio using Whisper (via Groq), and generates a formatted digest.

Supported Platforms

The analysis report is highly customizable. You can modify the dimensions used in the Multi-Dimensional Analysis (Step 2) and the Personalized Relevance (Step 3) to match your specific domain.

Examples of customization:

• Traders: Add dimensions like Market Impact and Risk Assessment.
• Developers: Add dimensions like Architecture Patterns and Tech Debt.

Standard Dimensions available for customization:

• Summary: Core thesis, source, and type.
• Key Insights: Includes Core Arguments, Tools & Methods, Workflow Ideas, Data & Numbers, Risks & Warnings, Resources, and Mental Model Shifts.
• Action Items: Categorized by Quick Wins, Deeper Work, and Exploration.
• Personalized Relevance: Mapping insights to My Projects, My Knowledge Base, and My Decision Log.

You can expose x-reader as Model Context Protocol (MCP) tools. This requires cloning the repository and installing the mcp extra.

Setup Steps:

1. Clone the repo and install with MCP support:

   git clone https://github.com/runesleo/x-reader.git
   cd x-reader
   pip install -e ".[mcp]"

2. Run the server:

   python mcp_server.py

Claude Desktop Configuration: Add the following to your ~/.claude/claude_desktop_config.json to use it in Claude Desktop:

{
    "mcpServers": {
        "x-reader": {
            "command": "python",
            "args": ["/path/to/x-reader/mcp_server.py"]
        }
    }
}

Exposed MCP Tools:

• read_url(url): Fetch any URL.
• read_batch(urls): Fetch multiple URLs concurrently.
• list_inbox(): View previously fetched content.
• detect_platform(url): Identify platform from URL.

The Content Analyzer Skill transforms any content (URLs, text, or transcripts) into a structured analysis report with actionable insights. It follows a three-step pipeline: fetching the content, performing multi-dimensional analysis, and mapping insights to your personal context.

Triggers

You can trigger the analysis using the following methods:

• Use the command: /analyze [URL]
• Use natural language: "Analyze this article" or "What are the key takeaways?"
• Automatic trigger: The skill is automatically triggered after a video or podcast transcription is completed by the video skill.

/analyze https://example.com/article

For video platforms, the skill first attempts to extract existing subtitles to avoid the cost/time of transcription.

YouTube: Prefers English, falls back to Chinese. Bilibili: Uses Chinese subtitles.

# YouTube (prefer English, fallback Chinese)
yt-dlp --skip-download --write-auto-sub --sub-lang "en,zh-Hans" -o "/tmp/media_sub" "VIDEO_URL"

# Bilibili
yt-dlp --skip-download --write-auto-sub --sub-lang "zh-Hans,zh" -o "/tmp/media_sub" "VIDEO_URL"

The Groq API has a maximum file size limit of 25MB. If the audio file exceeds this, it must be split into segments (e.g., 10-minute chunks) using ffmpeg before transcription. Segments must be processed sequentially to avoid Groq 524 timeouts.

# Get total duration
DURATION=$(ffprobe -v error -show_entries format=duration -of csv=p=0 /tmp/media_audio.* | head -1)

# Split into 10-minute segments (keeps each under 25MB)
SEGMENT_SEC=600
SEGMENTS=$(python3 -c "import math; print(math.ceil(float('$DURATION')/$SEGMENT_SEC))")

# Cut segments
for i in $(seq 0 $((SEGMENTS-1))); do
  START=$((i * SEGMENT_SEC))
  ffmpeg -y -i /tmp/media_audio.* -ss $START -t $SEGMENT_SEC -acodec libmp3lame -q:a 5 \
    "/tmp/media_segment_${i}.mp3" 2>/dev/null
done

To enable video/podcast transcription (via Whisper) and AI-powered content analysis in Claude Code, you must manually copy the skills/ directory from the cloned repository to your Claude Code skills directory. This is not included in the pip installation.

Installation Steps:

1. Clone the repository.
2. Set your skills directory environment variable.
3. Copy the video and analyzer skill folders.

export CLAUDE_SKILLS_DIR="/path/to/claude-code-skills"
mkdir -p "$CLAUDE_SKILLS_DIR"
cp -r skills/video "$CLAUDE_SKILLS_DIR/video"
cp -r skills/analyzer "$CLAUDE_SKILLS_DIR/analyzer"

Xiaoyuzhou is a Next.js SPA. The audio URL can be extracted from the __NEXT_DATA__ object within the initial HTML using curl and grep.

# Extract CDN direct link from __NEXT_DATA__
AUDIO_URL=$(curl -sL -H "User-Agent: Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36" \
  "EPISODE_URL" \
  | grep -oE 'https://media\.xyzcdn\.net/[^"]+\.(m4a|mp3)' \
  | head -1)

echo "Audio URL: $AUDIO_URL"

# Download audio
curl -L -o /tmp/media_audio.mp3 "$AUDIO_URL"

To read X Articles or login-required X pages, use the Playwright fallback by logging in once via the CLI.

Login Command:

x-reader login twitter

Reading gated content:

x-reader "https://x.com/user/status/123"

External Session Cookies: By default, local X cookies stay local. If you want to allow Jina Reader to use your saved X session for gated Articles, set the following environment variable:

export X_READER_ALLOW_EXTERNAL_SESSION_COOKIES=1