clawdi-ai/store/summarize-pro

Overview

This Skill converts various input content into a structured analysis report for the Agent to summarize, in two phases:

1. Skill handles: Content acquisition (download/read) → Transcription → Scenario recognition → Output analysis report
2. Agent handles: Read analysis report → Call user-configured LLM → Generate final summary

Supported Input Types

PDF, images, and web pages are handled directly by the Agent — no need to call summarize.py.

Core Scripts

scripts/summarize.py — Main Entry Point

Typical usage (Agent call):

python3 skills/summarize-pro/scripts/summarize.py <input_file_or_URL> --full --quiet

Parameters:

Output files (default path):

./summarizer-files/<timestamp>/<filename>-transcript.txt
./summarizer-files/<timestamp>/<filename>-summary.md

The last line of stdout outputs the absolute path to summary.md for the Agent to read.

scripts/transcribe.py — Transcription Sub-script

Called internally by summarize.py — Agent does not need to call this directly.

python3 scripts/transcribe.py <audio_file> [--language en]

Transcription result is output to stdout; summarize.py captures it and writes to *-transcript.txt.

Dependencies

Scenario Recognition Rules

Logic: Count keyword matches in the first 3000 characters; 3+ hits → classify as that scenario.

Use --type to force a specific scenario type if auto-detection isn't accurate.

Workflow

flowchart TD
    A[User Input] --> B{Input Type?}
 
    B -->|Local audio/video| C[summarize.py transcribes]
    B -->|Online video URL| E[yt-dlp download → transcribe]
    B -->|Text file txt/md| F[Read directly, skip transcription]
    B -->|PDF/image/web page| G[Agent handles directly]
 
    C --> H[Scenario keyword recognition]
    E --> H
    F --> H
 
    H --> I[Generate analysis report *-summary.md]
    G --> I
 
    I --> J[Agent reads report]
    J --> K[Call user LLM for summary]
    K --> L[Output structured Markdown]

Analysis Report Format

*-summary.md structure:

# 📊 Transcription Analysis Report
 
Filename, processing time, content type, character count, estimated duration
 
## 🎯 Scenario Recognition Result
Scenario type + summary strategy guidance
 
## 📝 Full Transcript
Complete transcription text
 
## 🤖 Next Step
Agent operation prompt

The Agent reads this report, then follows the "Summary Strategy" to call the LLM and generate the final user summary.

Error Handling