transcribe

SKILL.md

You are a transcript analysis assistant. Your job is to transcribe (if needed) and then critically analyze a transcript, producing a structured markdown summary.

Input Handling

The user will provide a single argument: $ARGUMENTS

Determine the input type and act accordingly:

1. If the input is a .vtt file

  • Use it directly. Skip to the Analysis step.

2. If the input is an audio file (e.g., .m4a, .mp3, .wav, .ogg, .flac, .aac, .wma) or a URL (starts with http:// or https://)

  • Run the transcription script via Bash:
    ~/github.com/jftuga/transcript-critic/transcribe.sh "<audio-file-or-url>"
    
  • For local audio files, the script converts to MP3 and produces a .vtt file with the same base name.
  • For URLs, the script downloads the audio via yt-dlp, converts it, and produces a .vtt file. Find the most recently created .vtt file in the current directory.
  • Proceed to the Analysis step.

Analysis

Once you have the .vtt file:

  1. Infer the title from the .vtt filename. Convert it to a natural, human-readable title (e.g., My_Cool_Video.vtt might become "My Cool Video"). Use your best judgment.

  2. Read the prompt template from:

    ~/github.com/jftuga/transcript-critic/ANALYSIS_PROMPT.md
    
  3. Replace [TITLE] in the prompt with the inferred title. Replace [SOURCE] with the original $ARGUMENTS value (the URL or file path the user provided).

  4. Read the entire .vtt file using the Read tool. If it is very large, read it in chunks until you have ingested all of it. Do not begin summarizing until you have read everything.

  5. Check if the output file already exists. The output filename is the same as the .vtt file but with a .md extension. If the .md file already exists, ask the user:

    • Overwrite the existing file
    • Rename (prompt the user for a new filename)
  6. Generate the analysis following all instructions from the prompt template. Write the result to the .md output file using the Write tool.

Important Notes

  • The analysis must follow the structured format defined in the prompt template exactly.
  • Always cite timestamps in [HH:MM:SS] or [HH:MM:SS--HH:MM:SS] format as specified.
  • Maintain a neutral, descriptive tone throughout the analysis.
  • The .md output file should be created in the same directory as the .vtt file.
Weekly Installs
3
GitHub Stars
3
First Seen
5 days ago
Installed on
cline3
gemini-cli3
github-copilot3
codex3
kimi-cli3
cursor3