Instagram Transcriber (`/ig-transcribe`)

Instagram Transcriber Workflow

When the user gives you a video link (e.g. /ig-transcribe ), follow these steps using the fully local, isolated python script.

Zero-LLM-Contact Security Model

CRITICAL: Never ask the user to paste their OpenAI API key in the chat. The Python script handles authentication transparently.

System Requirements

This script relies on yt-dlp which requires FFmpeg to be installed on the user's host machine.

Step 1: Execute Python Command

The python wrapper will download the audio (via yt-dlp), send it to the Whisper API, and print the raw transcribed text.

CRITICAL: VIRTUAL ENVIRONMENT To ensure the user's host machine is not polluted, this skill uses an isolated virtual environment.

Run the script exclusively through the provided run.sh bash wrapper. The wrapper will automatically build the environment if it doesn't exist and pass your arguments to the script transparently.

# Setup authentication (if auth error occurs):
/absolute/path/to/Obsidian-AI-Skills/instagram_transcribe/run.sh auth

# Transcribe URL:
/absolute/path/to/Obsidian-AI-Skills/instagram_transcribe/run.sh fetch "URL_HERE"

Step 2: Read and Format

The underlying Whisper model might lack punctuation or have spelling errors.

Your job as an AI is to act as an expert editor on the returned raw text:

1. Fix any obvious speech-to-text recognition errors.
2. Add proper punctuation.
3. Add markdown formatting (Headers, bold text).
4. Do NOT hallucinate content not in the transcript, just clean it up.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: tpitsunov
• Source: tpitsunov/obsidian-skills
• License: MIT

Install and usage instructions live in the source repository linked above.