extracting-ideas

name	extracting-ideas
description	Extracts 1-2 digital product, startup, or content ideas from podcast episode transcripts. Use when asked to extract ideas, generate product concepts, or mine episodes for business opportunities.

Extracting Ideas from Episode Transcripts

Extract actionable ideas for digital products, startups, or content from Ngobrolin WEB podcast episode transcripts.

Overview

This skill reads episode transcripts from src/data/transcripts/, analyzes them, and extracts 1-2 concrete ideas per episode. Results are saved to src/data/ideas.json.

Output Format

The output file src/data/ideas.json is a JSON array with entries:

[
  {
    "videoId": "abc123",
    "title": "Episode Title",
    "ideas": [
      {
        "idea": "A short, actionable description of the idea",
        "category": "product"
      }
    ]
  }
]

Categories: product (digital product/SaaS/tool), startup (business/startup opportunity), content (content creation/course/community idea).

Workflow

Step 1: Identify target episodes

• If user specifies an episode (by videoId, title, or slug), use that one.
• If user says "all", process all episodes with transcripts in src/data/transcripts/.
• Default: automatically pick unextracted episodes. List transcript files in src/data/transcripts/, read src/data/ideas.json, and find videoIds that have a transcript but are NOT yet in ideas.json. Pick the first unextracted episode and process it. If all episodes are already extracted, inform the user.

Step 2: Read the transcript

Read the transcript JSON from src/data/transcripts/{videoId}.json. The transcript contains segments with text fields. Concatenate all segment texts to get the full transcript.

Step 3: Extract ideas

Analyze the transcript and extract 1-2 ideas. For each idea:

1. Identify pain points, wishes, or opportunities discussed in the episode.
2. Formulate a concrete idea — not vague ("build an app") but specific ("a browser extension that auto-generates alt text for images using AI").
3. Categorize as product, startup, or content.

Guidelines for categorization:

• product: Tools, SaaS, browser extensions, libraries, CLI tools, mobile apps that solve a developer or user problem.
• startup: Business models, marketplace ideas, service companies, platforms with revenue potential.
• content: Course ideas, newsletter topics, YouTube series, community/event concepts, blog series.

Step 4: Save results

• Read existing src/data/ideas.json if it exists.
• Merge new ideas (update existing entries by videoId, append new ones).
• Write back to src/data/ideas.json with pretty formatting (2-space indent).

Step 5: Report

Show the user the extracted ideas in a readable format.

Running the extraction script

For batch processing all episodes, run: npx tsx .agents/skills/extracting-ideas/scripts/extract-ideas.ts

This script reads all transcripts, concatenates segment texts, and outputs a template ideas.json with episode titles and empty idea arrays for the agent to fill in.

Important Notes

• Transcripts are in Indonesian (Bahasa Indonesia). Ideas should be written in English.
• Focus on ideas that are actionable and specific, not generic.
• Each episode should have exactly 1-2 ideas — pick the strongest ones.
• Look for moments where hosts discuss problems, wishlists, missing tools, or market gaps.