No explanation of why it's different or better. Only Interesting for a minute, not enough GOOD info, all vague fluff.

[ScottzCodez]

After a 15 minutes trying to understand this repo. I'm not enlightened...
Why is it needed beyond normal AI?
All text is super vague, it's seem all generated by a low quality AI.
Needs examples of Input / Output.
Whare are benefits, beyond what asking AI with PDFs in context can do?
Why process 10's of thousands of Documents? Into what better form than the original?

[yusufkaraaslan]

Fair feedback - let me give you the concrete answer.

What it does: Turns documentation/GitHub repos into Claude Skills that persist across conversations. Instead of uploading a 10MB PDF every time you ask Claude a question, you create a skill once and Claude remembers it forever.

Real example:

• Input: python3 cli/doc_scraper.py --config configs/react.json
• Output: A React skill with 500+ pages organized by topic
• Result: Ask Claude "how do I use useEffect" and it instantly knows, no PDF upload needed

Why not just use PDFs?

• PDFs count against context window (limited space)
• Skills don't - they're permanent knowledge
• You can combine multiple sources (docs + GitHub code + PDF) into one skill
• Faster - no re-reading 500 pages each conversation

The 10k documents thing: That's for huge documentation sites like Kubernetes (40k+ pages). Most people use it for 100-500 pages. It just handles big docs if needed.

Try the quickstart - takes 5 minutes to make a real skill: https://github.com/yusufkaraaslan/Skill_Seekers/blob/main/BULLETPROOF_QUICKSTART.md

[ScottzCodez]

That helps, But Claude Sonnet 4.5 and ChatGPT seem to disagree (If you read all this, I think you'll agree, it can be expensive)
You should do some real world analysis of your costs and make a Costs.MD file.
Here are my questions and Shared Chats to AI:

=====[Asked AI] ChatGPT
If you use a Claude code skill, I know it's out of context. but will you be charged for example if the skill is used to ask questions about 500 pages of documentation?
Is it just an agent using you API credits to reason about the 500 pages?
Or can it be free in this scenario?
== https://chatgpt.com/s/t_690303c2d2ec81918bf4d6e4084fa1c0

=====[Asked AI] Claude Sonnet 4.5 above question also,
It seemed it didn't search the web, so I asked.... all these question,
==[Q]= Can you research Anthropic "Skills" on the web, it something new.

==[Q]= So the skill agent uses your API credits, correct?
Does it use RAG to find the right context?
Are you charged for the Tokens used for a skill?

==[Q]= You said it doesn't use RAG, but a file system.
Is it semantic search then, keyword search, or what?
Are relevant parts from the search based on your question loaded into the Skill context window to be processed by the AI?
I assume you get charged for the file search results of the 500 pages correct?
== https://claude.ai/share/340e83ff-cb9a-488e-97b3-a0b3325af645

[yusufkaraaslan]

@ScottzCodez - ChatGPT and Claude both got this completely wrong.
USING skills costs ZERO API credits.
Skills are markdown files on your computer. Once created and uploaded to Claude.ai, they sit in /mnt/skills/. When Claude reads them, it's like reading any file - no API calls happen.
Cost breakdown:
Create skill (one-time):
python3 cli/doc_scraper.py --config configs/react.json --enhance-local
Cost: $0 (uses your Claude Code Max plan for enhancement)
Use skill (forever): $0 per use
The --enhance flag (requires API key) is optional. The recommended --enhance-local flag is free and uses Claude Code instead.
ChatGPT/Claude assumed skills work like RAG (query on every use = expensive). They don't. They're just files Claude reads.