Claude Code SKILLS.md are a token trap.

Tim Carambat · Beginner ·🧠 Large Language Models ·3mo ago

Key Takeaways

Criticizes the use of SKILLS.md files as a token trap in LLMs like Claude Code and Codex

Original Description

Just a video where I monologue about my feelings on why SKILLS.md files are not the future, why they are intentionally a dark pattern to lock you to a cloud vendor, and how you can avoid them. Skills are a way to, with written language, give instructions and "code" to tools like Claude Code/Cowork, Codex, and more - except they actually lead to you burning tokens massively to build a behavior that will be hard to break or expensive to maintain in the near future. Could be my own personal tin-foil hat - could just be a text file. **Chapters** 0:00 Intro 0:55 Skills.md - what is that? 2:17 SKILLS.md are just prompt engineering again 4:00 Skills bloat and the token trap 4:21 Examples of Skill bloat 5:30 How this works against you 7:00 Lets talk about token cost 8:24 Token costs are NOT going down 10:10 The bloat is the lock-in 12:22 The prime example of context bloat - GStack 14:40 Rich people - you can ignore this tbh 15:38 These tools are built with cloud in mind 16:58 MCPs actually might be the way to save us???! 18:29 That’s basically it
Watch on YouTube ↗ (saves to browser)
Sign in to unlock AI tutor explanation · ⚡30

Related Reads

📰
RAG in Laravel: Embeddings and pgvector for a Knowledge-Base Bot
Learn to integrate RAG in Laravel using embeddings and pgvector for a knowledge-base bot, improving its ability to answer questions about your specific data
Dev.to AI
 📰
Kimi K3, and what we can still learn from the pelican benchmark
Learn about Kimi K3, a 2.8 trillion parameter AI model, and its potential impact on the field, plus insights from the pelican benchmark
Simon Willison's Blog
 📰
Kimi K3 Max Is Here: Moonshot’s 2.8-Trillion-Parameter
Learn about Kimi K3 Max, a 2.8-trillion-parameter model, and its significance in the AI landscape
Medium · Programming
 📰
LLMs, Sports Prediction Markets, and the Joy of Finding the Least Bad Path
Learn how LLMs can be applied to sports prediction markets to find the least bad path and improve decision-making
Medium · LLM

Chapters (14)

Intro
0:55 Skills.md - what is that?
2:17 SKILLS.md are just prompt engineering again
4:00 Skills bloat and the token trap
4:21 Examples of Skill bloat
5:30 How this works against you
7:00 Lets talk about token cost
8:24 Token costs are NOT going down
10:10 The bloat is the lock-in
12:22 The prime example of context bloat - GStack
14:40 Rich people - you can ignore this tbh
15:38 These tools are built with cloud in mind
16:58 MCPs actually might be the way to save us???!
18:29 That’s basically it
Up next
5 Levels of AI Agents - From Simple LLM Calls to Multi-Agent Systems
Dave Ebbelaar (LLM Eng)
Watch →