Urgent.News

What's breaking now, across thousands of outlets.

AI

My Claude Code config costs 9,857 tokens before I type anything

I installed 107 skills, 38 agents and 15 commands into Claude Code over a few months. Standard stuff — you see a skill recommended somewhere, it looks useful, you drop it in. Nobody ever tells you to take one out. Last week I finally measured what that pile costs. The answer is 9,857 tokens, and I pay it on every single session before I type a character. Here is how to check yours. Skills have…

Claude Code's configuration costs 9,857 tokens before any input is provided. This cost accumulates for each session, regardless of whether the user types a single character or not. To understand how to check this cost, it's essential to know that skills within Claude Code have two types of expenses. The first expense is the actual code associated with the skill, which is loaded when the skill is triggered.

This cost is visible and considered fair, as it is only paid when the skill is actually used. However, the second expense is the description of each installed skill, agent, and command. This description is included in the model's context window throughout the entire session, even if the skill never gets triggered. This "rent" must be paid continuously, as it allows the model to determine which skills, agents, and commands are available.

The script provided scans the user's Claude Code configuration, counting the number of characters in the description of each skill, agent, and command. It then calculates the total cost in tokens, which is the accumulated expense for each session.

Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

This story

This is one outlet's version. Read the fullest account.

Read the original at dev.to →

More in AI

Tech firm urges youths to embrace AI

A technology firm, TechRise DTI, has called on youths to embrace Artificial Intelligence and digital skills, saying those who combine human expertise with AI tools will be favoured in the future.

Qwen3.8-27B vs Muse Glimmer 30B: Which Permissive Open-Weight Model Fits Your Local GPU?

The August 2026 open-weights pivot For developers and machine learning engineers running inference locally, the open-weight landscape in 2026 has often presented a frustrating compromise.

  • Qwen3.8-27B and Muse Glimmer 30B are permissive open-weight models released in August 2026
  • Qwen3.8-27B has larger context window of 262,144 tokens for longer sequences
  • Both models can run on single 24 GB GPU with 4-bit quantization

More from Sunday 30 August →