Unused skills and connectors load into every chat and cost tokens. According to the author, a periodic clean-up can cut that sharply.
Why this is useful to know
Ben van Sprundel from the Ben AI community noticed that his Claude setup got heavier the more he added to it. Old rules, unused skills and connectors load into every new chat, before you type anything. According to him, cleaning up saved around 15,000 tokens per new chat. Of his 25 connectors, 19 had not been used in thirty days. Why this matters: you pay, in tokens and so in money and speed, for things you never call. That is why he reviews his setup once every two weeks. Mind you: these are figures from one personal setup, not research.
When you need this
Sound familiar? Claude responds more slowly, you have dozens of skills and connections switched on, and nobody remembers what they are for. Start with a list of your skills, connectors and your own instruction files such as CLAUDE.md. Switch off what you have not used in thirty days. Remove rules written for older models: according to him, new models need fewer instructions and old rules can actually hold them back. Skills you only want to start on purpose should be set up so Claude does not pick them by itself. Also check scheduled tasks for silent failures. Be careful with an external tool that changes your setup automatically. Look at the report first, approve changes yourself and check what rights the tool gets on your accounts.
How it works under the hood
In every conversation Claude loads the descriptions of all available skills, the tool definitions of connectors, the memory and project files such as CLAUDE.md into the context window. That is the fixed base weight of your setup, before the first question. So unused skills still cost you, because their description always comes along. According to him, connectors are one of the biggest items, because every tool definition counts. He says the setting disable-model-invocation in the frontmatter of a skill stops the model from calling that skill by itself; the skill then only runs through the slash command. Note: the audit skill starts five subagents that each use their own context, and those cost tokens too. How much you save depends on your own setup.
This news touches chapter 9 and 10 and 11 of the course AI(l) or nothing by Newway. If you follow the course, this subject returns there at a calmer pace.
What you can do with this
Want to know what this means for your business? Plan an introduction: half an hour, no sales pitch, but an honest answer to the question where to begin. Rather read on first? The course AI(l) or nothing is free and starts with the basics.
Written in response to a post by Ben van Sprundel in the Ben AI community on Circle.
