108 tests show when Claude Code Skills actually trigger

On October 26, 2026, an article originally published on khadim.tech and later republished on DEV Community disclosed the results of a hands-on test of how Claude Code Skills get triggered. It explains that a Claude Code skill is a folder containing a SKILL.md file: a short YAML header whose name and description tell Claude when to use it, followed by instructions that load only when the skill actually runs. Claude picks a skill up on its own when the user's request matches it, or the user can run it directly by invoking it as /skill-name. The author ran 108 tests on Claude Code 2.1.310. When either the skill's name or its description stated what it does, the skill loaded automatically on all 15 matching requests. When neither field described the purpose, it loaded on only 5 out of the same 15 requests. With disable-model-invocation set to true, the skill never loaded at all. The article also walks through the folder structure of a skill and the mechanism behind its triggering. So far this is the only published account of the testing, and no independent verification or follow-up findings have appeared in the available sources.