Will Claude Code call your skill on its own? 38 runs, one skill, seven ways to describe it

A recent experiment by developers testing Claude Code 2.1.273 reveals that the model's decision to trigger custom skills is heavily dependent on the specific wording in the skill's description. By running 38 headless tests with varying descriptions, researchers found that vague labels like 'Helps with git stuff' often resulted in the skill being ignored, even when the request was relevant. Conversely, specific descriptions or the inclusion of 'when_to_use' fields significantly increased invocation rates. The study highlights that Claude matches requests against the single line visible in the skill listing, meaning that even a file's first heading can act as a trigger if it contains relevant keywords. The findings suggest that developers should prioritize clear, keyword-rich descriptions in their skill metadata to ensure reliable auto-invocation, rather than relying on complex logic or implicit context.