I gave the 9 most popular Claude Code skills a sugar pill. 2 beat it, 1 did worse than the pill.

A developer tested 9 popular Claude Code skills against a placebo on 15 public tasks from SWE-bench Verified, Terminal-Bench, and TBLite. Each skill and its placebo underwent 30 runs, totaling 450 runs. The methodology was established before any results were collected. The findings indicated that 2 skills outperformed the placebo, while "planning-with-files" performed worse, achieving an 80% pass rate compared to the placebo's 100%. The developer is seeking suggestions for future skills to test and alternative control methods.