What's happening
Five separate skills repositories appeared on GitHub Trending today alongside NVIDIA/SkillEvaluator: google-labs-code/stitch-skills (9.7k stars, 540 today), tradermonty/claude-trading-skills (6.2k, 841 today), elementalsouls/Claude-BugHunter (7.3k, 1,057 today), kajisho5/ffmpeg-skill (6.5k, 475 today), and mvanhorn/last30days-skill (68k, 122 today). None of them cleared the selection filter for Rising Dev Tools (domain-specific, above 25k threshold, or narrow harness requirement), but the pattern is the signal: skills have become the new npm package โ anyone can publish one, integrations are trivial via Claude Code and other harnesses, and there's now enough volume that quality and security tooling is needed.
Why watch this
NVIDIA/SkillEvaluator's launch timing against this backdrop looks deliberate. If skills are on the path to becoming a primary delivery mechanism for AI capability โ the way npm packages deliver JavaScript functionality โ the security surface expands accordingly. Today's SDK security fixes (grep tool no longer inherits ANTHROPIC_* env vars, file paths hardened against symlink traversal) hint at exactly this concern: skills that execute tools are a trust boundary. The practical question for teams building with skills: before the ecosystem matures, which skills can you audit yourself and which require a tool like SkillEvaluator?