Agents
A single-author agent skill claiming a 45% failure cut on agentic coding hit 851 stars in eight days
Spielewoy/autoprompt-skill was created 17 August, was pushed to as recently as today, and stands at 851 stars. Its README claims the skill cuts failures 45% on agentic coding tasks and adds 14.61 points on Terminal-Bench 2.1 across nine supported agent providers. Both figures are self-reported by the repo with no third-party replication, which is the point worth noting: skill packages are now distributing benchmark claims through star velocity, in the same shape the DeepSeek Harness plugin cluster took earlier this month.
↳ Follow the thread