Autoprompt Skill Publishes a Measured Terminal-Bench 2.1 Run: OpenCode Goes From 60/89 to 73/89 Solves

GitHub·high signalXBlueskyLinkedInCopy link

Spielewoy/autoprompt-skill, created 2026-08-17 and already at 407 stars with three releases in three days (v1.0.0 on 08-18 through v1.0.2 on 08-19), is a coding-agent skill claiming 45% fewer failures. The README publishes the actual run rather than a marketing number: OpenCode 1.18.7 solved 60 of 89 tasks (67.42%) and OpenCode plus Autoprompt solved 73 of 89 (82.02%), a gain of 14.61 points and a drop from 29 failures to 16. Unusually, the repo also states the cost honestly, roughly 3x the time and 2x the tokens, and flags that those trade-off figures are planning estimates from user reports rather than retained logs.